For the IPO, I assume they're going to try to get everyone to focus on inference - which is profitable in isolation. But I don't know how they'll distract the market from the other half of the equation.
The bet is that cost per intelligence will keep going down and demand for intelligence will grow at the same time, if this bet holds, a user could get the same or more done with a cheaper model (running-cost wise) while paying the same subscription and the lab gets improved margins.
Open source is a challenge but their current moat is the fact that they're ahead, and as long as they stay ahead they'll be able to charge a premium.
"AI use cases" is a moving target! If you try to define "use cases" by "what AIs today can be used to do", you miss things AIs would do in the future. The more capable and price-performant AIs get, the more there are places where AI has a role.
This is a big part of what the frontier buys you.
The other big part is that going from a local AI to a frontier model often makes the difference between "this use case works 60% of the time, good tech demo" and "this use case works 99% of the time, good business decision". The difference between "clunky and needs human in the loop" and "just works" is worth paying the API price.
Models, however, are completely interchangeable on demand. That’s why companies are building these realtime internal marketplaces to route tokens to different models. Anthropic may win the occasional “bleeding edge” use case but the money is in the day to day use cases that get routed to the cheap-but-works options. There’s zero cost to having all the forklift sizes at your disposal and only using the size needed when it’s truly needed.
Anthropic is trying to sell the biggest baddest forklift in a world where companies can swap out cheaper ones easily and on demand. Hence why Anthropic doesn’t appear to have a viable business model in its current form.
And crane, despite being bigger, more powerful and more awesome (they lift skyscrapers and do things nothing else can do), the Forklift is actually more useful to more people.
I think the argument carries solid weight. Making bigger and better things doesn't necessarily lead to bigger and better businesses.
This is the problem.
While the frontier labs will always have a market for AI that can do things the other models can't, if the sliver of work that can be done only by the frontier models is narrow enough, then what they spend to train those frontier models may not let them capture all that much extra revenue.
I absolutely think there will be a market for frontier models pushing further and further.
But it's not yet a given whether or not being ahead on capabilities is going to buy them all that much more higher earnings relative to lower tier providers.
If it stays the case that going from local AI to a frontier model makes the difference between 60% and 99%, then they probably will, but there's nothing to suggest local or open model, or second-tier closed models won't keep moving ahead as well, and eat away at the use cases where only a frontier model will do.
Do you think that we are anywhere near that point?
We've already seeing things - like what's happening with AI reverse engineering now. I'm a reverse engineer - and I know that the reason why RE was used as "last resort option" was that expert reverse engineers are rare, expensive, and work slow.
But now that a lot of RE workloads can be tackled by AI instead of a human? The demand for reverse engineering is at a never before seen high. AI made reverse engineering much more viable - and thus, a lot more reverse engineering is happening now.
There are myriads of fields like this - things that go overlooked not because no one wants them, but because the economics of human expertise make them nonviable in most cases. As AI gets better, it can begin to enter those fields.
Except you, of course, need to push the frontier to do it.
Much of humanity doesn’t think at all btw - that’s why they’ll never pay for these products and Google search had to be free.
Furthermore two things could be true a) market for frontier is large b) Anthropic still continues to generate net loss’s
What’s your goal here
Interesting. Are you an Anthropic investor?
And that, in turn, makes AI even more valuable if it can produce thinking on demand.
That's like saying if 99% of birds don't have huge colorful tails that makes peacocks very valuable. In each case it may or may not be true but you are just correlating. (i.e. It's possible that intelligence just isn't as valuable as 'intelligent' people think it is)
Researchers specifically were never highly paid. And positions were competitive - there were way more people willing and able to do those positions then positions themselves. Same goes for art btw.
I could be wrong, of course.
No, I don't. But for any given task there is a ceiling above which for that task there is no value in additional intelligence. E.g. it takes very little intelligence (none, in fact) to sort a list. More intelligence does not result in a more sorted list.
And so for a huge proportion of what we do there is a ceiling of value that can be extracted by selling AI tools to do any given task.
Combine that with there being at any given time a ceiling of available capital to be apportioned to buy intelligence to carry out the remaining tasks, and the floor of tasks that non-frontier models can do well enough rising, and there will at any given time be a finite amount of capital that will be accessible to frontier models.
