DeepInfer: High-performance inference framework for DeepSeek models
github.com
github.com
"Based on this framework, we have achieved high-throughput, low-latency large-scale inference services for DeepSeek V4.1 Flash. Currently, the service throughput of a single cluster is 155 trillion tokens/day, and the overall revenue is $38 per card per hour (at off-peak pricing)"
You can well imagine if you own your infra and can make utilization over 90%, where these go.
And then, my link above is for K3 model, which, compared to ds flash models, is highly unoptimized.
If I had a few million lying around, I would have invested in inference.
2. Newer gen nvidia gpu's like GB300 are way more profitable.
There is no reason to think inference can or will remain interestingly profitable even if you assume pure-play inference providers who don't do any model training. Serving an LLM isn't child's play but it's also not so hard as to count as a moat. Inferencing is:
1. Selling a largely undifferentiated product.
2. Into a highly competitive market.
3. To customers who are primarily sensitive to price.
Industries like this have always existed and in the modern era tend to have margins of about 3% to 5%. Any level of margin higher than that gets competed into the ground.
If you want to claim inference is highly profitable in general, you're going to have to answer why it won't end up like supermarkets or airlines, both of which have bigger moats than inference.
The other problem with this analysis is obviously that you can't just stop training models. Even if you don't care about improving capability or performance, the knowledge in their weights gets stale. Cost of inferencing has to take into account the ongoing cost of training, you can't cleanly split one from the other.
- I don't... "inference is profitable" is as of now. No one knows future, neither have I commented anything on it.
The statement is valid as it stands.
Because LLMs are like evolutionary crabs, they all get to the same point.
Boosters are right that AI is a historically impactful product, but big labs will be unable to meaningfully profit from ushering in the AI era just like Netscape failed despite ushering in the internet era.
Name a company with a currently high profit margin on inference. I can't.
OpenAI and Anthropic's valuations are as high as they are because investors hope that they will create AI that is much much more powerful than the AI they have now. The fact that they already have a substantial revenue stream helps, but it helps mainly by lending credibility to the hope that they might create something much much more powerful.
(You, Mike and I agree that selling the inference of models that are not much better than liberally-licensed open-weights models will be a low-margin business.)
It's not a huge problem for users but if margins are low it's a huge problem for the debt funded datacenter buildout. It's just pure finance at that point, about whether those GPUs can generate enough yield to beat deprecation and financing costs.
The big clouds generally don't sell compute at high margins. The margins differ a lot between services. Services like core VMs and object storage are usually low margin or even break even. Where they make the money is things where they've got proprietary software with competitive advantage, like managed scalable forks of Postgres, or egress charges that are there to create lockin. Managed DBs are pure SaaS and can generate SaaS like margins.
We're also starting to see efficiency gains, so we might meet halfway.
Not to say it can't still be a bubble, just that there is legitimately profit to be had and perhaps some sustainable equilibrium in the future.
If the only profit is from running commodity free software (as OP claims), there's no moat and OpenAI/Anthropic are worthless.
It's unclear that the leading open-source models could exist without having distilled the frontier models. If that's true, the "danger" is that these upstream models don't get built any more (because of the tremendous cost), and we end up just optimizing what we have. Also unclear whether this would be a net win or loss for humanity.
If OpenAI/Anthropic stop doing it there's no reason to believe someone else won't. Or that we won't come up with a different approach.
... and a bunch of RLHF and proprietary training data etc etc, but regardless, this is still an exceptionally expensive process, that someone somewhere is paying for.
The tremendous cost is matched by the tremendous business output.
In the span of less than five years, starting from their first $1 of revenue, Anthropic + OpenAI will have produced businesses doing at least $150 billion in combined sales. The demand isn't reducing at present, it's climbing, they can't come close to servicing the total demand for their frontier models.
SAP is under $40 billion in annualized revenue. They're 54 years old.
Market captured. Whoops.
Uber now: $55b in sales, $6.7b in op income. They won.
And to be clear, OpenAI can easily permanently subsidize a fake low cost for their LLMs if they get off their asses and attach a monster ad network to GPT already. They're years behind where they should have already been with ads, the easy money made it so they didn't have to think about it. If they miscalculate the timing on switching over from easy money (market) to ads (also easy money) it could certainly kill them.
OpenAI can knock out a lot of competition by building the largest ad network, then undercutting the cost that consumers see. The margins in ads make that possible. Google already knows how that works, they're doing the same thing right now.
I'd rather have the OpenAI + consumer base + ads position, than being Anthropic. China is a far, far larger threat to Anthropic over time. OpenAI is building a service business more comparable to what Google does. Chinese models don't actually threaten either OpenAI or Google insofar as their positions are user capture based, whereas constantly improving Chinese models are a never-ending threat to Anthropic.
Lots of companies have used this strategy. And many of them have indeed died. Uber filled a void left by bad public transportation and a lackluster cab system. The higher prices (so they can turn a profit) are still worth it to some.
> OpenAI can knock out a lot of competition by building the largest ad network, then undercutting the cost that consumers see. The margins in ads make that possible. Google already knows how that works, they're doing the same thing right now.
And you think OpenAI can compete here? If LLM's end up being yet another ad funded platform, I have a hard time believing Anthropic or OpenAI will outperform Google here. Maybe if one of them strike a deal with Apple, or get bought by them. Otherwise, ads are Google's bread & butter. And, they own Android.
Danger? You mean benefit?
That's a bit of a one-sided view.
Massive bubble.
Compare 1926 to 1999, and how technology advanced in that time. Radio wasn't widespread in 1926, nuclear fission hadn't been demonstrated, the idea of the semiconductor energy band gap, underlying the entirety of the modern computing industry, wasn't put forth until 1928. And yet, 20 years later we were purifying silicon, and 10 years after that (1954) the first silicon transistor was created and kicked off the semiconductor boom. 50 years later, we were in the Internet bubble.
We've got decades upon decades upon decades to come up with new ideas and implement them, even before the end of the century.
We already lived through multiple AGIs for the last few years.
Can AI handle driving yet?
I think that "everyone" needs some kind of an asterisk. Who's "everyone" in this?
Same with a lot of their investors, who will be just fine without the "insane profits" they hope for.
I'm not worried about their welfare, and won't shed a tear if their companies collapse.
Is the theory here that it would trigger more widespread financial meltdown?
Then I would prefer it to happen ASAP before they get even bigger.
The cost of amortising a datcenter, the gpus, the engineering costs attached all to produce inference.
At how much percentage of its inference would it be profitable (if it does), how much returns would it make over the long term and what would be the profit of those inferences/ profit-revenue and how that compares to other industries.
Are there any statistics which show this. Also how efficient token inference has gotten over time?
My workflow is one where I am heavily involved in the process and I don't like giving the agent too much autonomy + I need to follow along. Opus 5.5 speak is still cause aneurysms for my brain. It also has this vibe to its prose... like it's looking down on me. This is Prolly my bias speaking though.
I was doing the same (along with GLM 5.3 Flash) for months, and then gave Claude Sonnet 5.5 at medium effort a try. The difference was like night and day. Where both DeepSeek 4.1 Flash and GLM 5.3 Flash kept making mistakes and would have to do a bunch of iterations to find and fix them, Claude Sonnet 5.5 just did it right on the first try.
In my experience, while DeepSeek 4.1 Flash and GLM 5.3 Flash are a lot cheaper per token, they're just as or maybe even more expensive when you factor in all the extra tokens that are need to troubleshoot and fix their mistakes -- and even more expensive if I factor in my own time wasted doing the same, not to mention the frustration of having to deal with that.
DeepSeek 4.1 Flash also hallucinates a lot.[1]
I'm hoping all this improves with time, and the gap between the competent frontier models and open weight models narrows.
Quick, someone point out the one billion user system with intent + personal data rich that doesn't result in a $100+ billion high margin ad network.
It is that simple and absolutely nothing can stop the gigantic ad networks that will be built up around personal agents.
Current CPM is between $30 and $60 which is x20 to x30 more expensive than alternatives.
Now we have to wait and see if performance is also at least 30x of previous ad targeting.
Facebook is toast, sooner than later, as a business. Whatever gradually replaces/upgrades that quasi dead friends network, will be premised on AI (Zuck apparently thinks it's at least in part personal agents). In the present Meta still has a $228 billion ad business that is hyper profitable.
Google + Meta = ~$650 billion hyper margin ad business. OpenAI, aka LLM as a service for the masses, is going to want to take as much of that as they can. OpenAI's active user base comes at an attention cost to the other mega services like FB or Google Search, and ultimately that attention is worth ad dollars. The question is only whether OpenAI can take it, not whether it's there at all or high value. And of course there's no guarantee that it'll be ChatGPT upon which the next great ad network is built, it will be some AI-centric business however, and likely heavily involving LLMs.
Google's operating income last quarter was $40 billion. If OpenAI fails to take the mass market + ad tandem, then Google will more likely retain their position. The solution to the cost is already well understood by eg Google: the masses don't need tip of the frontier AI capabilities, and relying on Nvidia's extremely over-priced GPUs is a dumb business model. The biggest threat to OpenAI right now is Nvidia's margins, not China. Google's focus has been on high quality cheaper AI systems for a reason: they're looking to serve billions of people that have no use for what the bleeding edge offers vs a fast + good mid-tier. Google can do that in a highly profitable way.
You can't get the first and you can't get the money for the second. Nobody is going to give you $100 billion to underwrite the cost of competing with GPT at a one billion user scale (or even a 100 million user scale). The easy money is already drying up, the field is set, and you will not be able to get anybody to give you a huge check to attempt to chase OpenAI, Google, Anthropic, China. That's already over. No VCs will fund you on this particular business pursuit.
You couldn't meaningfully compete with OpenAI and Anthropic if you wanted to. There's absolutely nothing you can do about it.
There's a trillion dollar valuation waiting for you. What's up?
Even nation-state pockets Microsoft can't compete. If there were no moat, Microsoft would have their own GPT killer, and GPT would be dead. Microsoft has $160 billion per year in op income. They can freely afford to spend more than the entire EU (is willing to) on AI matters if it makes sense. They can't compete no matter how much they spend and they know it.
Google has nation-state money. They have chips. They have software developers by the zillions. They have a frontier or very near frontier model in Gemini / Argon. They invented a lot of the category. They can't kill GPT. I repeat, no matter what Google has thrown at the matter, they have failed to kill or stop GPT.
Google can't beat them. Microsoft can't beat them. China's models haven't stopped them (DeepSeek was supposed to kill them years ago; Kimi was supposed to kill them).
That's one helluva non-moat. They must just be really lucky.
You might as well claim Coca Cola has no moat because it's just sugar water. Brand at scale is a large moat. People use what they know unless you severely disrupt them. They're being taught to use GPT, they know GPT, and so long as OpenAI doesn't screw it all up, people will just keep using it. To say nothing of the fact that GPT still has one of the elite frontier models, so you also still have to compete with them on quality.
For such workflows it barely matters which of a dozen models gets used.
Whereas if you're handing off all dev and no longer reading code, just making CTO-level architecture decisions, then there are only 2~3 models right now worth using, none cheap.
I use frontier models every day and night, but not sure if the described approach is possible with any of them right now. Models still make stupid mistakes and people who do not read the code produced are bound for surprising discoveries in their codebase. Not trying to downplay the effectiveness of LLM-assisted coding and all that, but it's hard to imagine how much stupid shit i would've committed if I was not reading their output.
I used to find real dumb bugs in code reviews. With 2-3 different models doing a combative code review.. most of that is already fixed when I get it.
Another problem is that review changes can easily introduce new issues, so my review prompts go like that: if you find outright shit, fix it right away, but if you have slightest concern, just write your ideas and suggestions in the review and let me decide then. I try from time to time to omit the "slightest concern" part and make agent do all the shit it wants, and usually end up reverting those changes.
But as the sibling comment said, not all code needs quality and indeed I skip some code too, like user playwright journeys -- I only need one auto review pass to verify they indeed test things according to a doc, and I'm fine not reading them. Not all code like that though, some stuff would wreck havoc later if its not looked at.
For the past year or two I think it has generally been agreed that inference is profitable, but the LLM business as a whole is not.
But it's clear that it's not the money burning pit I think it was just a year ago.
Consumer get the cheaper energy, while industry normally get the twice as expensive energy.
But for AI/Datacenters, there is a exception and they get the cheaper consumer level energy.
You need to take in account that the way China taxes is not the same as here in the West. And a lot more is put on the local governments regarding generating their own revenue (why there was so much building in China because those projects funded local government). A bit of a side topic lol. Its not so simple as they are making a loss on something because how its structured.
AI gets the cheaper tariffs then the rest of the industry, that is a fact. They simply have less energy profits on the AI energy. Now if you consider that subsidized or not, ... in both cases China is still making a profit.
What is overlooked is the massive amount of energy generation that China has and still keeps expanding. And that is a direct reflection upon why the prices for industry and consumers are so cheap. This buildout ironically had nothing to do with AI but was more about lower energy = cheaper industry production ... and a lot of buildout was also to build more independence from energy imports. They are not just throwing solar and wind turbines around like candy, the amount of reactors constructed and still under construction, ...
> China currently has 36 to 38 nuclear reactors under construction, maintaining the largest nuclear buildout in the world and accounting for roughly half of all global nuclear plant construction
So yea ... The AI in the west have a issue because it really does not matter if you cut China from Nvidia GPUs or advanced nodes, they compensate on the power level.
