For some reason they need the $100B today at the cost of dilution, rather than additional private rounds and/or using cashflow+profits to continue the advancement.
It hard to say they are doing this now because they need the money now - they are committed to spending close to $1T ($1000B) on datacenters/etc, and the IPO will only raise them around $50B (5% of what they need).
The timing seems more decided by the fact that growth may never look better than it does now as tokenmaxxing gives way to customers looking for ROI and cheaper alternatives to frontier LLMs, such as decision models and non-frontier LLMs which are rapidly becoming good enough for many use cases.
To your second point, it's been commonly theorized that this is the real reason based on how they have tried so hard to emphasize a specific 6 month period for revenue while it immediately started taking a nose-dive from large companies doing a 180 on tokenmaxxing after that short period.
https://en.sedaily.com/world/2026/10/19/anthropic-offers-10-... says, "According to the Financial Times on Thursday, Blackstone was offered a return of more than 10% on subordinated investment in a $60 billion financing designed to secure Google TPUs for Anthropic [...] The double-digit return is a striking concession given that four months ago Anthropic issued about $4.4 billion of debt at 8.5% in a similar deal known as "Project Big Sky."," so we can estimate 10.5%. 8.5% is already well into junk-bond territory.
Can you explain your calculations?
Their current revenue is US$17 billion per year (https://en.wikipedia.org/wiki/Anthropic) out of the US$128 trillion per year world economy (https://en.wikipedia.org/wiki/List_of_countries_by_GDP_(nomi...). If Anthropic were to continue to grow at 47% per year, they would reach the size of the entire world economy in 02049.
If Anthropic were to have to pay a double-junk-bond-level 15% yield to borrow the 1 trillion [US?] dollars that HarHarVeryFunny says they need, they'd need US$150 billion per year in revenues to service that debt. At 47% growth per year, they would reach that level around Q3 02031. That seems like a relatively unambitious timeline to build out the envisioned trillion dollars' worth of data centers and incur that debt.
This is wrong. See downthread.
I have no idea what is going to happen.
As to your gloss of my comment, that wasn't what I meant. They would only be losing money until 02031 if they sold one trillion dollars of bonds in the near future, at 15% yield (something like 45% higher than what they're currently paying). This would be an unreasonably bad way for them to fund several years of datacenter construction. In fact, it's roughly the worst possible way. That's why I did that calculation: however bad their actual cash-flow situation is, it won't be that bad.
What they'll presumably actually do is sell those bonds (or additional shares) over the next several years, so that they're never paying interest on money they didn't yet need to borrow.
The interest on US$60 billion at ≈10.5% is only US$6.3 billion per year, which is about a third of their current revenue. Given that they're currently spending US$42 billion per year, this doesn't seem out of proportion.
I agree that nobody has any idea what the AI landscape is going to look like in five years' time. We don't even know if there will be humans.
However, Anthropic's existence as a company is kind of conditioned on explosive growth anyway. If their growth goes into single digits per year, or goes negative, they're dead even if they don't borrow money.
I agree and this is really my only point. It only works if everything goes according to plan. If it doesn't I think the company fails spectacularly. Not that OpenAI is any different, they are in the same boat, maybe worse off. I don't know if there is any scenario where both can thrive considering their commitments. How much bigger is the addressable market that will pay big to use AI? These are of course questions no one has the answer to, but
Anyway, I have been typing from my phone, so here is my admittedly back of napkin math. Anthropic has $450 billion in near to medium term commitments (through 2030) and another 500 billion after that through 2040 or so, most if not all of which they cannot cancel. So just in per existing commitments they owe already 110 billion+ per year for the next 4 years. I don't know if that is front loaded or back loaded, or evenly spread, but that is the average. That likely does not include all of their costs such as their 3.2 billion annual marketing spend, employee costs, and other annual costs such as their debt servicing which we will put at 6.3 billion per your math, all of which is undoubtedly going to grow. So they need at least 125-150 billion a year minimum for the next four years just to keep the lights on. That is not profit; just break even.