It is by no means given that that slice of capital will grow rapidly enough for frontier providers to be able to create an attractive financial return. That depends on frontier labs being able to continue to deliver sufficiently more value than cheaper models at any given time that enough people won't simply wait.
I'm extremely bullish on AI in terms of its ability to transform society over time, but that does not automatically translate into assuming that the current frontier labs are a safe investment bet.
If "sort a list" becomes a trivial task, everything that was bottlenecked on "sort a list" is unbottlenecked, and other parts of the process become bottlenecks instead. Thus, the demand for non-sorting tasks grows. The saturation of low end feeds the value into the high end.
Sometimes it is. But that is also irrelevant. Replace "sort a list" with any given goal and the argument remains the same.
> The saturation of low end feeds the value into the high end.
But that process is not immediate. There's a reason I was careful to include "at any given time" in my comment.
For the purpose of whether or not a frontier lab is a good investment, it doesn't necessarily help them if there is an ever rising demand for intelligence if their market is eaten from below just as fast by cheaper models from competitors catching up.
It could be they manage to out-pace their competitors sufficiently that they will produce returns high enough to justify their valuations.
But irrespective of whether the demand for more and more intelligence continues to rise perpetually if 2nd tier labs hollows out the bottom of their market fast enough, it could equally be that most of the market ends up being low-margin, highly-competitive inference that won't justify their valuations.
Given the size of their valuations, even if they turn out to be tremendously profitable businesses, it's still not a given they'll give a good return on investment for investors getting in now.
I'm not saying they won't. I'm saying we can't easily judge that based on just assuming growth at the high end will continue without understanding what their moat is likely to look like.
Your assumption here is that the other parts of the process are intelligence. In the real world they usually are not e.g. raw materials, politics, zoning, manual labor, clinical trials, traffic, energy ... And usually those things are such huge bottlenecks that Amdahl's Law renders the speed of delivery of the intelligence part irrelavant.
There is no business model for competing companies to build forklifts that can lift an infinite amount.
That is not to say attempting to do so would not find new use cases for such powerful forklifts. The problem is the cost of building ever increasingly powerful forklifts while the TAM for that power shrinks.
You don't even have to bring in the enormous difference between forklift rusting theta vs gpu theta.
That it would be profitable if only there were no competition is not a really strong starting point for an IPO, though, is it?
(It is, however, the winner-takes-all world view Anthropic still seems to operate with)
So, yes, it's high risk - you're basically betting on who will be in the strongest position when one or more of the competitors are forced to slow spending and the rest are able to ease off on theirs as well as a consequence.
If you think either that Anthropic is in a strong enough position to come out on top or that we'll end up with a detente like with the big cloud providers where they big providers feel they have a strong enough position to beat off upstarts and collectively basically stop competing other than on the margins because they all know they'll earn more by protecting their margins than by internal competition, then the IPO might be worth putting money on.
I think the latter scenario is slightly lower risk, but still risky given how small the moat appears to be vs. the Chinese labs (I say this as I just had Kimi re-design a website in 10 minutes by just giving it the URL to the current one and tell it to replicate everything; they might not be as smart as the frontier models yet, but they're capable enough that they're taking over an increasing proportion of my token budget)
It's a sort of weird chicken game in that there's a meta-chicken game going on too — they are both, sort of separately, betting that the outcome of their chicken game will be too devastating for the US economy for it to be allowed to happen.
(I say "sort of separately". But the one guy used to work for the other guy. You know how people who split up from partners spend the next few years trying to shake their ex-partner's framing of the world? They know each other's thinking; they are aligned where this meta-chicken game is concerned)
Especially if it is finely calibrated enough that they get to continue at a slower pace, but end up with enough expensive red tape to slow down upstart competition more because of high compliance costs.
If they can't beat OpenAI etc., the second best is to near-freeze market shares in place at a lower rate of investment - hence my comparison to cloud providers where none of the big providers makes too large reductions in prices etc. because they all know they'll all lose on a price war.
So they make a profit on each burger they sell you, but however they keep limiting how many they are willing to sell you?