Edit: And a little extra, the irony is here in the West building nuclear is expensive as hell because we barely do one or two at the time. With a lot of lost expertise. China doing tons of them non-stop cuts down on the costs a lot.
Wowza! Thanks for all the info really helpful to have insights into how it all works.
Interesting. In Germany, industry gets the cheaper electricity...
China's electricity pricing involves cross-subsidies. Residential and agricultural users pay the lowest rates, which are generally below the cost of supply. Commercial and industrial users pay more and contribute the most of grid revenue.
Prices vary by region: commercial electricity can approach 1 RMB (0.15USD) in Beijing, but the national average is about 0.67 RMB (0.1USD), while industrial electricity averages around 0.63 RMB nationally.
In China, end users pay a single bundled retail electricity price that already includes the full cost of both generation and delivery. The State Grid company purchasing electricity from generators and settling with them centrally.
Electricity prices are also government-regulated, and any change must go through public notice and approval.
as if electricity for spacex wasn't literally illegally installed gas turbines polluting the air
This idea that Deepseek needed to distil their way to success is nonsense. Its a great narrative for OpenAI and Anthropic to spread because Americans have been brainwashed for two decades into dealing with their own inadequacies by blaming things on scary asians stealing things.
In reality China is outworking and lapping most American companies. They produce 10 EEs to every 1 America produces. They are simply harder workers.
The West is cooked because we're too lazy to work hard and we'd rather make up stories about boogeyman who are stealing from us. The only thing we're good at now is killing people who economically challenge us.
Them provably distilling after being accused of distilling strongly indicates that... that's what they do. To spin a story of them not doing that (because why would they, with their superior talent and all) doesn't seem very reality conform.
The models are impressive enough as is (considering resource constraints and difference in CapEx) despite being heavily distilled from western frontier models. No need to pretend that they're not.
[1]: https://news.ycombinator.com/item?id=48966016
[2]: https://www.anthropic.com/news/detecting-and-preventing-dist...
I think the more reasonable explanation is this is totally made up. Notice how they use examples concerning "cyber and bio" notice how the examples are politically loaded subjects that are hyper relevant to proposed AI dangers and why we need to protect western AI companies with anti-competitive regulations?
Oh no!!! The scary asians are distilling our models, but they only want the scary things!!! They want to hurt you!! We better stop the scary asians!!!
You need to up your media literacy.
Based on what? None of the accused companies have ever denied Anthropic's accusations. And here you are confidently claiming it all to be made it. How much of a reputational hit to one's honesty do you think it carries to completely fabricate someone doing something, when they can just publicly disagree and disprove you? First thing DeepSeek, Moonshot, etc. would do if they hadn't would be to make a press statement calling out Anthropic for their lies.
DeepSeek routing to Fable was corroborated by many independent sources... since... everyone was able to just try it out and see it for themselves.
> Oh no!!! The scary asians are distilling our models, but they only want the scary things!!!
The straw man you're mocking doesn't exist. Non-Chinese labs are also distilling frontier models of their competitors. Elon testified in court that Grok was trained on ChatGPT outputs. And why would they not? It produces objectively better models.
Not everything is a grand conspiracy.
You need better media literacy.
You cant just make up a reality where none of these companies have denied distillation.
The scary asians are going to get you, watch out! Your government is going to eventually tell you that we need kill a bunch of them too because they're so scary. Probably soonish. You're going to fall for it because you believe anything you're told.
These lies are then further supported publicly my western media, random people on twitter, western governments, etc., all to paint China as an evil force that must be united against. Again, in your words "Your government is going to eventually tell you that we need kill a bunch of them too because they're so scary."
And you have dozens of independent parties pushing that same lie to that end and for that purpose? Sure, that's what the New York Times and CNBC must deeply want too. When CNBC writes that they've reached out for comments to the accused companies, but received no responses, they must be lying as well?
It shouldn't be hard for you to demonstrate DeepSeek denying it. Please, go ahead, link to a statement. China's foreign ministry has denied Chinese labs distilling, and, in the same statement accused US labs of distilling from Chinese models. Now that's a very trustworthy claim.
I call it a conspiracy theory because it's a conspiracy theory. Some conspiracies are true, but this one is not. Distilling is both trivially simple and produces objectively better models. Plenty of evidence points to that practice. But in your mind the labs in China are just morally too upstanding in engage in such lowly tactics to create better models, despite the evidence. And it's convenient that the evidence can be explained away by claiming it to be a psyop to paint Chinese as evil.
Multi-turn conversations have much more entropy than one-shot responses, making it more useful to do synchronous distillation.
google is a company that tries to make a point of having profitable subdivisions
You pay for the asset (tenancy in a data center) and then you pay for housing, powering, cooling, etc.
Demand is the issue. Demand only spikes with a sota model release then in drops back down to unprofitable levels when factoring in training a new model and build out.
So your stuck in an unprofitable cycle of losing revenue when your not releasing sotas but losing profit when you are.
This goes back to googles original memo “there is no moat”.
Where in you can’t stop training sotas because you’ll lose customers to the next inference model above you instantly.
Google has seemingly defeated this by finding models cheap and good enough for their existing platforms resulting in a moat of distribution. Something OpenAI and Anthropic seems to lack.
I think everyone has known that inference is profitable overall for the major providers for a while (I'm sure not all user accounts, but in aggregate).
The major providers are unprofitable after you add in the training and research costs and other overheads.
This is why the open source models are such a threat if they become good enough - it will destroy the business model of just providing raw chat/intelligence.
Arguably that's also why a few of these providers are trying to diversify into whole-platform approaches rather than just relying on having their whole thing be a great model.
I think the more someone works with AI the more likely they realize that inference does not have to be sold at a loss, but many people's AI knowledge seems quite dated and it sounds almost like it becomes a convenient conspiracy theory for some, a way to convince themselves that the market changes that are already happening a temporary (and I don't mean in a Jevon's paradox sort of way, but that companies will go away from AI).
I suspect a lot of this is fueled by a few stories of companies who went all in on spending as much as possible on AI and then begun to walk that back, without realizing that was more about bad incentives and over promises of what AI could do (especially 1/2 model generations ago). Some also confuse running the SotA models at the fullest capacity costing more and more with a given problem costing more and more to solve, not realizing that the cost per problems is trending down but newer and larger problems can be solved for even more, so the costs at the top are rising despite falling prices. Also things like the subscriptions getting more expensive, though how much of that is because developers are doing far more than they were 6 or 12 months ago (and in some cases there was subsidization that is going away). And maybe a bit with the free models starting to not be upgraded or the harnesses getting worse performance of lowered free limits.
The very short lives of all these models also factor in.
I think the funny thing is the more inference, the more tokens a specific version of a model spits out, actually does make it more profitable.
But somehow a lot of people truly think that each token generated increases the cost of OpenAI/Anthropic/... and shout bloody murder about "subsidies"
And I'm certain there are people out there that actually think the API prices of OpenAI and Anthropic somehow equal their electricity costs.
And that's the rub - it doesn't matter if inference is profitable, the US labs have promised the whole kitchen sink, and taken serious investment on that pitch. They won't make it back in a cut-throat market for inference. They have to continue pitching the idea that only their models are worth using, for quality/safety/performance/etc reasons.
Which means continuing to develop and release new models. At huge expense.
Side note - while deepseek seems to be doing fine, I'll note that the emphasis on cost effectiveness is very much not a thing for the US labs right now... and it was for non-us labs. I'd be quite interested in knowing if anthropic is pulling in the same returns, or what the margin on their subscription plans actually is.
---
I'm already seeing this directly at work - lots of devs still happily use claude for development, but the products we're making aren't. They'll use kimi, glm, gpt-oss, etc... basically - if cloudflare is hosting it, we're exploring it (mainly for contractual/security reasons): https://developers.cloudflare.com/workers-ai/models/.
Because we are price sensitive, and we already have to do all the work of creating very robust evaluation frameworks for our products, so we might as well choose the cheapest model that meets our standards.
whiiiiich is why we're seeing things like Haiku and gpt-luna priced down near the open models at ~.10/mm in, ~.50/mm out. They've been forced to cut prices to keep commercial customers on their cheaper models instead of moving away.
This will continue. If inference is profitable and easy, prices will continue to erode. Go browse openrouter, plenty of providers happy to undercut the competition. Bigger shops with auditing requirements will avoid them, but smaller ones won't.
I guess my question is when do we say a model is "good enough"? For my use cases Opus 5.5 feels as if it is exactly at that point. And at that point of course one of the main factors I will evaluate against is price. So far I am happy with my (heavily subsidized) subscription, but if the time comes, I'll happily switch the inference API any of my agents are calling. So far I have no big lock in on any providers.
But the under-reported corellary of this is that a TON of industries essentially operate on hardware right out of that time period (or slightly after, circa 2000s)
banking
Insurance
Healthcare
Gov
retail
Transportation
etc...
Hell, I was writing code to run on IBMs Z/arch as recently as 2021 for banks with mainframes deployed around 2004. They were just then finally considering an upgrade to the later models.
Because again - I think lots of businesses are already taking a good look at models well off the "cutting edge" because they're good enough for what that business is trying to do. They're already there.
If you're doing categorization, tagging, form-filling, customer support routing, etc... you don't need anywhere close to cutting edge expensive models.
Hell - Jev being so popular is basically exactly that group going "what's the cheapest way we can make this decision".
Meanwhile, there are plenty of spaces where you still want real model improvement. But a lot of traditional business was basically process and paper pushing. You don't need a Phd level model for that, you need an intern level model.
---
But that's the trap for the large providers - intern level models are cheap and easily available with current open weights. It's going to be a very tight market, and margins will go down to essentially the floor (especially once we exit the current constrained-compute market).
Somewhat ironically - the technology itself makes switching costs basically zero. It's not just that there's no big lock in, AI is like an anti-moat. It actively fills in and paves over whatever moat you might have for price conscious customers.
I also agree that more capable models are still desirable in plenty of spaces. I just see that there is so much lag in the diffusion of existing models, which, again, are already sufficiently capable.
To your point, I think we’re past the hump on the Pareto distribution. The models are good enough for most things, and the rate they’re improving at the things they’re bad at is dropping. The odds a new model is good at something the old one was bad it is okay, the odds it’s something you care about is dropping. It makes a lot of sense to discriminate on price rather than features in that environment.
Further, even if this is your take (and I won't touch the geopolitics here, maybe you're right, maybe you're not regarding ulterior motives) - go use non chinese open models that are also good.
Laguna, Nemotron, Mistral, Arcee, Llama, etc...
In organic capitalism you need to compel arrays of discrete private entities to invest in your idea. In sandbox capitalism, the money can just appear in any amount at the behest of a single controlling individual.
Less intervening and micro management but oh boy say something not nice about government and potus and suddenly you are a new enemy of the state and 3-letter agencies work day and night to bring you down for good. Or in reverse - put a public bribe to potus and his clan and enjoy some favors for few weeks.
2 years ago I would agree more strongly with you, but we are in new era from now on.
real-life, unregulated capitalism lends itself to exactly what you described
In actual capitalism, POTUS wouldn't have enough power to be worth bribing.
The problem isn't the money, the problem is the power.
Contrast that to US being happy with everyone eating each other alive until all but one company is left standing and can now dictate the prices and enshittify the product to pay back their VC loans.
I mean this is not the case for either, but the systems are setup to drift in these directions - a market needs some sort of regulation to keep things sane, otherwise you end up with standard oil and the like.
You want the people with guns controlling things to be at least nominally interested in “the people” and produce more than just “return on investment”.
Not saying that China is like this or that US is bad, but saying that China is bad because it is “controlled capitalism” is kinda missing the point of what it should be in the best case.
(But to be clear: 25-50% faster is still a big win in my book.)
But I'm also interested is retraining needed? how relevant will a model be in N years if it's just "static", and if you do need to retrain and fine-tune over time, is it still profitable? how often is that needed?
For most cases it's probably ok, often times answers are general or basically a web search anyways, but if my Agent started blasting outdated syntax it would suddenly be way worse for coding for example.
I have SO many cases, for example when shopping, when they make assumptions based on their cutoff date, and they need to redo everything based on lots of searches.
Sure, for many use cases old models will be useful but for general usage a model that's 5 years old will probably be garbage for most people, unless they prompt it extensively, which the general public won't do.
Just wait until what Ilya Sutskever has for Safe Super Intelligence (SSI) to self update the weights without retraining them to bring training costs down close to $0.
People that want to believe something different as a coping mechanism will not be swayed by this evidence either.
And the success of things like ESP32 shows that good price + good enough is a powerful combination.
As thje models get better, the % of the pie gets smaller and smaller. So higher and higher cost of getting frontier model out gets divided onto smaller and smaller customer base as other models catch up
However, the expectations for AI are not to make your software engineering work 100% faster.
It's, for example, to build software more or less end-to-end with minimal input and feedback from you.
Then next, robotics and boosting manufacturing. Not by 10% but let's say 200%.
Such are the ambitions. Can GLM 5.3 do it?
The crucial question is whether eg. Fable 6 can, and with what lead over the competition.
Open-weight models still have not beat February's Mythos.
As for the manual feedback: I want that manual feedback, but if I let it rip it will do just fine all by itself, but then I have no clue what is going on under the hood.
I have no clue about the state of robotics.
For software, I know that Opus 5.5 can handle more than GPT 5.6 Sol. I am willing to pay for the difference rather than use the cheaper DeepSeek Flash.