From a revenue growth standpoint they went from 144% QoQ revenue growth in Q2 to 44% QoQ revenue growth Q3. I don't think we can ignore the downward trajectory of their revenue growth. And there are signs that the revenue growth story might get a lot worse in the short term as competition gets more fierce and with companies becoming more discretionary on AI spending meanwhile the cost of computer continues to rise. Short term 15% is probably realistic sustained QoQ growth and after that 5-10% Medium term.
Q1 26 - 5B +132% Q2 26 - 12B +144% Q3 26 - 17B +47% Q4 26 - 20B +20% // Year 26 - 54B Revenue (125B) //
Q1 27 - 23B +15% Q2 27 - 25B +10% Q3 27 - 27B +10% Q4 27 - 30B +10% // Year 27 - 95B Revenue (30B) //
Longer term who knows! Like I said, if they can't keep growing I think think they immediately run into financial and confidence issues and I think the risk of other cheaper models being good enough or OpenAI catching up and surpassing Anthropic is a real risk.
https://explodingtopics.com/blog/anthropic-revenue confirms: "Anthropic made $33.2 billion in revenue in the first nine months of 2026, up from $2.5 billion in the same period of 2025 (Anthropic S-1, November 2026) Anthropic’s Q3 2026 revenue was $17.0 billion, up 1,127% from $1.38 billion in Q3 2025 (Anthropic S-1, November 2026)".
It also gives quarterly revenue figures for 02025 (US$0.376B, US$0.787B, US$1.38B, US$2.04B) and for 02026 (US$4.73B, US$11.6B, US$17.0B), all sourced from their S-1/A. Those last two are where we get the 47% revenue growth per quarter, while the Q1 to Q2 growth was 145%.
So evidently I was erring by using quarterly figures not only for growth but also for revenue.
If that growth were to continue, they really would have no trouble at all in paying US$450 billion of expenses. They'd make US$37 trillion dollars between now and 02031, US$11.9 trillion of it in the last quarter of 02030.
This is unlikely to happen, but I really don't think they're about to implode.
In any case, though, I don't think they have US$450 billion of debt currently, and I don't imagine they'll pay for those commitments out of their revenue; I imagine they'll float bonds just like any other business.
I find it hard to believe US$11.9 trillion of that will go to one AI company. Unless it is Brawndo.
I think you might be trying to implicitly use historical analogies like the advent of electrical utilities in the late 19th century, or of the internet in the late 20th century. Possibly a closer historical analogy is the arrival of the Spanish in the Aztec Empire in 01519.
My calculations above are to understand Anthropic's current and near future financial situation, not to predict a most likely figure for 02030.
The IPO has to happen ASAP because financial conditions are deteriorating for Anthropic and private capital has tapped out. They will need cash, they're out of people to get it from, and the cost to get it goes up every day.
I agree the fundamental question of whether Anthropic can grow past its fixed commitments is still a big unanswered question and that there are a lot of fundamental headwinds that you correctly brought up. But at least the IPO takes the business as it exists today seriously.
This is a bubble, the question is whether it will fill out fast enough not to implode, taking earlier bubbles into account I don't think it will, and if it does it will wreck the economy and then it will still implode, only a bit later.
Sketch me a scenario please where they can grow a couple of orders of magnitude and the rest of the economy survives?
The lore is that Ford created the assembly line and paid his people wages high enough that they could afford the product they were making and effectively jumpstarted a massive upgrade in welfare. This is the exact opposite.
This has been true for 4-5 decades, not just anthropic. I wonder if the is what Warren buffet had in mind when he came up with his rule of never participating in IPOs.
I doubt OpenAI will wait very long - it seems both companies are timing this based on trying to finesse the timing for when they look best, not based on need.
- rsi won't happen immediately, and is inherently unpredictable, so they need to bridge the gap
- it makes it easier to trade their shares/ownership with their suppliers. they're fighting over compute allocation like everyone else.
- their "public benefit" plans have always involved trying to share the upside with the public before it goes "exponential"
LOL. Apparently they haven’t seen the chart put out by their own captured safety org, METR, showing that marginal capability per billion dollars spent is plateauing. There will be no exponential anything.