Being ahead isn't much of a moat in a game where playing catch-up is significantly easier than being ahead. Also, how much ahead of Chinese labs are they at best? A year? Two? That's a moat for ants!
But profitable enough to justify the ludicrous valuation?
I think there will be a few forces driving prices lower if anything.
Everyone is currently looking at their spend on AI and the revenue growth for their own companies that this spend is not yielding.
I also think that if there is any competition, they won’t be able to lock in prices where they stand. Especially with open source models.
the problem is that they are not ahead. they are ahead on producing one-shot demos, which are useless party tricks, but when it comes to shipping real products these large models are just a big expensive waste of time and what you need is deterministic systems that integrate small efficient purpose built models in the places where they are actually the best solution for the problem.
the anthropic game plan was to wrap the entire software industry up in a magic box called "AI" and then rule the world but these people are delusional and their house of cards is rapidly collapsing. the technology is real, but they didn't invent it and they don't own it. it is kind of amazing to think of a $2T company with zero IP ... they didn't invent a single thing that they are selling, and for the most part they are implementing it worse, and serving it less efficiently than at least a half-dozen other competitors.
In intrinsic valuation, the value of the operating assets is determined by ebit(1-t) - reinvestment.
Gross profit doesn’t mean a whole lot if a) your operating income is negative b) your reinvestment needs are mega. You have to make some interesting assumptions Re. Anthropic to get a 2T valuation.
I almost have to hold my nose reading finance posts on here.
Like dot-com, there are many long-term winners, but cash hungry frontier labs are not among them.
Prediction:
OpenAI, Anthropic, Oracle, and SpaceX[1] go down and assets be sold or companies acquired for a fraction of their valuation. The same goes for most neoclouds. Profitable "old" tech companies like MS, Google, Meta, and Amazon are fine and scoop up the valuable parts for pennies.
Nvidia will be in Cisco's position (Cisco was the world's most valuable company in the dot-com era). Both are good companies that are not going anywhere, but stock prices get decimated when markets collapse (Nvidia liabilities are not high enough to take company down, they are mostly tied to there not being a bubble and data centers are getting built)
[1] Everyone already knows that, according to SpaceX, > 80% of SpaceX's valuation is in AI, not in space stuff, right?
It's fabless, right? So they haven't even built out wasted manufacturing capacity.
As long as the GPUs they have sold are really sitting in a warehouse in Taiwan or somewhere in the USA, they are making properly concrete money.
The fact that nobody else is making real money, that will eventually come back to bite them, but Nvidia will just see its business shrink; it will have to scale back if the bubble bursts, but it will get out with much of that money intact.
For September:
> In IT, semiconductor exports rose 262.8 percent to $60.30 billion, exceeding $60.0 billion for the first time as memory demand continued to grow, with export volumes continuing to expand and contract prices maintaining their upward trend.
https://english.motir.go.kr/eng/article/EATCLdfa319ada/2743/...
Not too shabby for a rainy day fund.
According to SEC filings Nvidia has $100 Billion in total commitments. They are mostly purchase obligations, advance supply agreements, and multi-billion-dollar prepayments for fab capacity (wafers, HBM, packaging, substrates and interconnects).
A general rule in the AI bubble appears to be this: Nvidia aren't waiting to get paid for the things they make. Rather, they are being paid very up front for things that cannot be installed.
> Unlike the dot-com era, this is a bubble that everyone saw coming.
As far as I recall, the dot-com bubble was widely seen as such prior to it bursting. E.g. Greenspan and Buffet were both quite vocal about the irrationality of the values being assigned to companies.
The vc’s have to be allowed to cash out - so that money remains a strategic weapon for US venture funds over any other nation re. Funding startups in future - esp capital intensive.
The valuation is a joke but they will be allowed to go on and cash out - just like the investors of spacex.
Hard to run local modals when sam altman continues to drive up the cost of compute.
These companies believe they're a year from AGI, and that's how they'll be profitable.
What I don't get is that if Company A is a year from AGI (I don't think that's true, but then again, I don't think anyone even has a good definition of AGI), then surely Companies B–ZZ are only a few years (at most!) from AGI as well. There is no moat. Or if there is a moat, the moat is hardware scarcity – but that's systemically outside of the control of these companies.
They have not actually demonstrated this, really.