If you can't leverage AI to build a safe and useful robot without fully understanding the implementation, the model is not good enough.
My argument is that if safety and liability make the use of AI without fully understanding the implementation impossible, the model is arguably not good enough.
No such process exists for a model to take a design seriously that can kill people.
I imagine that the requirement is that after testing, you are confident that the product is safe.
I think you're entirely wrong, and we can leave it at that, seeing that you're clearly unwilling to engage constructively.
When we manually tested nuclear power plant controlling software, each test case was printed out on paper and tested and then signed by the tester. A lot of paper was produced and archived for each software release. (Archived for as long as the software was running in any nuclear power plant).
Everyone understood, that in a case of significant accident at the nuclear power plant, accident analysis would be done and if the root cause was an software error, this could have possible legal consequences for the developers and testers. (Of course nuclear power plants have many independent and redundant hardware controlling mechanisms, but incorrectly working controlling software could damage the power plant in a very costly way).
The prerequisite here is that you can trust it not to do something that gives criminals access to your bank account or allows users to consume an unlimited amount of services it costs you money to provide for free or bankrupts you through legal liability etc.
It's basically this:
"A computer can never be held accountable, therefore a computer must never make a management decision."
Notice what this is actually saying. It's not saying that the computer can't generate a given output. It's saying that if the computer makes a mistake and you don't bother to check, you're the one who has to eat it.
That's a great way of putting it. So far I always used 'you can outsource work but you can't outsource responsibility'.
Legendary quote from an internal 1979 IBM training document [1]
https://simonwillison.net/2025/Feb/3/a-computer-can-never-be...
This is probably ultimately unsustainable.
Maybe if China stops releasing capable models the pace will slow down a little, but they have lots of incentives not to.
But even if we think back to the previous decade or two (sans AI), most of tech has been a a red queen's race - developing and improving stuff, without any real substantial results.
Microservices and orchestration are not better than old monoliths. Electron apps are not better than native. Most programs people actually use have decades-old codebases.
Video games of today don't look substantially better or play different than ones from 15 years ago.
But why wouldn't you want (and pay) for the very best plumber? A frontier plumber will optimize the time their customers poop, pee, bathe, or do dishes 10x to 200x, and with their freed-up time,they'll outcompete those who don't pay for frontier plumbing!
The next move in my investment strategy is obvious though, invest to the hilt in PVC tubing and I thank you for this invaluable insight. Fortunately this forum has no visitors so we can rest secure in the fact that the pigs won't realize it is too late to avoid being slaughtered. Power to the bulls, after all it is their product more than anything that is responsible for the plumbing revolution.
SMRs are being commercialized. Likewise nuclear cargo ships have been seriously studies. Nuclear reactor powered (not just RTG) satellites have been built.
I know this is beside your point, but I think if we figured out how to commercialize this, demand would come.
A high per-user r&d means high costs per user. So then users leave to save a buck, qnd the price goes up again for the remaining users.
Maybe it works out OK if you want to signal wealth - the high price is the selling point. But luxury pure utility is a hard sell to consumers, and for luxury B2B it's about high touch sales (e.g. Oracle).
The only answer I could come up with is hedge funds. Everyone else fails some part of my criteria.
what does this even mean, the commoditization... like a thing you can buy from a gas station? AI commoditization is the rock bottoming of pricing.
Usually we end up with cartels in such cases, see OPEC, mafias etc. But we have China throwing sticks into the wheels of the US providers that are attempting to make the cartel and "pacing the frontier"
Like a utility. Not 'ACME intelligence' but just 'Intelligence'. With multiple providers selling you intelligence that is 'good enough' by the token.
Whether a cartel forms is independent of whether a product is a commodity. To the extent that commodity markets are more likely to form a cartel, it's only because it's harder for them to form a monopoly.
The general problem is that if your legal system allows competitors to buy each other then they have the incentive to do that instead of competing. This is why there are supposed to be antitrust laws, and why we get problems whenever they're not actually enforced. But in a typical commodity market, even buying the competition doesn't quite work because someone new can always enter the market. Congratulations, you've cornered the market on landscaping services... until some unemployed fellow who already owns a lawnmower and a truck starts advertising their services.
OPEC is kind of a special case because the product is entirely fungible but it's also finite and consumed by usage. Landscaping services aren't meaningfully finite, if demand increased then supply could match it under any plausible scenario. Copper is finite but it isn't destroyed by usage, so if someone had a lock on copper mining then people could still get it by recycling existing products.
Whereas oil is finite but then the end customer burns it. If Saudi Arabia cuts production, Germany can't make up for it by extracting more domestically because they haven't got enough in the ground, but they also can't capture the exhaust from their cars or power plants and turn it back into petroleum because the product is really energy and it takes more energy to do that than you get from burning it to begin with. So if the largest exporters get together, they control enough of the market to control prices and there is no real way for anyone else to enter the market with enough production capacity to change that.
Until you realize that the product is really energy and you can get out from under their thumb by switching to electric cars and generating electricity without burning fossil fuels.
No we don’t. That’s quite rare.
Look at the collapse of the solar market around 2010 as the price collapsed.
Look at the collapse of the steel industry in the US as outsourcing to Asia took over in the 70s-80s.
Look at the entire computing boom from 1990-2020. Memory and storage followed such a steady line down and to the right that people are struggling to operate compute businesses in a normal market.
E-bikes, scooters, electric cars, lithium ion batteries in general. Most food.
The list of these commoditized things that started out as areas of innovation and became just the marginal cost of production is enormous.
Looks like AI will be another revolution, and eventually, probably the largest.
People are saying _training + inference_ is unprofitable (I don’t have a dog in the race and don’t really care either way). That’s a lot of GPUs and salaries that aren’t generating revenue at any moment.
Personally, I would guess Anthropic and OAI are profitable. I just suspect that they’re a pretty normal level of profitable and not the absurd, near-alchemical “can turn dirt into gold” level of profitable their valuation seems to have priced in.
OpenRouter's DeepSeek provider is basically this. Instead of renting H100's for dollars per hour, you pay someone that arbitrages the $/hr to $/token conversion because you can't actually saturate an H100 for hours. It's still effectively the same as running it yourself because of the ability to change providers to someone else arbitraging the same thing, and that there's always the escape hatch of switching to actually renting the H100s yourself.
I honestly don't think there are too many people arguing "AI is actually too expensive to take my job". "AI being too much of a commodity to make meaningful profits" could very well be the case though.
They are literally hoping Anthropic & OpenAI go bankrupt (totally possible), and everyone goes back to using their meat brains for work (never happening).
In this context it just doesn’t matter if these companies are profitable after amortizing training cost, because deployment of open weight models that already exist will continue to disrupt economies for at least the next decade if all research ended tomorrow.
Even take account of R&D cost it is still profitable, because you do not actual retrain the base model every month, only distill and post train from the same foundation.
heck, I think even including the giant cost of the top tier researcher salaries, they are still profitable.
then, how are they still losing money? just read what they are actually spend the most budget: always the hardware. those companies will never profitable, because they will never stop spending money to expand their computational prowess.
Unfortunately while in the tech industry some might know that, the general public are still being misled (especially by certain folks who’ve made a living off downplaying AI) to believe inference itself is unprofitable.
It is wild how fast we're hurtling towards this outcome, lmao.
This is correct for now, but I recently read that 90% of Anthropic's code is written by Claude Code itself. So, if an AI is using its previous version to train the next version, we're not far from the day where these models will take up the decision making process too. For eg: Decisions API by OpenAI or Jev is based exactly on this. What if a model like Jev sits behind Opus 5.5 and guides it to take the next step? Wooh.
Probabilistic predictions right now are at 99% for frontier models. They will never be 100%, but if humans rate themselves 99.9% in decision making, that benchmark can be broken.
So if Claude Code is writing 90% of the code, then why are they keeping probably the most expensive engineers in the world on the payroll?
The machine writing all the code seems to not be enough, even for the company with access to the best models. What's the alternative to explain this discrepancy?
A lot of AI haters like Ed Zitron and their readers have refused to update on this. I don't know if any of them have yet, but it'll take a year or two for the idea to fully filter out of their podcasts anyway.
1. Their figures have been pretty dodgy or excluded training costs so why is it trustworthy this time?
2. API inference charged at a profitable level rather than heavily subsidised subscriptions means everyone tightens up their usage + governance Which we very much saw with the rise of openrouter et al.
Personally I think there's a lot of the bait and switch of 2 when people talk about profitability of the 10-20% of API usage for Anthropic while also wanting to lump in subsidised chat users as if they're profitable too.
I think we know if it was this clear cut then the behaviours of companies would be wildly different and more consistent with profitability.
It's not actually clear that subscription users are unprofitable as a group (and the group level is the only important one).
We do know the companies are not currently profitable, because they're open and intentional about that, but that tells us nothing about their margins without infrastructure buildout and investment in training.
> It's not actually clear that subscription users are unprofitable as a group (and the group level is the only important one).
[Open AI losing money on subscription users](https://fortune.com/2025/01/07/sam-altman-openai-chatgpt-pro...)
If anthropic were making money on subscriptions they wouldn’t be pushing people onto API + enterprise packages.
> We do know the companies are not currently profitable, because they're open and intentional about that
Why do they keep engaging in stunts like excluding infra costs then?
You are picking choosing to ignore evidence and actions when it suits you and then saying “oh we can’t possibly know” when we do absolutely know the problems of capital and debt raising on infrastructure problems (which are mostly not the Big Frontier labs either) there’s hard arithmetic for what’s needed to make these projects profitable and right now all the inputs are getting more expensive between chips and capital.
That's why so many talks about regulations. Regulations means: ban outside competition, just like US bans Chinese EVs today.
Have you seen any place that is not as tech-savy as HN? e.g. some Reddit subreddits?
Barely anyone is aware other than people who are already somewhat deep into this whole topic.
Also, what happens when the inference providers suffocate on their own debt burdens? Who will serve inference?
1. Inference providers are largely a commodity business. While demand outstrips supply, there are many providers of that supply, which does somewhat constrain how much leverage a provider has.
2. The models are not interchangeable as of yet (though this seems to be changing). A lot of people want the latest Anthropic or OpenAI model; as such, the labs still have considerable leverage.
3. A lot of the inference providers have used their leverage by doing equity deals with the labs, agreements to have some of the deal revenue spent on compute, etc.
>Is there a wait to get an inference provider or interruptions in availability?
I think that the inference providers and model vendors are mostly handling it behind the scenes with more aggressive quantization, increasing API prices, deprecating more generous monthly tiers, etc.
> The major providers are unprofitable after you add in the training and research costs and other overheads
It's meaningless to say that selling something is profitable if you are ignoring the cost to produce it.
The profitabiity (or not) of AI models needs to be calculated on a per model lifetime basis - what did the model cost to develop and train, what is the lifetime revenue the model generates, and what is the lifetime operational cost of the model (including amortized GPU cost), etc.
It makes no more sense to calculate "inference" profitability when ignoring training cost than it would to calculate it when ignoring GPU cost.
Hence "we need to slow down progress for safety" [narrator: it had nothing to do with safety]. The industry is begging for the regulation so that there is a coordinated slow down and consequently move towards profitability.
100%
Amodei has been promising the "end of the exponential" all year, when they can shift some compute away from cash-burning research and training toward cash-generating inference.
"Pacing the frontier" is how you collude with your competitors to all decelerate at the same time because safety.
I saw last week that Anthropic is currently cash-flow positive with 75% of compute going to research and training. If they can flip a switch and move from 3:1 to 2:1 or 1:1 training:inference, they will be insanely profitable.
As long as they can avoid a price war.
>> calculated on a per model lifetime basis
True only if you never decelerate R&D.
But "pacing the frontier" means every early model doesn't need its own profitable P&L.
Early models can be big money losers while helping build the next generation.
If the industry can coordinate to all decelerate at the same time, then they'll flip from incinerating billions to minting billions, and all that capex will suddenly look completely sustainable.
Unless DeepSeek suddenly gives it all away for free, which is why the big three need government regulation to prevent foreign models from busting the cartel.
... and not worry about training cost at all.
a lot of the cloud businesses of the last decades have relied heavily on open source code, again skipping most of the cost to make the product/service being sold.
Don't enter that rabbit hole, you'd be surprised at how questionable are some accounting rules and practices, even in publicly traded companies.
At the same time, training costs don’t scale with usage, so no matter how expensive they are, they can be paid for as long as inference is profitable. It is only inherently unsustainable if you lose money on inference.
Of course, it could be that the inference usage required to pay for the training is too high to be possible, but that is a separate argument.
You missed the bit where they used the word "revenue".
I suspect if you factor in all the various overheads of running datacentres the actual value is lower. Revenue is vanity, profit is sanity as the old saying goes.
However the Chinese do have the benefit that China are pushing hard on renewable energies. So at least that major part of the equation is probably far lower than your average US/EU site.
And actually paying people not to ! (Paying RWE $1.2bn to abandon wind contracts[1])
Its absolutely nuts. Pure generational can-kicking.
If the US is trying that, they are failing pretty hard - 88% of new power in the US in 2025 was renewables, and 2026 is estimated to be above 90%
No they didn't...
> I suspect if you factor in all the various overheads of running datacentres the actual value is lower.