If you're worried about competitiveness, note that if R&D costs go up, it can sometimes hurt the mid-tier competitors more than the leaders, as it requires a bigger investment to catch up. Chip fabbing economics became worse in R&D spend and we ended up with (almost) only one firm with the scale. No idea if this will apply to AI though.
I've never seen it solve a problem by "thinking harder", but as soon as I give it access to a feedback loop where it has control it can solve a lot of problems, It sort of feels like brute force, but it's not because it's informed about what it focuses on.
theory > change > test > assess > rollback and retry)
I think the bigger issue is how you convince people to give those findings back to the system. Either you lie about what you're training on or you have to give people a stake in the benefits. For example, I still needed some level of expertise to track down my issue, so why should I publish that so Anthropic can train the next model on it and rent the knowledge back to me?
(SCNR)
I guess you're hinting at the difference between model training (including RLHF etc) and models interacting with code, and improving on that.
I guess the real scenario is "simply" a coding agent pointed at code that integrates collected data, likely with the training code using distributed compute. And distributing the training input improvements is surely also already a solved problem, just maybe not OSS.
They probably won't even need to pay them. They'll just make a useful tool in the domains they're focused on, hand it out to a bunch of freelancers and corporates at a discount, and keep it easy enough to teach the agents knowledge work that simply having better agents is sufficient "stake" for people.
They'll call it something beginning with C - for some reason software engineers trust anything that starts with C. Claude Code, or Copilot, or Codex. Something like that.
Ummm... why? This doesn't really make sense. Let's see their current investors:
Google (at least 5%), Amazon (at least 5%), Microsoft, Nvidia, Sequoia Capital, Altimeter Capital, Coatue, Dragoneer, GIC, and ICONIQ.
All of those combined have more economic power (and likely more political power) than most countries in the world.
IMO, this is anthropic conceding they can’t go any further and exiting. Actions always speak louder than words. If RSI was coming, you would be able to secure funding and continue developing independently. There must be nothing evident.
RSI doesn't mean self improvement for free.
Classic industrial learning curves are, arguably, in the same category as RSI, just with a smaller scaling factor.
they still need many billions in compute, while it is not clear if RSI will provide strong revenue gain, so they propose Wall Street to take part in gamble.
VC can wait next 5 years. There are 2 options: they need more capital and can't find it in private markets OR investors think their positions are too risk and want to "derisk"
usually they put money for 8 years.
They have got to the point where private capital markets are tapped out and only public markets can provide the liquidity that they want/need.
It is disappointing just how little recognition there is of this point. For some reason, AI maximalists settled on an exponential model of growth in their mind, and I've yet to see one seriously consider whether RSI might not be that. It's a conversation ender, like bringing up an alternative axiom that cannot be debated, rather than a testable hypothesis.
> Intuitively, it should always be possible to add one more processing unit to increase your intelligence (however it is measured) by a fixed amount.
That's reasonable. The question is how much return you get on your investment, and whether there are "diminishing returns" or "economies of scale" for IQ-per-dollar.
> I agree that exponential growth in particular is debatable, but if intelligence doesn't diverge to infinity then it must have an upper bound, which doesn't seem very believable ... I don't think it can be below linear.
Being able to always increase your intelligence by some amount, at some cost, is compatible with any of "fixed upper bound" (e.g. exponential decay), "sublinear unbounded growth" (e.g. logarithmic), "linear growth," or "superlinear growth" (e.g. exponential growth). Sublinear growth and diminishing returns certainly seem consistent with every self-optimizing system we've seen so far, including agentic AI.
That's false for almost all well defined algorithms out there. Why do you think that should be true for "intelligence"?
Though the truest of true believers also see themselves as one day eliminating scarcity as a concept so it might just be that profit doesn't motivate them.
But when it comes to Anthropic, I am almost certain that it does not make sense to argue based on the typical arguments on maximizing profits. Everything I know about the people working there suggests that they are not primarily motivated by money, but by their mission. I bet they would prefer a lower-strings attached private offering, but have no issue with diluting their own stock, if they think it will serve to move Anthropic further ahead on bringing about aligned AGI. Basically, the diluted money will just be missed donations to EA causes for the most part anyway.
AI can be a "tool", but only so long as you understand how it works and what the implications of its use are. Being an enormously powerful tool implies, the consequences of not understanding and not knowing are proportionally enormous as well.
Currently, LLMs are used practically exclusively by "professionals" who, more or less, know what they're doing. The idea of applying them "everywhere" completely misses the gap in understanding and responsibility for the vast majority.
Not Anthropic specifically, but there's this thing that I recently read:
https://www.nytimes.com/2026/11/09/opinion/ai-anthropic-doom...:
> A.I. company leaders often talk about how their work might lead to human extinction soon. This is astonishing. Even if you wanted to destroy humanity, you wouldn’t say so. If, however, you wanted to save the human species, wouldn’t you stop developing this technology?
> There is an explanation for why some tech leaders, so seemingly worried about an extinction-level event, are plunging ahead with A.I. development: Mankind’s survival, while important to many of them, is less critical than building something they believe to be not only conscious but godlike. A substantial portion of the A.I. community, in celebrating this belief, is quick to support any action or statement that reinforces the idea that the technology is like a deity. They might call for regulation, but only in specific ways — perhaps by deploying more A.I. to control A.I. — that adore the technology’s powers. It is that adoration that is nonnegotiable.
It's by the guy who apparently invented VR and was apparently mentored by Marvin Minsky. So he likely has the kind of access that gives his opinions credibility.
I think it's unlikely that an AI enthusiasts would literally claim "they are building a god," but such an attitude (or something very similar) is implied by their statements and attitudes, e.g.
1. Talk/hope of building something that so exceeds human capability, it's beyond human comprehension and control.
2. Taking existential risks for humanity (by their own statements) to try to get there, against all reason.
3. etc.
I guess they think AI will become smarter than people but I'm not sure that's a god. I've always thought of gods as the dictionary "a supreme being, creator, or a supernatural entity" rather than a souped up chess computer.
That's a marketing page. It sounds like their view is more like "we're building a god, but only we can be trusted to make a good god instead of a devil."
https://www.nytimes.com/2026/11/05/technology/dario-amodei-a...:
> These schools of thought fused for Dr. Amodei into an unyielding conviction that A.I. could be so dangerous that only he and like-minded technologists could be entrusted with building it. That catalyzed his stated efforts to create A.I. before people without concerns about the safety of humanity did, and spurred a desire to control the technology, his friends and colleagues said.
> I guess they think AI will become smarter than people but I'm not sure that's a god. I've always thought of gods as the dictionary "a supreme being, creator, or a supernatural entity" rather than a souped up chess computer.
We'll seeing God as "a supreme being, creator" is only a particular conception of god. IIRC, polytheists often gods as just powerful supernatural entities that are owed deference (because otherwise they might cause you trouble).
> or a supernatural entity" rather than a souped up chess computer
Obviously a AI machine god would fail a "being supernatural" test, but I'm sure the thinking goes that a super-intelligent AI would be so powerful, it's natural-ness wouldn't matter
If you're looking at history for an answer, the end result is somewhere between total destruction and something approaching world domination.
So, not that informative, really.
Anthropic desribes itself as a "public benefit corporation" in its IPO. The irony. Or is it sarcasm?
A god isn't your subordinate. You cannot "program" your wishes into it, like a genie/jinn. They (or their investors) want to make themselves into gods. Over you.
Actually, most LLM usage right now comes from vibe coders who don't even read the code, much less know what it does.
The consequences of "AI use" are manifold and very poorly understood. That doesn't make them less impactful, on the contrary.
Whether you look at "humanoid robots" or incorporeal LLMs, when they are more capable than yourself, you become their dependent counterpart, not their "master".
Does that mean I'm not in control of when and what I use them for?
https://www.tradingkey.com/analysis/stocks/us-stocks/2623001...
This was always the case with ML, even absent RSI. Big bag of weights is a pain to figure out exactly why it can do whatever it is that it does, leading to all kinds of blind spots about how they go wrong.
> Currently, LLMs are used practically exclusively by "professionals" who, more or less, know what they're doing.