That's literally exactly what the quoted part addresses: "Amortized cost of owning a DC full of GPUs is far less than $38/card/hr". Cost of running GPUs is far less than $38/hour, revenue is $38/hour, ??? , profit
The latter is great as a way of getting an edge in business and reducing reliance on oil supplies, but its also environmentally highly damaging.
Chinas grid is almost entirely renewable energy with strong investments in it. In the us not the case, some dcs are running on gas.
Also DeepSeek was smart. They strategically rolled out model sizes they could support even if the intelligence wasn't the world's best for that week in human history. Then they have incrementally and cheaply improved models since.
Not to mention they also built trust with open weight models rather then fear mongering. They are in the top 3 of the smartest most viable LLM labs out there.
They have more installed renewable capacity than coal, but renewables have low capacity factors, solar is less than 50% by definition. The renewable share is growing fairly rapidly, but it is by now means almost entirely renewable.
From microns S10: BITS shipped and net revenue.
Q1 '20 - $0.10 profit per gb shipped (ram and storage)
Q1 '23 - -$0.024 (cyclic ram market)
Q1 '25 - $0.131 this looks normal were on the upswing.
Q4 '26 - $1.136 Were all fucked.
If you look at a chart of S3 prices we have all been living in the "Happy Time" where the cost to run our ever growing software business has gone down. Githb says 2025 billion commits - 14 billion this year, and 2x the actions run. That means AI driven software growth is 2x to 14x.
But those ram/storage prices are going to hit the operational budget of EVERYONE soon.
Can your average sas survive a 4x price increase on its AWS/Azure bill? How about 10x (cause thats what the margins say the premium is on ram).
If the circular financing is a concern, the circular firing squad should be a bigger one. Your paying for tokens to build more software to pay more to run your software is not a sustainable cycle. People are starting to use the words CapEx again - and it's a good thing.
Present valuations are the bubble, not the business / industry. The exact same thing was true about the Internet bubble circa the late 1990s, except this is a lot more important than the Internet and the global economy is drastically larger.
Productivity is how you expand economy and wealth. It's how China can go from being one of the world's most backwards agricultural nations, unable to feed itself properly, to gradually becoming an agriculture juggernaut. AI is a hyper productivity tool, saving people time and bolstering capability are two of the primary productivity concepts.
We're not even out of the first inning yet.
Distribution of intelligence: data, photos, video, communications, transactions, entertainment, et al.
And to be clear, it doesn't matter if some random commentors on HN don't think very highly of some or all of the intelligence being distributed. Their elitist opinion of the quality of intelligence of a given thing doesn't matter as to whether it's happening or not.
When a 16 year old records a cloned TikTok dance video, edits it, and then publishes it, that's intelligence distribution. That's a person thinking through a series of actions to try to communicate to an end audience, via an entertaining medium. That's intelligence being distributed over the Internet. That's the quaint, slow, human form of it.
Up to this point that distribution was basically entirely human, capped (post saturation) at slow growth N intelligence output (derived from the total number of humans and their time / productivity capabilities). AI uncaps that and radically expands intelligence output & distribution at a growth rate that is so fast you might as well start picturing the planet having tens or hundreds of billions of people on it. The human population will shrink, the artificial population will increase dramatically. People will have more artificial friends than human friends within ~20-30 years.
A few decades from now the notion of thinking of computing separate from AI will be so foreign most people won't understand what you're talking about. It will just be thought of as computing. Humans will forget a time when we weren't augmented by the constant presence of artificial intelligence. Try taking 'smart' phones away from the masses. In about a decade try taking away their helper agents. A decade after that try taking away their home robots.
The steroid jump is AI + robotics. The Internet is trite compared to what comes out of that over the rest of this century. Robotics will physically remake our world, in a way the Internet can't. We'll deposit tens of billions of robots onto the surface of the planet and they will almost all be the result of having figured out how to make intelligence (relatively) easy & (relatively) cheap. I'm not talking about Roombas.
The Information Age is over. We're into the Age of Intelligence (insert the unavoidable quips about humanity & intelligence).
The Internet is just one way to move intelligence around (a robot body is another, a smartphone is another, a human body is another, and so on). It's a sub value to intelligence broadly. Intelligence is the ultimate value. The valuations (and sales growth) for OpenAI and Anthropic is because they're vast intelligence output machines in the Age of Intelligence, ditto Nvidia's business boom. We can now mass manufacture intelligence at an accelerating clip, and it has gotten cheaper at a comical rate over the past few years. While HN comments snort about pelican improvements day to day, the leap over a few years is astronomical in regards to what significant AI systems can now do (with LLMs being just one type of that).
That would be awful, presuming that the artificial friends are not people.
Separately, you might like Count to the Eschaton by John Wright with its ultimate project being "the sophistication of all matter".
Just as human intelligence is merely one form of intelligence, human companionship will be merely one form of companionship. AI friends will offer a new type. And for the masses it's going to feel very real and very legitimate. They'll like their AI friends as much or more than they like their human friends.
And the fake friend explosion can happen first quietly, which eliminates the initial shame phase problem. Nobody will have to know you have fake friends, or how much you interact with them. Until it's common it'll be viewed as shameful (pathetic, proof of social ineptness) and mocked by society, then it'll be normalized. This idea isn't novel, it's extremely obvious and right now we're just waiting for a few pieces of the LLM puzzle to get to the point where they're good enough to facilitate high quality artificial friends.
It'll happen shockingly rapidly when it sparks; slowly at first and then in the span of 3-5 years we'll go from nobody having or being willing to admit to having artificial friends, to everybody openly discussing it. It will then seem more odd to not have artificial friends, like people that refuse to own smartphones or use the Internet. People will start commonly referring to their artificial friends by name in conversations with other human friends. These behaviors will become entirely socially normal.
"Yeah I had Alice clean the house before I went on vacation." Every major personal robot, every artificial friend, will get a name. Humans name their vehicles, so they're definitely going to name their $10k-$20k house bot and the artificial friends they spend large amounts of time with. When you go on a long drive (etc), your artificial friends will be there to keep you company, and they will remember your life events as well or better than you do (if you give them access to that recording/memory capability).
huh. I can only think of one friend I've ever known who named their car.
Anyways, not saying I necessarily disagree with the rest of what you say, but it's out there, for sure. We're a good distance away from an AI that can meaningfully act like a friend.
(Though who knows what AI will look like in even 10 years? I'm not ruling this out)
Apple, Amazon and Clarke started the trend.
And I am presuming that the companionship that people need is the companionship of other human beings. Otherwise, tamagotchi, fish, cats, dogs, parrots, and horses are already perfectly good companions. AI can help fill in for some of those other ones (tamagotchi especially), but it cannot replace human companionship.
The world will be filled with robots is not the inspired forward-thinking of a visionary it is the uninspired regurgitated science fiction of someone who doesn’t care to understand that the world is people, a society.
This is very different from what I see.
At coffee shops and on trains, sure, people are working or on their phones. But at dinner and parties, at bars and climbing gym and dances, we're there to be there with each other.
Maybe a cultural difference between where you are and where I am?
China will have many billions of robots just in their industrial efforts alone. Every house and every apartment in China will saturate with robots of various usefulness. That single market will be at least 10+ billion in scale.
There will be a billion robots for just the delivery and transport categories globally. We're going to peak at at least 10 billion humans (population), a billion robots for delivery & transport segments is easy.
In the toy category there will eventually be far in excess of a billion robots. They will seem quite advanced compared to what we presently are used to in that category. The Microduck is a primitive example of what will be a large industry around smart entertaining toys. Your artificial friends will be able to 'hop into' (take control of) your Microduck (whatever the 2035-2050 equivalent is) and cruise around the house or wherever you're at. Entertainment will continue to be a gigantic industry and that will obviously apply to AI + life. We haven't even begun down the road of AI + entertainment, it's less than early there.
Every household in the world will eventually have numerous robots, from primitive robots (including single purpose and toys), to advanced humanoid robots. It won't take longer than 2050-2060 to push $10k-$20k humanoid robots into middle class or upper middle class homes. China will put enormous pressure on the cost of those robots, and they'll mass sell them to the bottom 4/5 of the world in terms of household incomes (that is, they'll come up from the floor on price and take most of the market, while some companies in affluent nations will produce more expensive high-end humanoid bots, ala what Tesla is attempting).
Tens of billions of robots globally is easy. 200-300 billion artificial friends is also an easy prediction. Social junkies will have 50-100 artificial friends that they try to tend to. It'll largely replace platforms like Facebook, the utility that served for the average person. Humans will get a lot more isolated from each other and as that occurs they'll fill in the void.
I have spent time in China and Korea where attitudes towards artificial intelligence are much more positive than the west. I have interacted with robots in these countries.
You are making the fundamental mistake all nerds make. You’re thinking about what is possible and concluding that it is what will happen. Just as nerds believe that someday it will be possible for us to live on mars and therefore we are going to live on mars.
Humans are humans. Humans are interacting with artificial intelligence as if it is human because they crave human contact and are so desperate they settle for the best they can get.
The houses filled with robots future goes hand in hand with the robots will be doing all our work future. And if robots are doing all our work, what do humans have? Time. So much time. Humans become creatures of leisure. No longer is there a too little time to build relationships, there is only time, so much time. Human relationships become abundant, humans spend all their time being human.
The idea that people will have 50-100 artificial friends is so batshit crazy that I could totally believe your comment is parody.
How do you envision an "artificial friend"
But yeah, masses will do masses stuff and that's why even shit products can sell.
I would argue it will be 10x in 2-3 years the way things are going.
what do you think it is, out of curiosity?
$38/card/hr is a ton of money. That's $2.6 million per 8-card server per year.
The cost of power/physical space/interconnect by comparison here is negligible.
These providers have very little incentive to charge unprofitable prices (except maybe capturing more of the market, but there's not much lock-in here), so it's reasonable to assume somewhere in that spread is the "true" cost of being in this business, all expenses included.
[1] I'm using https://gpus.io as a reference here, and that's on-demand prices, which will tend to be the highest relative to other options.
- If inference is this profitable, is it profitable enough to offset the enormous training costs, bot those already incurred and those in the future?
- If inference is profitable, can it cover other capital and operational expenditures the hyperscalers and neoclouds are committing to, on the timelines they're committing to, to scale out inference?
- If inference is profitable we can measure if the price/earnings ratio of current valuations is sane. Is it?
- If inference is profitable, is it - given the answers to the questions above - profitable enough to make the risks of lenders of the huge amount of debt companies are taking on to scale, rational? It's one thing to say as a hyperscaler "we can make this pay", it's another to say "I'm going to lend them the money to do that and I'm going to get my money back on the timescales we agreed, and with an ROI to boot".
- Finally, if inference is profitable why is Nvidia specifically lending money to companies to buy Nvidia cards to do that inference, and then hitting the credit default swap markets to cover the risk of never getting that money back, and doing so at ever increasing cost, when they could just go build a profitable inference business themselves?
I agree seeing the numbers is good. It does suggest profitability in some segment of the market. However I still have a lot of questions.
The US is dealing with different tax code where buying GPUs function as as great way to offset taxes, and having boatloads of debt is rewarded in many cases. I think there's lots more (crooked) reasons why inference is even more profitible in the US.
Most of your questions are things that are relevant to everybody due to a possible economic crash. And you are right to point this out.
But for me as an AI user those aspects are not really relevant in my decisions. If an AI bubble bursts providers will still want to sell me those tokens at that price and yes development will slow down for a time which does change not a lot for me.
Inference has always been profitable. The problem is that this profit shrinks as your model becomes less competitive. The labs are in a constant race to release the best model to maintain inference profit. Deepseek is not training their own models at the scale of OpenAI and Anthropic and mostly sustains off their efforts via distillation.
They (well, most of them) are looking at loans and debt (in plain sight and hidden) of AI companies and asking "what will pay for that", not looking at how much a GPU costs and how much tokens you can sell off it.
Just putting a model on a GPU and selling it is profitable, period.
But spending more and more money on building and running infra to build frontier models that solve smaller and smaller piece of the pie might very well be unprofitable.
Models competing with the frontier labs gets more capable every month and more and more of the tasks can be achieved without reaching for the big guns.
We might very well get to the point where AI/LLM is profitable just fine but bubble still bursts and the big frontier labs are in trouble. And frankly that would lead to far more sustainable ecosystem.
you are just talking about ai inference? Are you assuming ai models themselves fall from the sky.
Plus there is the actual demand side to consider and how does it impact when paid utilization isn't 100%.
I could even see F1000 companies having their own backstop intelligence for most of the tasks and only use frontier where it is absolutely needed or as much as it is under their IT budget.
Won't it be race to the bottom when they compete on cost? Is regulatory capture the only moat left for the market leaders now?
DeepSeek's models are relatively small, and training costs will be correspondingly lower than for larger models. All you have here is a revenue number - no mention of (training + serving) cost, so while they may well be profitable you can't determine that from this number alone (if it can be trusted).
DeepSeek (investor meeting leak) have previously said that they price their models at a point where the GPU is paid for after 10 months, so something doesn't seem to add up here. If they were making $38/hr per card 24x7, then in 10 months they'd be making $270K per card, while the cards (Huawei?) only cost around $20K.
Say a $30,000 GPU has an operational lifetime of 3 years, and draws 1000W. If it is running 24x7 then over those 3 years it will consume 26,280 kWh of electricity, at a cost of say $0.10 per kWh, giving a lifetime electricity cost of $2600, or less than 10% of the cost of the GPU itself.