Even the people who made them don't really know what they're doing, hence https://www.felonybench.com
He distinguishes between weak recursion (using sotfware to build more software), and strong recursion (handing over model development to AI which does everything), and makes a pretty strong case for the fact that only weak recursion is happening.
This maps to my experience with AI models, they reduce the cost of experimentation massively, but they can't (currently) design experiments themselves.
Even then, training RL is very data hungry, and most profitable tasks don't have good data so even if we got superintelligence (whatever that means), it's entirely possible that progress would still be bottlenecked by lack of high quality training data.
The hope was that weak recursion could produce strong recursion by automating the iteration of ML pipelines.
Theres also the idea of LLMs producing synthetic data to contribute to new models. Perhaps that is not as effective as hoped.
Wasn't model collapse/model decay a big issue since the beginning? i know that RL kinda mitigate the effect, but if this was close to not be an issue anymore, wouldn't we have a paper on that already? All that i've seen on the subject was about mitigation, never about solving the issue.
Sorry if my question is dumb, I need to do some research about the difference between image generator and text generator, i feel like my understanding of LLM is skewed by my usage and understanding of Z-turbo, but i only have that much time in my week.
This never made sense to me. Seems analogous to inbreeding in biological organisms.
More complex synthetic data generates things that could be in the training data but aren't for some reason. For example, machine translation training corpuses often contain a lot of formal language (because formal documents are the ones that get translated), so you might generate a bunch of casual conversation to improve the translation model's ability to handle that.
do you mean ppl using claude to improve claude?
And what is that mission?
I'd almost prefer it if they were motivated by money, then at least we know what we're in for. All this 'mission' stuff seems to effectively amount to 'we're saving the planet, sorry you got in our way'.
Bring about ww3?
sniffles
Because private rounds still dilute you.
And VCs tap out well before $100B.
And they'll need more than $100B to accelerate widespread economic diffusion.
And RSI can't hallucinate gigawatt power plants and TSMC fabs into existence.
This is no longer the days of Apple, Amazon, Intel, Google, NVIDIA and Tesla.
Apple, Amazon, Intel, and NVIDIA all went public in an entirely different world, before the JOBS act.
I'm not necessarily taking the same position as the person you are replying to, but using these companies' IPOs as comparisons for what Anthropic is doing is not exactly Apples to Apples as it were.
I was excited when I got a good metric system implemented because that meant concrete signals for Codex to use for optimization.
It attempted ~5 things. All of them did not improve performance past the baseline.
How much more for the unclear goals, general ML research, etc. My takeaway from all this: You need domain expertise to drive the models.
I've had this just today with both Fable 5.2 and Astra. It was an extremely simple task too, basically just "Here's the extremely detailed ticket (a large ticket, but extremely well documented, one of those tickets where even an intern new to the team could just follow along and get it done in a reasonable timeframe despite the size), here's the MR, let's make sure that we trim some of the fat in this MR from all the iterations we did on this ticket over the last month it's been in review". Both models shat themselves monumentally, started changing random crap all over the place, and in the end this exercise in removing some code only ended up adding a LOT more useless code that was doing fucking nothing, not to mention how unreadable and low quality it all was.
Time and time again, my actual real-world experiences with this tooling is a "meh" at its very best. We have all the harnesses and MCPs and all the other bullshit the AI bros like to shove on you, good CLI tooling, sandboxed environments, basically unlimited access to SotA models, excellent human-written documentation for the context, a mature, stable and battle-tested CI environment which makes changes extremely easy to test. Anything other than the most simple possible translation tasks, and none of these tools ever actually save any time whatsoever.
Every fucking time it's just introducing 3 billion useless unit tests that assert x == x despite who knows how many instructions in the "harness" detailing in excruciating detail how to write tests. Or it's introducing incomprehensible garbage comments that leave you more confused than if there weren't a comment to begin with, or the infinite other plethora of rancid bullshit this tooling leaves behind
Or Repetitive Strain Injury. If LLMs acheive RSI then programmers won't have to type as much, so the risks of RSI will decrease.