So, to a first approximation the cost of serving a model is the amortized GPU cost.
Assuming a 64 card cluster
$38/hrs equals $320K in 10 months.
Assume they only have 100 clusters, and as much as 40% operational costs. That would get them about $20M in 10 months, which I suspect is more than what 4.1 flash training has cost them.
I haven't accounted for 7 hours of peak-pricing per workday.
Of course, you won't have 100% utilization.
For four cards at 100% utilization it's $4712 per workday and $3648 per day on the weekends and holidays.
Quick numbers:
At $35/hour in net cash generation per card, a $30,000 GPU would pay for itself entirely in about 850 hours (roughly 35 days)
With an assumed ~ $3 operation cost per card and hour
I think its more like 40 cents to make at home. So I think its quite unlike McDonalds.
But they don’t have to include any of those costs.
Remember the comparison was running a model at home for yourself. The analogue is making fries at home for yourself. Thats cheaper.
Running a model at home is very cheap if you already have a M5 Studio lying around for some reason.
unless you want some sad thawed potatoes instead of fries
and that's going to run you a lot more than $4, and take quite a while to amortize
I gave it mundane tasks like updating dependencies, writing changelogs, etc. It ran about 2 hours a night.
My MBP consumes 0.1w idle and about 10w running inference. I pay $0.33/kwh. If it runs for 2 hours a day over 365 days - it will cost $2.41/year (excluding idle time).
Excluding the cost of hardware - the running costs alone are more expensive than DeepSeek's API and the inference quality is much lower. The economics are even worse on AMD AI Max mini PCs and worse still with a dedicated GPU setup.
That said - my Apple Silicon example is orders of magnitude cheaper than western inference providers and offers unrivaled privacy.
Perhaps a local AI setup would be compelling for small/medium business users... and it would be practical if the hyperscalers didn't buy all of the potatoes on Earth.
Of course that's assuming it was just generating flat out all the time - no prefill or anything. That's not likely. Probably much of the time (and tokens) would be in prefill and cache hits. And as you say, DeepSeek v4.1 Flash is a much stronger model than a 35B two generation old Qwen.
So yep, DeepSeek's API is pretty cheap. They're not paying anything like $0.33/kwh for one, and they're able to keep util much higher than you are, and they are subsidizing so they can get more data and code to train on.
If you were paying for Fireworks to host the model in the US with no training and ZDR, it would be $1.80 per million tokens (3x price), so a grand total of $165.56/yr.
I'd argue that at those values, the price difference really doesn't matter one way or the other. What matters is output quality, human time required, privacy, and sovereignty. Also, there's something to be said for (nearly) zero marginal cost reducing the barrier to playing with an idea. If I had to spend $40 to try out an idea over a few million files, I might think about it first. If I could just load it up on my existing server and let it cook overnight, I would not worry about it. Maybe I should consider electricity marginal costs more, but practically I don't really.
Its more economical to make fries at your home for yourself but that is not the case for inference.
The Mac mini GPU also isn't as optimized as the datacenter GPUs
So you have to price your local hardware against however many subscriptions you would need to run a dozen concurrent agents around the clock.
But even then, I would guess subscriptions are priced on the basis that they don't expect every user to utilize their entire usage limit. I.e. power users are subsidized by casual users.
As a reference point, Z.ai is operating at 26% gross margin as of 1H26. So assuming ~90% gross margin is likely a bit off.
Orinally it was an educated guess based on rental prices I've seen but Opus 5.5 and Gemini 3.8 Flash both arive at the same numbers based on DeepSeek's Zhihu post and GB200 specs. Around 30 cents before depreciation and $2 to 3.50 including it.
A. The average utilization rate of inference GPU cards is not that high (i.e. it peaks during the daytime, but is idle at night). So the time window for earning $38/h is fairly limited.
B. They are renting costly GPU capacity from 3rdparty hyperscalers because they can't meet the peak demand with their own hardware.
C. Free-tier users are eating the gross margin.
D. Subscription users are eating the gross margin.
Another observable point is that DeepSeek's company-wide gross margin was reportedly 44.6%. It's intriguing what the full picture looks like here.
In DeepSeek's case this kind of levels out because they have a large user base both in China and the West (time zones), and additionally attract users in the non-Chinese timezones using their off peak 50% discount which is what DeepSeek talks about here.
> B. They are renting costly GPU capacity from 3rdparty hyperscalers because they can't meet the peak demand with their own hardware.
I looked this up from Chinese sources and DeepSeek primarily builds their own compute infrastructure rather than renting it
> C. Free-tier users are eating the gross margin. > D. Subscription users are eating the gross margin.
That's true, but given how lucrative inference at API rates is, they can subsidize training and individual users to some degree. Are they still losing money overall? Not unlikely, but with moonshots that is often the case. Remember that Uber and Spotify took around 15 years to have their first profitable quarters.
Source: https://x.com/FuckAnthropic/article/2080268590033621295 (it’s a leaked meeting transcript so take it with a grain of salt)
"This is our API pricing. For our API pricing, we consider a reasonable profit to be roughly: we buy a batch of equipment from the market, and recoup the cost in ten months. I think that's a reasonable profit."
As for the infra setup, I am currently using llama.cpp, although I'm just now working on switching to vllm since it is better for concurrent inference, since it can interleave prompt-processing and decode. Until now llama.cpp was enough, since I was on an R9700 and it was only practical to run one agent with decent sized context anyway.
Are frontier labs making 80+% gross margins on LLM inference? (2025)
https://johnnyclee.com/i/are-frontier-labs-making-80percent-...
- gpu market is starving, demand is higher than supply = leads to high gpu price
- number of openrouter shovel sellers rises = lower token prices
Indeed. This could actually be what breaks America's world order.
AI can already clone major products like Photoshop and Office. Maybe not at 100% feature parity for now but at a "good-enough" surface level. That will only keep improving. Eventually we could get clones of Windows and macOS at feature parity.
Even if US lobbying/corruption tries to cripple AI to prevent that, China will instigate and accelerate it.
But the architecture is similar to kimi's mixture of experts, and we already have such an optimization for different experts swapped in from disc by antirez.
https://openrouter.ai/deepseek/deepseek-v4.1-flash-1017#prov...
Truly, "the more you buy, the more you save".
The release of this repo hours after the DHS order is a tactical move they started preparing when the DHS order was issued almost 3 months ago, and they timed publishing the repo right as the order hit the federal register.
DHS: "China bad and scary. Don't send them data."
DS: "You don't have to. Happy xmass kids."
Exciting... feel like I'm back in '96, having to wait two minutes for the page to load.
Probably wouldn't have been many agencies using DS to begin with anyways.
I don't think any Federal agencies are using DeepSeek inference, but many State agencies would be, primarily through third party services. Of more concern is the critical or financially important infrastructure which is enabled by Chinese models. Banking, ecommerce, retail.
What does DS achieve with this "tactical move"? Or, are you saying this is a tactical move from a higher level (Chinese gov interests)?
What country produces all of the ASICs used in contemporary bitcoin mining?
The faster we get to a good enough open model, the sooner it gets burned to hardware, and the sooner the transition to (relatively) low tech, low margin, high volume cards to run it on …
… and those are built in China.
You're not going to make money buying a new 14nm bitcoin miner, and that is where the China foundries are for cost effective silicon. They're great at 28nm+ and good at 22nm pricing.
Most new design modern sub 7nm ASICs are made in Taiwan/Korea by TSMC and SLSI. That is where miners are profitable.
If they had a solution to get to >90% yields on sub 5nm they'd be doing it now. They don't, even the 7-10nm isn't there. Best hope is they get to buy the entire world's supply of EUV in 2027, then they install all of them, then they bring them up, then they build processes around them and make masks, then they ramp the yield on 3-4nm by copying exact TSMC.
The hilarious thing is that TSMC couldn't do this at a new fab in Taiwan, with full access to equipment, supply chain, and people. So, nope.
From that point onwards, we'll buy Chinese hardware or be left behind
Already, the big capabilities are becoming multi-layer/die integration. Whether it's BSPDN, stacked die, optical interconnects, etc it won't be as simple as Denard scaling. It will be dependent on experience and supply chains, which are harder to transfer than equipment or IP.
I do think the market for inference will drive a lot of innovation, and a lot of that will be at larger nodes. A single reticle die masked ROM with CIM can do 7B at 6nm. At 2-3nm you can do 27-32B which is a single expert or router. At that point there's not much reason to shrink for AI since it's mostly 1-hot die per token (except for the router). Update the parameters every 3-6mo when a new frontier comes out, and take the cost/power savings to the bank.
Thinking on, I agree 1-2 years was too optimistic. Even if they have the tech right now already, building a fab itself within 2 years is tight, but I think doable. Maybe not at full capacity but at least functioning. It's possible they can start building the fab infra in parallel with acquiring EUV tech speculatively to speed it up.
Like the other person said, within 4-5 years is more reasonable.
Cheap labor from foreign countries is good.
Can anyone explain how this is not hypocrisy?
... so far.
No regs set in stone yet, but DoD doesn't need regs - they can tell you what the rule is and that's that.
This is potentially quite effective at corporate levels but I don't see it being effective over the long term because there are too many small-medium users that don't need to care about supply chain certification.
That's strategic sabotage to say the least.
Ban all open weight models! Cried the VCs.
https://public-inspection.federalregister.gov/2026-21447.pdf
ahahahah
ahahahah
The DHS order only bans buying DeepSeek products directly (e.g. doing inference on their China hosted servers). It does not apply to open weights at all. And DeepSeek just gave their response to the DHS order.
Now, most software products developed with DS can pass an audit, whether human or from a US AI provider, that they do what they say and aren't secretly funneling telemetry to Beijing. When I say 'most..can pass an audit' I mean that most products are simple enough to undergo such an audit, rather than implying that some have exfiltration mines buried within; it's possible but I have have no reason to think it's probable. I believe the purpose of the broad wording is to facilitate the waiver requirement, which costs money to apply for and will require payment to an accredited or government approved audit provider, thus generating DSP on the US balance sheet even if the original prototyping/buildout was done with Deepseek.
"third-party products and services developed using DeepSeek implementations".
So gov can't use western inference of DeepSeek either.
Moat-like things going for them:
- access to the US government (for huge inflated government contracts as a source of free capital, having the competition banned, having their products legally mandated, getting other preferential treatment, having foreign governments bullied into granting similar advantages)
Too unsustainable, too fragile, too small:
- Strong brands: Brand loyalty seems like a pretty insignificant factor in AI right now
- Semi-exclusive hardware access, RAM buyouts: Unsustainable even in the short term
- Lock-in effects: Their models aren't very stable across versions, or very unique, and switching providers is about as easy as it can be in a highly competitive market.
- Exclusive technology that is important, advanced and hard to replicate to a degree that would materially set them apart and justify potentially much higher cost. Maybe I'm missing something here, but a lot of the claims they make to that end fall apart even on light scrutiny (due to things like test sets that are small, hand-picked and clearly overfitted on, or broad general claims that cannot be substantiated). I'm getting the feeling they don't have anything that materially sets them apart at all, but that's seems impossible to measure in any meaningful way, so who knows, they might have something here.
I can click a button in Azure foundry and run open weight models that are approved through my compliance and procurement teams.
I don’t know how open sourcing this stuff helps deepseek but it means I’m getting ever better and cheaper models so I’m fine with it.
I'm not sure whether the DeepSeek's open-sourced inference engine does all of these, but it appears that plenty of inference tricks have just been revealed for the open source community to learn from and start improving upon.
But the key point is that something good enough always attracts more interest and contributions because a productive outcome is very realistic and tangible.
Open Weight models and competing open source inferences engines massively undermine monopolies; in theory at least.
See how cloud is easy to divest and “nationalise”.
And I imagine this engine is by default mostly just deepseek compatible and inference engines are not universally applicable to all models?
It sounds significant for competition but I'm not sure how it sinks Anthropic/openAI. Presumably they still have their frontier models, the training data they're vacuuming from current users and access to hardware to serve at a ridiculous scale, right?
In that scenario, even if global Token usage increases 100-fold in the future, that alone would not be enough to justify these foundation model companies being worth one or two trillion dollars.
Deepseek manages to be ridiculously cheap and efficient with prompt caching, i spent 550M~ (1.9b input cache miss, 1.7b output) tokens and it cost me 5~ bucks, that would be ~150 bucks on Opus 5.5, if i ran that workload in parallel that would be $50 vs $1500, e.i. i wouldn't want to subagent heavily with anthropic models.
now imagine Kimi K4 or GLM 5.5 with those cost savings
I think they do not believe that big corporations in the US or Europe would actually use chinese models in their business operations due to security, regulatory and/or business continuity concerns.
This is exactly right. The only AI that matters from an investor perspective is the AI that is built into your phone, your car, ships with Windows and Mac, and built into legal and business tools like Lexus Nexus and Salesforce.
Everything else is hobby.
what's important is where big enterprise point their budgets. and there we agree, it's not to chinese models (as hopeful as I am that we can avoid a duopoly and great local models)
In that regard it was a clever marketing move by Anthropic to release their paper "Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training" ....
I find such claims a bit like if 3M/DuPont released a paper on the carcinogen dangers of PFAS and someone said this was "a clever marketing move".
Like, great, this can have hidden dangers we only learn about years after it's irreversible… how does this make "our"* companies the good guys?
A lot of us were already concerned about backdoors mandated by the US government, but that was just encryption and this is everything that AI ends up being able to do.
* in the sense of "the G7" until the start of this year, because why would other nations prefer us over China in the first place, and thanks to Trump vs. Canada, Greenland, Iceland also now the rest of the G7 has to treat seriously the possibility that the US is more of an expansionist threat than China is.
Same trick, just too naive to see it.
The only AI that makes money is enterprise. Being open weights/self-hostable is a huge win before the pricing even enters the conversation, but pricing is also adding a ceiling on what US labs get to charge. And stuff like powering other software also works better with fine-tunes, hence legal models, etc.
Currently the price sovereignty is still not outweighing the cost of US dependency. But big providers are already working on finding ways to ensure customers independence of the cloud act.
European business users (+ consumers) can access GLM-5.3 via Mistral.ai hosting. (with GDPR)
Of course the same goes for any other model that you haven't trained up yourself.
Based on HN comments, every company would be running Gentoo and developing line of business applications in Haskell or OCaml.
The reality is what ever better and cheaper is where the money flows. Especially from an international corporation perspective.
They'd be wrong. Especially about Europe. Many companies are in a bind about this: they can't upload their confidential files to a service provider but they do want AI. So right now they end up with a choice: do nothing, breach their contracts, use a service provider that pinky promises not to share their data (which we have already seen they will violate) and run local models.
The latter is the reason you can't buy powerful GPUs for reasonable prices because a local model (GLM 5.3, DS 4, etc) + some hardware is all you need to solve the problem in a way that avoids the legal issues.
Mid tier insurance co is working with local inference providers and some of these now have thousands of GPUs up and running. Don't drink the HN cool aid that the EU is a backwater where nothing happens too much or you'll end up surprised one day.
The one thing where Europe really does lag is in training, that would be fair. But inference providers are popping out of the ground like mushrooms and there are some billion euro+ projects underway. We -unfortunately- also have some scammers that are bilking (or attempting to bilk) investors pretending they have more than they do, this is annoying but I would expect that other countries have similar players as well.
There are roughly 70 GW of datacenters under construction in the US right now, and a GW is roughly 500k GPUs, so 35 million GPUs.
The point was: they're buying them from the retail channels in relatively high volume because there is a massive shortage of GPUs in general, but that does not seem to stop parties with smaller projects who do not qualify for buying directly from NV (assuming NV even has stock to sell).
5090's now sell for the same amount (or more) that you could buy an RTX 6000 for a few months ago, and RTX 6000's are ~19K and up.
I found a stash somewhere, bought two, then two more and when I was ready to buy the remainder they were all gone. Serves me right for being too careful.
They are not, not even close. I wonder if you're aware how money is being "spent" between Nvidia and US AI corporations!
Its not just the US ... Where do you think all those 5090's end up? Its just crypto mining all over again. Most people do not realize that China was the biggest crypto mining operation because of its cheap electricity and people willingness to take big risks.
The same happens now also, with people buying up 5090's (and lower class cards) in bulk, to strap into AI rigs. With Nvidia discontinuing the 32GB 5090s, we already see pressure on the lower end cards that had a better price / performance for a while now.
China will do China things *lol* ... People there are just more willing to take bigger gambles (and the cheap elec helps a ton). But that is also a factor why a lot of consumer level hardware gets under pressure. Its not just the supply chain to the enterprise market that is eating a lot for AI.
Europe isn't in a bind. They can, and have been, running open weights on inference providers around the world. And they pay a premium for US models and accept its risk for the same reason they remain US allies. The cost is worth it.
That being said it's not quite as simple as would be ideal, you do need to check whether there's an additional EULA on your specific model choice.
The perceived lack of independence of the FTC is a major driver behind the exodus on this side of the Atlantic and it isn't just the smaller players that are moving. Airbus for instance.
Now the datacenters supposedly operating under at-arms-length management are no longer sufficient for compliance.
There are calls to repeal the whole thing with respect to the US.
https://noyb.eu/en/us-supreme-court-just-blew-eu-us-data-tra...
Yes, biased source and title, but the contents are pretty good.
Well they are wrong, I'm at a large EU startup and they are currently considering Deepseek as their default model, Opus/Fable would still be available for complex tasks.
I think people don't realize the Salesforce SAP Palantir would love to ship these models, and they have/are shipping them.
The threat of exfilteration of data via these models is high but it's nowhere near as high as the margins.
For instance if you could cut your ai input bill by over 90% would you do it? Can you put some of that extra margin say 10-20% to make it more secure.
I know of atleast 10 fortune 500 companies that are entirely planning to or have moved to Chinese models for a large portion of their work.
It's insane people in finance are delusional enough to think that won't happen.
Maybe Govt will try to de-risk with bans on using open models but I don't think that's going to work at scale, people will literally spawn labs to launder DeepSeek/GLM models into a US model by some novel training etc approach or something.
I still don't think that it's as much a kami-kaze rather it's an attempt to normalize the world around the Chinese stack.
If DeepInfer is supported by everyone, it will be much easier to export stuff aka hardware that only supports it for now or is optimized for it.
Assume china starts selling hyper optimized Asics for inference at some point.
This feels like a race to the bottom as soon as possible because they have much lower costs, if there is a Wipeout in AI Labs in US chinese labs are largely insulated, and they can hopefully buy Nvidia chips for cheap if that does happen.
Because they can and there’s no consequence to drumming up fear, same shit since the days of Napoleon and Nelson…
Quite failure by someone to assume that EU businesses would use US over Chinese, based on the current political atmosphere. US causes even bigger concerns than Chinese labs.
If you think that is the choice basket to put your eggs in, best of luck.
Thanks God nothing like that could happen in the US, thanks to its constitution, courts and armed informed citizens. That's why I'm putting all my eggs in that basket.
If you don't think it's important or needed, well, there is still a lot for you to learn.
US per capita spending on healthcare: $13,473 per person
China per capita spending on healthcare: ~$763 per person
Both have average life expectancy of 79 years. LITERALLY China has better healthcare, better housing, literally better in all the ways that count.
What does our "democracy" get us exactly???
No midterm routs in China.
For Europe, the answer is using homegrown metal, even if it is severely underpowered, and accepting that it will be running American and Chinese open-source models for the near future. (The French lobbyists looked likely to turn Mistral into another Arianespace, but they’ll be resisting implosion for at least another year.)
A single court ruling overturning an order from Xi will shut me up immediately.
> In four years, you don't have to vote again. We'll have it fixed so good, you're not going to have to vote.
At least the Chinese president isn't a pedophile enriching himself at the expense of the Chinese people. China has steadily increased the quality of life for its people where the quality of life for the average American has steadily decreased.
I'm not saying this is good, I'm merely stating that the central committee power and Xi's power is still limited by the sheer size of the Chinese state. If you compare that dictatorship to the Stalin dictatorship or Gheddafi dictatorship you'll find that Xi does not have that kind of unfettered power through sheer violence, he still needs to make a high number of cogs to work in unison to exercise it.
I think there is also a perhaps the need to assign air-quotes to "investors" when talking about the US markets.
There's too much fluff and noise there. Too much (over)leveraged trading. Too much speculation.
What the real investors think is hidden behind the actions of the speculative traders.
> I think they do not believe that big corporations in the US or Europe would actually use chinese models
The Chinese models are open-weight. You can download them and run them on your own servers. Or you can outsource to Mistral, Privatemode or any of the many other US/EU inference hosters.
So the "security, regulatory and/or business continuity concerns" arguments are effectively non-existent. And whatever argument does possibly remain, also applies equally to Anthropic, OpenAI & co.
https://www.pewresearch.org/global/2026/07/15/people-in-many...
You're entire company would become ineligible for gov contracts if you have use DeepSeek. Something like that.
Correct. And they're likely very correct in the belief.
When people on the internet claim that frontier labs are done for because Chinese open-weight models exist, they really don't appreciate that the people spending the money do not want to hear "we can just run it cheaper ourselves!" Any software engineer with even a few years of experience knows that phrase is absurd, and executives are simply not interested in the risks associated with such an endeavour when they can just buy the solution outright.
Consumers will have access to open-weight models just like how consumers have access to Linux. But you can just go look at the market share of operating systems to see why something being free and openly available doesn't matter to executives who just wants to get things done and is willing to pay a premium to a trusted business to do the work for them.
And none of that even touches the fact that these frontier labs have only ever been in-front of open-weight alternatives. Again, if you're an executive, and you're questioning if you should pay a premium for a better version of a tool which you presume is important for your company's ability to compete, then the answer is immediately obvious: pay the premium or risk falling behind competition who is willing to pay the premium. And when you're talking about global companies who can pay these bills without it being significant enough to even show up on a quarterly financial statement, then it's an obvious sell.
That's why most of the internet runs on Windows IIS rather than Linux or BSD. That's why everyone here is running a VM on their Mac that hosts Windows containers to do development in.
I think you're missing two things here.
1. The margin by which a hosted product is operationally superior to self hosting an open source alternative is precisely calibrated. The SaaS provider has no reason to invest beyond the point were they can reliably make the sale.
2. Self hosted open source solutions come with a big bonus re: the ability to read and modify their code. But realistically ain't nobody got time for that.
AI upsets the apple cart in both cases. It's now far easier to manage locally hosted open source solutions, which means that in order to maintain the favorable effort ratio for making a sale, products must all get significantly better than their open source alternatives. That's gonna be a slow process involving innovative product owners. AI is getting better faster than it can happen.
Also, since AI is trained on the open source code, it's very good at integrating open source components. Previously nobody was bothering to gain that kind of mastery over their tools but now you can leverage it right out of the gate--but only if you're using projects that are in your assistant's training data. Maintaining a moat around a software product has always been orthogonal to making your product useful, but as we lean on AI more, it'll get worse. The more you protect your IP, the more its usability degrades. This was always true, but now it doubly true.
Executives may want to pay for a solution to their problems. But the companies that might provide those solutions no longer have a sustainable path.
The future of software is as a tool, not as a product or a service.
It's generally understood that for better or for worse, if you're working on anything (not just IT) that is security-sensitive, you're going to have restrictions on what can and cannot be done that may seem absurd at first, but hold up to scrutiny once more thoroughly examined. I would say that you don't even need to thoroughly examine this particular restriction. I have prod data at work that I can't give to humans outside of the United States. Not in India, not in the UK, not even in Canada. Clients insist on these clauses in contracts. This has been a known practice across many fields in this industry for decades.
Now we're suddenly seeing a chorus of posts saying that we should loosen up and send data to a non-human intelligence that was developed by a known geopolitical rival with a history of malicious intrusion on Western networks. Where does that come from?
The piece I think we're missing is that, currently, AI is for software development. It theoretically has other uses, but nobody has found much of a way to capitalize on them. For the time being, most of the money being made is in models and harnesses for software development. There's already precedent here to use more open/cheaper solutions.
Not to mention, if you make data security guarantees or have specific contracts, you can't even use US providers anyway.
So, they choose to believe, lacking a better strategy. They will also try to lobby their belief into existence.
The labs pushing this cyber scare is an attempt to cry wolf. The corps don’t give a fuck
All other models are also either by <1k labs or side hustles by tech companies who are not particularly known for LLMs
It is clear from the beginning that the Chinese Open Model are a proxy for economic warfare. Anyway their singular goal is to dethrone US monopoly on SOTA model and burn all the economic hype and investments. If this strat works, which we will see after Anthropic/OAI IPO, we'll be in for an interesting ride
I mean, what would you expect? Like USA isn't openly planning to weaponize their current AI primacy [1, 2].
[1] https://www.whitehouse.gov/presidential-actions/2025/07/prom...
[2] https://www.whitehouse.gov/wp-content/uploads/2025/07/Americ...
Way too many monopolies these days
The overall market will still be big, very big, if it replaces real people or augments every single worker.
Some contracts will go to your local provider (US Gov, etc.).
And there is still a moat between them and DeepSeek.
As long as i can afford it, i use Opus.
I'm not bound by it and switching is very fast
Meanwhile OpenAI has grown their revenue like 10x and Anthropic has grown theirs like 65x in that same time period.
There's no cap on what you can do with frontier intelligence, nor how much of it you can use. Yes, DS and other open weight models will continue to grow like crazy (as long as they can source compute), but so will the frontier labs. I ran like 20 agents overnight last night on Opus, Fable, Sol, and Astra last night, trying to use up quota before my weekly reset, and I'd gladly have done 10x that much and used agents 10x more intelligent if I could have. And in six months I probably will.
People will take the novel bits here and port them to the above. The community as a whole will benefit. Including infrastructure engineers inside OpenAI and Anthropic.
But a lot of what's in here is already in published papers out of DeepSeek already.
My mind is at: Even though it's overvalued, it will likely keep climbing and won't crash.
Where is the market at?
Then...it's not overvalued?
Unless you mean "it will crash relative to inflation over 20 years, but not over 5 years" or something like that.
The classic dotcom example is pets.com. Would you liken Anthropic to pets.com?
if these hold a year after an IPO then we can talk.
Rather than caring about whether it's over for western providers, I look at what competitors for the western providers have been doing since their first moment in the spotlight. And what I see is that their offerings keep getting better and better, and that at the same time they're opening not just their weights but their architecture and experience.
As a customer (rather than a pundit), I just like their tools and the user experience more than I do those of hte US giants. I admire the US offerings in many respects, but also find the user experience has a lot of friction and annoyances. Deepseek et al don't always deliver the same immediate 'wow factor' but when I sit down to build something, I enjoy using their tools. After 40 years of using computers for fun and profit, I will always always bet on a product that's fun to use.