These people are incentivised to do it ASAP to mitigate the risk of any kind of crash or scandal or whatever putting their insane wealth at risk
Basically, if you could snap your fingers and make $50M today, or have a 99% chance of making $200M in 5 years time, which would you choose?
All insider and outsider predictions seem to think "true RSI" will really begin in 2027 or 2028.
I just can't imagine this is true when their overreach problem (doing more than you ask), hallucination problem (lying), redefinition problem (doing something completely orthogonal to what you asked), and harm problem (doing something that is exactly the opposite of what you asked) are all far too high at this point, and by most measures - all of the recent models are worse on most (or all) of these metrics.
They are giving up here in RLHF to benchmaxx the metrics people are looking at to pretend like they're making huge gains, when they're largely running in circles.
They are getting better at pass@1 at the expense of far worse p90s on real agentic tasks.
You can do the math on the dilution, which will be on the order of ~10%, give or take. If the new shares are worth more than existing shareholders paid for theirs, each share can be worth more even though the percentage is smaller.
They need resources and $100B can buy them that. I don't think you need a conspiracy theory to explain that.
The whole idea of self-improving machine equals immediate singularity is just wrong and ridiculous. I don't know why so many people are attached to this idea. Perhaps just because "singularity" is such a cool sci-fi word?
No humans in the loop.
Deep learning had no humans in the loop, but there's always a bigger outer loop with a human somewhere.
You can place bets that these companies will be failures if you want.
Comment from a perma bear. So take it with a grain of salt. lol
2nd bullish thing seems to be that chinese labs are having difficulty in distilling latest frontier models. No deepseek pro for some time now!?
also government, defence, financial and similar institutions and companies will almost sure never use chinese open weights models.
Without this IPO they'll need to either stop the bleeding or coax yet another multi-billion funds (which most of them are already balls deep in AI)
fund or bust
It also doesn’t stop investors from being happy liquidating at an already good return, especially since they can choose to liquidate partially.
EDIT: Maybe before downvoting read the actual evaluations?
https://pbs.twimg.com/media/HWYbAX9b0AA-SWt?format=jpg&name=...
The gains between generations are smaller because model releases occur more frequently (this is primarily driven by them taking more frequent snapshots of the base model during training).
The title of your chart is literally: "As of October 2026, benchmark scores appear to be increasing at 1.9x their pre-October 2025 rate [95% CI 1.5x-2.8x] - when including recent internal models."
Fable is great, certainly, but it's no better at writing or reviewing code than the top-end Carbon or Astra models. To the extent Anthropic has a moat, it's Claude Code. Obviously that's not much of a moat, but their competitors seem content to camp on the other side of the drawbridge for some reason.
Codex is pretty close in a lot of ways, as the other poster says, but Astra consumes tokens at a higher rate, and when it compacts it completely lobotomizes itself. That used to be a problem in Claude Code as well, but it has been largely solved (although at the cost of making the compaction take much longer.)
Antigravity reminds me of the old comic strip where Firefox and Chrome are duking it out, while Internet Explorer is sitting in the corner wearing a helmet and eating out of a jar of paste.
Normally I run the models that the harness is intended to support. You can't run Claude models anywhere else besides Anthropic's own CLI and desktop harnesses -- or at least you're not supposed to -- without paying out the nose for API tokens.
I do use DeepSeek 4.1 Flash 1017 in Claude Code when I want an abliterated model for security-related work or unhobbled browser control, though. I know it's possible to get OpenAI models working in CC, but I haven't bothered. Gemini I'm not sure about.
If you’ve ever don’t consulting work, you know how dangerous it is when the majority of your revenue comes from a single customer. You stop being independent, you stop having your own set of values, and if anything happens to that company, you’ve already proven you lack the ability to court as large a contract from other customers.
It always ends in a loss of identity and huge layoffs.
Making vendors product triple their current production rate for the next five years isn’t even possible to start this year, because they have to build facilities and train employees. Nobody runs businesses at 50% capacity anymore. That’s considered wasteful, even stealing profits from your shareholders.
They can’t even get memory chip companies on board with their koolaid.
Both Anthropic and OpenAI seem to be burning cash as if it was a game of startup chicken. Of course they all desperately need cash.