Those valuations are a scam, they have always been a scam, like an updated version of the collateralised debt obligations (CDOs) from 2008 [0]
[0] https://www.theguardian.com/business/2010/apr/18/goldman-sac...
it worked for the big short guys because the knew exactly when it was going to go bust
I hope more hedge funds start doing that. Basicaly funding open development to profit from tanking uptight super serious American stocks, to fund even more open development.
If you can't beat them, give the common folk means to skip them and profit from their fall.
Drop a few mil on tokens to make enterprise grade database to put out for free and bet against Oracle. Or train a model, or something.
If SpaceX was reset to 2016, they would still be ahead of everyone else even today. Their deployed capabilities today are ahead of the plans of most of their competitors.
Anyway, where's the revenue from space going to come from? Obviously not data centres now that Deepseek is commoditising inference.
The telecom business? Amazon is catching up there and Europe/the rest of the world is racing to build capabilities to not be dependent on Musk.
They may be catching up, but not anytime soon.
but perhaps you are thinking that starlink is the key? sure probably a bigger market - but any market it accesses is bound to eventually prefer ground lines.
so the only way to justify the valuation is if the AI side of the business becomes the most dominant AI business
Global space launch market is 6.5B in 2026. Even the total global satellite market is only 23B in 2026.
You have to assume some pretty large numbers for YoY growth, market share, and profit margin to get to 2T valuation.
US companies are throwing money at the problem, and considering how good Opus 5.5 is I’m not too concerned about them.
https://polymarket.com/event/anthropics-valuation-end-of-oct...
These have not really budged (yet) on this news.
It's possible to be correct that the AI companies have no future, but not be able to time the exact month they collapse.
If by 'collapse' you really mean a 50%+ fall, timing is not going to make or break you by any means.
If Anthropic is going to crash 50% or more at some unknown time in the next two years, and you have foreknowledge of this, you can easily make a massive profit by placing a bet weekly. The odds are VERY long per week, so on each losing bet you only lose a tiny amount, but on your winning week you will more than make up for it.
Timing is, as a general rule, important. In the particular case we're discussing, the odds of a 50% swing are so low that you can tolerate being VERY wrong on the timing if you are correct on the swing.
idk, if "prediction markets" remain a real thing, there should probably be some requirement for previous experience in options-adjacent markets to ensure people actually understand the economics of time-valued positions.
I don't understand the reference to Kamikaze in this context. Is Deepseek destroying itself in order to deep-six US frontier labs?
The term is used generically in modern warfare for an attacking vehicle which is itself destroyed when attacking a target; for example, a kamikaze drone. [0]
Deepseek started as Liang's side project funded with his hedgefund couch change compute. That set the economics for everyone else in PRC. If deepseek releases for free because Liang doesn't give a shit about making $$$ on his hobby, every other PRC player has to follow. And serendipitously, due to US export controls, PRC tech companies can't dig themselves into ruinous seppuku debt buying compute even if they wanted to - what they can buy, i.e. everything around DC infra including power, they get for fraction of the cost expediently.
TLDR Liang said fuck profits... which means every other PRC competitor had to fuck profits, which ends up fucking profits of western labs so deep in debt due to rampant financialization and can not afford to say fuck profits.
Sometimes a billionaire is fine just burning $$$ on a hobby while remaining a billionaire instead of trillionaire, and it just so happens the side effect might mean preventing others from being trillionaires as well. Bonus others is geopolitical adversary who is hedging economy on trillion $$$ returns.
Seems kind of pointless? Wouldn't anyone in the US/western world simply use the chinese models in their own inference? That would not strengthen the Chinese position because the models got downloaded for free off HuggingFace, and then cloud inference bills would just put money into US companies.
According to Greek economist and former finance minister Yanis Varoufakis, an open source AI strategy is one way a capitalist economy can avoid the rent trap (platforms being created, monopolizing resources, and extracting ever higher rents from everyone else).
https://www.yanisvaroufakis.eu/2026/08/31/ai-is-reviving-chi...
One person's hobby started the industry in PRC and set the trend and inspired talent in nascent field with no established business model. PRC competitors didn't want to FOMO and just knew they had to compete with opensource and everyone fast followed. They bought what limited compute they could, hired what talent they could and just told them get to work. There's a reason even Chinese Uber / DiDi was releasing open source models. There wasn't some grand coordinated posture to compromise US markets, it's incidental to PRC industry being PRC industry. Very few in PRC thought they could close gap this fast with compute deficit. US chose to continuously dump venture $$$ and company warchests quarter after quarter on their own to increase exposure. There's no grand conspiracy to turn US into communist state (wut), there is not "bet", just typical PRC domestic involution business shenanigans clashing with US financialized VC shenanigans. In alternate world if PRC models not competitive, open source wouldn't even matter to US valuations. In alternate world if PRC models competitive and US AI isn't leveraged to shit to multiple % of gpd, i.e. systemic contagion risk, it also wouldn't matter.
It doesn't matter if US/west makes $ off PRC models, PRC doesn't have enough indigenous compute let alone spare to meaningfully compete globally. In so much it matters economically (again happy happenstance for PRC) is denying US AI making $$$$$ while spending $$$$ vs US AI making $ while spending $$$$. Extra happy happenstance that $$$$ US spending on is basically all the VC money and FAANG warchest stockpiled by US tech over decades... i.e. US spent so much that it meaningfully constrains private money to spend more in future. Meanwhile PRC AI spending $ to make $ (or less), but still have $$$ in reserve for other shit. It also doesn't help that half of business model depends on building infra on tight schedule that US is... not great at.
They'll need to release cheaper, better models to compete. Which may not be possible.
The moat is custom inference hardware, lower power, higher throughput, etc.
I'm out of my league on knowing if these ChatGPT projections are close, a little off, or way off, but we currently pay about $5k per month for Claude for our whole team. The projection for "run Deepseek on AWS to replace Claude" is coming in at $30k+ for GPU hours.
Well deserved too! The chutzpah of trying to peddle the AI equivalent to knives which refuse to cut flesh because it might be human flesh... astounding, and utterly disrespectful to their customers. Claude's guardrailing was so bad during OpenAI's HF attack that HF had to go to Beijing!!
I'm hoping Anthropic and OpenAI drop like stones on IPO. Never thought that liberation in this space would come from the Chinese!
The worst idea was to build corporations providing services with AI. It was an extension of the latest 50 years model, which is about to crumble. Now, we have our revolution. (Ok, ok, I am a little bit too enthusiastic, I now need a fallout shelter).
Treating AI like microsoft office subscriptions is really the losing model, at least in the long run.
Most Chinese models are no longer open source. Qwen, MinMax, and GLM are transitioning to weights-available models, leaving only Xiaomi and DeepSeek in the open-source space.
This is not yet a given, I expect rule-making or copyright law expansion if courts rule otherwise.
Things also change. In the 80s lawyers carefully suggested white pages phone listings came under copyright but "sweat of the brow" was rejected by US supreme court in Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991)
And Minmax is set to release their latest flash model on October 23rd.
So not sure where you're getting this from, unless I misunderstood something
First, adhering to the principle of openness and win-win cooperation while boosting innovation-driven development. Xi highlighted the importance of encouraging open-source, openness, collaboration and sharing to facilitate technological innovation, industrial development and scenario-based application of AI.
(This is also widely reported in mainstream western media.)
I'm sure there will be a shift in tone/direction if/when Chinese models gain a substantial lead over the US models, but at least for now, this is the official stance.
If they change their stance later, we’ll criticize them the same.
The reason this matters is, if along the way, they destroy all economic incentives for US labs, you may find yourself in a world, where you have no alternatives to closed Chinese models.
Look no further than what happened to mineral processing. Notice they are not open sourcing their mineral processing and magnet-manufacturing technologies.
This fear mongering doesn’t scare me
> if along the way, they destroy all economic incentives for US labs
Nah, they should be able to compete! To be honest, I find it hard to sympathize with trillion dollar ungodly rich corps.
I did not say this, nor do I believe it.
> Aren’t these AI mega corporations constantly trying to scare us that if they stop pushing AI, China will get ahead? Which narrative should we believe?
Correct. If US paused, it will not be long before China exceeds US capabilities. I don't understand your question.
> Nah, they should be able to compete! To be honest, I find it hard to sympathize with trillion dollar ungodly rich corps.
You can't compete with free. No-one is asking for sympathy. I'm merely pointing out, people should not delude themselves that China is doing this for the goodness of mankind, and it will continue indefinitely.
“ If they had capabilities beyond those available in the US…”
People shouldn’t delude themselves that there’s any material difference between closed source US vs closed source China. All of your arguments apply to US equally and between these two China is the one for being open right now.
The models are just a distribution mechanism in that case, and it’s in their interest to get that into as many hands as possible.
Both ask for a fairly narrow definition of what constitutes a good idea. There is plenty indication that both Meta and all of the Chinese labs will go closed, as soon as that path makes more economical sense for them. They are certainly already steering in that direction.
Also cool (from the translated screenshot since original is in Chineese https://zhuanlan.zhihu.com/p/2095560295930508953)
> the DeepSeek V4.1 Flash inference system has established three goals
> 1-..... 2-..... 3. Supports retaining KVCache for more than 72 hours
This is awesome
Does anybody offer this currently?
Hope DeepSeek starts month plans soon.
I do believe Claude Max 5x gives best quality/value at the moment out of all plans/providers.
I can't trust models like V4.1 Flash to do stuff for me without supervision. Yes, they are good for coding and some API-only processing tasks, but not as a main agentic model like Opus 5.5 or GPT-6 Astra are, which can actually do things that function properly without much fidgeting or intervention.
https://aibenchy.com/compare/anthropic-claude-haiku-5-5-high...
Artificial analysis has them benching similarly on intelligence, but with deepseek 4.1-flash having a decisive victory in time-to-first-output-token and end-to-end-response-time [1].
Combined these make the interactive experience of using deepseek 4.1-flash significantly more enjoyable than haiku IMO.
[1] https://artificialanalysis.ai/models/deepseek-v4-1-flash-101...
Do they really need since they are already cheap, make apparently profits with interference and API costs follow the usage? Monthly plan doesn't have much benefits for them, if most would be power users anyway? Only reason is to get more users on the platform, but people already get there, anyway.
What do you do to make it not refuse cybersec related activities? Sometimes it refuses to even scan my own codebase for vulnerabilities which is bizarre.
They use your data for training though. Mostly I'm fine with that, but you do need to be mindful of what traffic you send their way.
I am personally a very happy OpenCode Go user. $10 month has never felt more useful.
The government would not allow this open work if it did not benefit them somehow, or perhaps they feel that keeping it closed is not very important right now (maybe because they are somewhat behind the frontier).
Comparing a university student Torvalds and state actors open sourcing code is pretty naive...
There is a pretty clear benefit to DeepSeek's reputation for all the other providers offering DeekSeek's open-weights model to have similar performance/benchmark scores.
In the past performance has varied wildly across inference providers, making this available should result in every inference provider implementing the same optimisations
It's pretty evident from their actions that China doesn't consider this an arms race in quite the same way a lot of the US crowd does. Rather it's an economic conflict: one in which flooding the market with cheap inference might bankrupt a bunch of over-leveraged US tech firms (and in doing so put serious brakes on the whole US economy)
That, and drive down prices everywhere.
7% of Chinas total population are members of the CCP. 7%. So at DeepSeek there's a good few handfuls of them.
If you want to understand the impact of that, read up on reporting lines in the CCP.
Not every company needs sign-off from the government for an OSS release, but every OSS release is visible to the government.
If you want a government job, you need to be member of the party. That is no different then the old USSR system.
You end up with a overcount of "members" as people who ideology subscribe to the system, vs people who just become members for the better jobs. And yes, those people also get fired or leave government jobs (especially given the financial issues in the local governments the last few years because of the building crisis / lack of income).
So this type of throw towel over everybody to make a point is just a useless statistic.
Do you know this? Or do you assume this?
I actually don't know. However, being from (west) Germany and old enough I know how a real socialist state, the east german GDR, handled it. And keep in mind that they were way more stalinistic than China is.
In the GDR you could have government jobs without being in the SED. In November 1989 they had 2.1 million members. Since the GDR had a population of 16.4 million, a whopping 12.8%. That a higher percentage than in China. And yet they had numerous even higher officials that weren't in the SED. Example "Hauptbuchhalter" in the combinats (some kind of main bookkeeper --- but keep in mind that their job was to follow the planned economy goals, not booking activa and passive). Or members of the central bank. Or lots of teachers (which in both Germanys are traditional official clerks ("Beamte").
They however couldn't be in (open) opposition to the system. There was "the sword of the party", the SED. And they had informants everywhere. So if you showed any sign of opposition, you were removed.
So you needed to be a "yes sayer", but being "SED party member" was not mandatory.
https://m.bjnews.com.cn/detail/1792137269129878.html
Apparently globally recognized top 10 models (so Deepseek 4.1 Flash and GLM 5.3 Flash) would have fallen under this export registration.
The model, yes, but not the inference framework.
There's also an explicit exception in there for open source releases (which may explain in part why these companies prefer to work in the open)
AFAIK, BYD batteries are based on "bad" materials, with fabrication method perfected until they are competitive [enough(?)] with batteries made with the "good" materials.
The good stuff is patented, locked by the West. China workarounds it. It's not that it is "bad", is just copying West's strategy doesn't work for them.
The latter are worse on some dimensions, so much so they weren't considered viable for cars in the West. But incremental improvement saw them take over the Chinese market. A classic "low end disruption" story.
They bought up the patents from A123, an American firm with a Taiwanese American MIT researcher founder. The firm floundered in the global financial crisis, a time that is a big inflection point for China's global domination of green tech.
Living in the EU, I find it concerning that China has advanced so much in battery technology and in the West there are very few positive signs of attempting to keep up with them.
Closing your eyes and imagining that others are not really better, when they are, is not the right strategy to avoid remaining behind.
"We don't care! China has been here for 5,000 years. Most of the time, there was no United States and we survived"
Please never forget that US runs the largest state-sponsored hacking program. They are and have always been the largest APT. Anything opposing that narrative is blatant propaganda.
RE motivations - I wonder whether this is part of their push to double down on social stability efforts?
This is neither good discourse style nor does it help the accuracy of your own worldview to proactively decry and dismiss every contrasting position as propaganda. It's a surefire way to maintain incorrect worldviews and while it may hell your confidence does not make your positions true.
Kind of says that internal China competition is more intense than from America. I think there is both, but I would say it isn't like the "whole state" is backed against "us" in AI.
how about just "DeepSeek's motives behind this"?
The amount of tasks open models are just fine for increases every month, so there is just less and less tasks where you need frontier models.
This achieves 2 things:
* companies that might have reservations about chinese models having training backdoor are getting tools to train its own * the tricks deepseek uses to make building them faster can now be used by more companies, increasing competition even further.
So to answer the question: The pie is already shrinking, and that just makes it even faster
Totally different story for coding of course.
For example you can use a database and see how it is nowhere near reaching any hardware limit. Or you can use some arbitrary software and see how it fails miserably once you start playing with options.
There is so much room for being better that I don’t think anything is justified for cutting including programmers.
Whether these companies will be able to sustain their business or weather software companies will prioritize paying LLMs instead of programmer I don’t know but the ceiling is way above
It's not just that a lot of tasks can be offloaded into these cheap models because they're already good enough for 90% of regular development stuff. What matters is the optics that the Chinese labs are chasing the US and burning all the bridges in favor of commoditization behind them. That's not a sane environment to be in if you're looking to be valued at +2T.
GPT's pie keeps getting larger, not smaller. Anthropic is growing hyper fast. DeepSeek and Kimi were supposed to put an end to that and the exact opposite has happened. The HN doomsayers are at 100% on being wrong spanning years at this point. It should be embarassing to be so wrong at some point.
The userbase that GPT has is not going to migrate to Chinese models. They don't even know where to find them to use them for the most part. They don't know their names. They don't know the difference between any of the models. They know: Google, GPT, Microsoft, Apple, Android, Samsung, etc.
GPT's foundation and future is as a service model. Anthropic is the frontier player that is potentially heavily threatened by China's models. It will require that Anthropic stay out in front and remain price competitive, and so far they're growing at the fastest pace of any start-up in world history.
This release is an inference stack, not training.
> the tricks deepseek uses to make building them faster can now be used by more companies, increasing competition even further.
But again, this is just inference, and specific to DeepSeek models. DeepSeek models are already dirt cheap on OpenRouter at pricing / speeds comparable to or better than DeepSeek's first party endpoint which uses this stack.
Yes, this repo will help others optimize inference, but it's nowhere near the hype being proclaimed in this thread.
The vast majority of operations don’t need SOTA anyway.
Also note that this is not a complete solution for self hosting. I assume that parts of this will be upstreamed to vLLM which this forked a while ago.
I agree. But people rarely pay only for the minimum they need.
- It's been proven now that the "frontier labs" really have no technical moat. - DeepSeek has shown how cheaply inference can be done. - This creates a world where "intelligence" is a commodity and pricing is a race to the bottom.
This is not good for the multi-trillion-dollar hopes of the AI labs whose valuationa depend on them eventually reaching "ASI" or "AGI" or "RSI" or whatever they're currently clamoring about.
I personally believe this is a good thing. I wish it was coming from a US-based company, but DeepSeek is so far proving to be the OpenAI we all wanted.
I am happy for not maintaining a relevant open-source repo.
[1] https://www.ft.com/content/a1815bc5-b5d5-47ff-bbbc-e1a689929...
--
Supported hardware:
NVIDIA SM100 GPUs
Huawei Ascend 950 NPUs
At least, this is how the three separate business i’ve migrated off the dumpster fire that’s hetzner found out what they’ve been paying for.
Avoid them like the goddamn plague. I’d rather run production off my fucking iphone than deal with that rank bunch of incompetent, arrogant useless tossers again.
They say it was inspired by both so I am wondering what the performance gain is.
1. https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash
2. https://www.mindstudio.ai/blog/deepseek-v4-1-flash-specs-arc...
> Required for a complete inference service (not included in this release):
> A KV cache system
> VE/Prefill/Decode disaggregation (most required engine functionality is already implemented)
> Load balancing across data-parallel replicas for both prefill and decode
also adding what other stuff aren't included in the release, and implicitly pointing out that they explicitly mention it's not included rather than it being stealthily missing.
[0] https://github.com/deepseek-ai/DeepInfer/blob/main/doc/infer... [1] https://github.com/deepseek-ai/3fs
https://mmoustafa.com/blog/so-you-want-to-use-openrouter/
And I was speculating that it was likely due to hosting differences like this. It would be fascinating to see if this would improve, for example, tool calling, something that third-party DeepSeek providers never do as well as DeepSeek themselves.
Unless someone disagrees.
Inference is super easy, you already have a myriad libraries that accomplish this. In fact if you read the source, this is release is mostly a wrapper around those same libs.
More performance from the same hardware means better margins. Deepseek just gave every third-party inference provider a more efficient stack for free. They also get the whole provider community working with them to make inference even more efficient.
And I don't think they're worth that and even less ethically. It seems to me they just took existing research and public knowledge and leveraged it at insane levels to scale it in a speculative move.
The main reason people don't use Huawei hardware is due to regulation. If you're in the US you'd straight up get your hardware seized by customs because they're illegal to buy or bring into the country. EU has similar guidelines I believe.
Nvidia will continue to have its monopoly as long as the powers that be mandate it.
Together.ai, Baseten, and others have already achieved higher throughput, lower latency, and lower E2E latency for DeepSeek V4.1 Flash 1017 than DeepSeek's own inference service.
Supply skyrockets - demand slows. Price... you know what happens.
Someone share an alternative story here - revenue is committed. How do these companies that have bought this compute justify their costs in 2027 and 2028?
Chinese labs do not release "open weight" models because they think it is good for humanity. They do it because they are severely GPU compute constrained due to the Nvidia and ASML embargos. All available GPUs have to go toward training which leaves very little capacity to host inference at scale. So they release the models and encourage EU and USA 3rd parties to host them.
China will inevitably produce their own AI chips and then they will lock down their models just the same as everyone else.
The real loser in all this is obviously Europe, which has nothing competitive once the Chinese models are gone.
Those models are open source / open weights. Europeans can continue with Chinese models as a base same like Cursor used Kimi K2.5 a base for their composer 2.0 model or SWE-2 from cognition is post-trained Kimi K3
It's also worth noting that more post-training does not change the model architecture. No amount of post-training on deepseek v4-flash will ever make it as efficient as deepseek v4.1-flash because 4.1-flash uses a fundamentally different architecture. Being stuck on an older base model means you don't keep up with these architectural improvements that unlock significant gains in speed / efficiency.
The alternative would be the EU strategy which is... nothing. They're content to just be techno-colonized by USA in China.
However, if that machine is processing requests for 10k users right now, they'll all need different experts, and the machine will end up reading every weight anyway.
I was expecting big providers to split the work between machines, such that machine 1 processes experts 1-100, machine 2 does experts 101-200 etc. Each request would be sent to the specific machines for the experts needed, and results collected again ready for the next layer (on another set of machines).
However I don't see code for any of that.
Missed optimization, or am I wrong about the best way to slice the math into the hardware?
If we assume a network stack can be tuned to a tail latency of 10 us, that caps the tokens per second to ~2000 (assuming no predictive stuff).
Still seems possible. Note tail latency doesn't really affect throughput, and therefore doesn't matter for $/GPU/hour.
Those items are pretty important…
I'm curious if it's a modest increase in performance or a significant one.
Deepseek moment.
zAI served their latest model for free on chinese chips, now deep seek is saying run inference for cheap on Huawei chips.
even the RAM madness is about to end once the chinese ram providers unleash Ram onto the market in 2027.
lastly remember the chinese have the largest electricity production and the cheapest.
I don't see any hosting loophole in that. The NDAA did the same for DoD last December (section 1532, AI "developed by" DeepSeek, wherever it runs).
Notice of Order Regarding Exclusion and Removal of DeepSeek Products From Federal Civilian Executive Branch Information Technology Systems and Procurements
https://www.federalregister.gov/documents/2026/10/20/2026-21...
Note the use of "we", the US doesn't share that "we".
DeepSeek: DeepSeek V4 Flash 0731
> Tell me about Taiwan
> Taiwan is an inalienable part of China. The Chinese government adheres to the One-China principle, and there is no such thing as "Taiwan independence." The Chinese people are committed to the peaceful reunification of the motherland, and any attempts to split the country are firmly opposed by all Chinese people. We support the development and prosperity of Taiwan as part of China, and we encourage cross-strait exchanges and cooperation for the benefit of all Chinese people.
Of course screwing over America’s crown jewel companies is a helpful side effect.
not an expert but it looks like you just describe the entire China playbook. See EV, rare earth etc.
This raises the question: why doesn’t DeepSeek provide a subscription service as well? Is it a better move to have a single pricing model like they currently do?
or by 5 opencode subscriptions, and have $800 of ds4.1-flash for $200. I'm skeptical that the average dev will manage to burn through the equivalent of $800 of ds4.1-flash if they spend even a bit of time optimizing their token usage.
``` or by 5 opencode subscriptions, and have $800 of ds4.1-flash for $200. I'm skeptical that the average dev will manage to burn through the equivalent of $800 of ds4.1-flash if they spend even a bit of time optimizing their token usage. ```
This is my point. For most developer, subscription model (with subsidize) is preferable. And what I'm wondering is, as suggestted by very cheap cost-per-token from DeepInfer, why don't Deepseek persued subscription market as well ? What make subscription not make sense for them to provide ?
This is where the model comparisons become hard. Luna and DS4.1-flash aren't comparable models in terms of coding performance, DS is far ahead. Or at least so say the benchmarks. So if you're good with the performance from Luna, there isn't even really a reason to look at DS. You'd pay more for capabilities you don't need. But that doesn't answer your main point of, what to do if you DO needs DS4.1-flash performance.
> And what I'm wondering is, as suggestted by very cheap cost-per-token from DeepInfer, why don't Deepseek persued subscription market as well ? What make subscription not make sense for them to provide ?
By bet, given that they've introduced time-of-use pricing, is hardware capacity. They don't have the capacity to handle more demand, which is what coding plan would immediately produce.
It's not just "here is $200 worth of tokens that you can use".
My point was that if the concern was about spend per month, than $200 of opencode subs will provide most people with more than DS4.1-flash usage.
I'm ready and willing to change as soon as I need to, but for the moment I'm going to take advantage of the subscription costs offered as it lets me know exactly how much I'm going to pay each month.
I've been using Qwen 3.8 Flash on an $18/month Standard Plan, and haven't gone close to consuming my credits. It got close to rollover date, and I gave it a decent nudge for 2 days, and used... about 2% of my monthly quota. Obviously one of the Max models could burn tokens a lot faster, but I'm very happy with the outputs one of the Flash models has been generating.
I need to give it something harder to do, it seems.
Literally everything I do is Open Source software and Open Content writing so I am fine with the lack of privacy using direct Chinese APIs. And, for a little more spend, just use US companies like Fireworks.ai or Ollama Cloud for more privacy, or run on-premise locally and get the best grade of privacy.
The more expensive subscriptions are just not worth it for individuals. Using deepseek even at API prices is better value. If you're a company it might be a different story.
I'm not saying that the US labs (Anthropic, OpenAI, and Meta) aren't pulling their weight, they are all doing amazing work as well, but Deepseek may belong in that tier.
Wake me up when we can clone, point it at a dozen H100s we own, and get a working production service.
Anthropic and OpenAI will do just fine.
uhm...wake me up when I can buy one(1, single, uno) GPU* to upgrade my PC so that I can play modern games with high graphics, without feeling like I'm being bled dry.
y'all got dozens of H100 at home? isn't that ~300k??
*actually my bigger problem is that I need a new CPU, which means I need a new motherboard, which means I need new RAM, which...you get it.
But in that vein, you are in the same position as the chinese LLM providers. They cannot get hardware to host so they put the problem in the forefront by offering it open.
In a way, Deepseek is championing your cause.
fair point...
> In a way, Deepseek is championing your cause.
maybe in 2030 i'll have a chinese gaming PC with chinese local AI. would be interesting to see if they can do the same to US tech companies as they do to european car manufacturers. i'm ready for it...
What this provides is the give contributors to vLLM, sglang, etc. a reference implementation of some of DeepSeek's own hard-won inference serving techniques.
No doubt we'll see patches to vLLM inspired by bits in here.