New Constructs is bearish on every tech IPO. Easy to reinforce your bias by selectively looking at instances where they were right, like WeWork, but they also predicted failed IPOs for Spotify (+219%), Uber (+60%), CrowdStrike (+3136%), Cloudflare (+2306%), Palantir (+1991%), Snowflake (+51%), DoorDash (+93%), Robinhood (+187%), ARM (+422%), Instacart (+59%), Reddit (+370%), CoreWeave (+105%), SpaceX (+20%) and lots more.<p>So whatever your stance on Anthropic this particular analysis is safe to ignore.
You dont need to criticize good research, you can just recognize the market can stay gullible, for longer than you stay solvent.<p>I like to think you can enjoy the ride wave as you are propped above the heads of an hysterical crowd, as they rush to the abyss...Just find some solid tree branch you can latch on, when they realize where the road ends.
I don't see how Anthropic can become profitable enough to justify its investments unless they can either stop training models ("Pace the Frontier!!1"), or magically eliminate open source competition so that their volume x margin on inference is high enough to dwarf the training costs.<p>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.
Their inference margins are already high, the thing bringing their net down isn't training per se, it's the fact that they have to train the next iteration right away due to competition, otherwise an extended inference cycle would've eventually recouped training costs.<p>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.<p>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.
>> Their inference margins are already high<p>So they make a profit on each burger they sell you, but however they keep limiting how many they are willing to sell you?
> it's the fact that they have to train the next iteration right away due to competition<p>That it would be profitable if only there were no competition is not a really strong starting point for an IPO, though, is it?<p>(It is, however, the winner-takes-all world view Anthropic still seems to operate with)
Their competition has the same problem. They're engaged in a giant game of chicken (or giant game of "who runs out of ability to raise more cheap capital first") with respect to overspending on training relative to inference revenue.<p>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.<p>If you think <i>either</i> that Anthropic is in a strong enough position to come out on top <i>or</i> 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, <i>then</i> the IPO might be worth putting money on.<p>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 <i>everything</i>; 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)
> Their competition has the same problem. They're engaged in a giant game of chicken (or giant game of "who runs out of ability to raise more cheap capital first") with respect to overspending on training relative to inference revenue.<p>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.<p>(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 <i>aligned</i> where this meta-chicken game is concerned)
The rush towards open weight models defeats this thesis. There’s no prize for being “ahead.” That’s academically interesting but a terrible business plan. 99% of AI use cases don’t need the latest and greatest model, and businesses increasingly understand this.
That's like saying "forklifts are a terrible idea, because a forklift is excessive for lifting 99% of the items that I can lift". See the problem?<p>"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.<p>This is a big part of what the frontier buys you.<p>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.
Except that’s simply not how it works. Companies just have one forklift. Yes it’s too much for most use cases but you don’t have a garage full of different sized forklifts. There’s a big cost to keeping all these extra forklifts on standby.<p>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.<p>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.
> inference - which is profitable in isolation<p>They have not actually demonstrated this, really.
They don’t care if they get to sell more equity.
If the models get good enough, the demand for them will soar.<p>These companies believe they're a year from AGI, and that's how they'll be profitable.
Question is really how many companies will stay due to inertia while competition catches up
Unlike the dot-com era, this is a bubble that everyone saw coming.<p>Like dot-com, there are many long-term winners, but cash hungry frontier labs are not among them.<p>Prediction:<p>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.<p>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)<p>[1] Everyone already knows that, according to SpaceX, > 80% of SpaceX's valuation is in AI, not in space stuff, right?
Nvidia probably doesn't really have really serious liabilities as long as it has <i>actually produced</i> all the GPUs it has sold on a buy-and-hold basis.<p>It's fabless, right? So they haven't even built out wasted manufacturing capacity.<p>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.<p>The fact that <i>nobody else is making real money</i>, 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.
My prediction is the same. I think MS, Google, and Amazon will be the winners here. They might take a hit when OpenAI and Anthropic blow up, bu they can absorb it, and are well positioned to provide those services in a reasonably sustainable fashion. I suspect the costs of using AI will be considerably higher at that point (subscriptions will be extremely limited or just non-existent). Meta does not benefit in this case since it doesn't provide compute, but they have money to waste (as they did with Metaverse, VR, etc).
Interesting take. How high is that probability within the next five years, do you think?
Given <a href="https://www.abc.net.au/news/2026-10-10/how-firmus-the-biggest-ipo-in-three-decades-fell-over/107247100" rel="nofollow">https://www.abc.net.au/news/2026-10-10/how-firmus-the-bigges...</a> and the above, I'd say we're already primed for a fall. Given 5 years is a pretty long time (that's 3 US elections cycles where politics can increase uncertainty about AI), and the length it took for the subprime market to create the GFC (see e.g. <a href="https://en.wikipedia.org/wiki/File:Subprime_mortgage_originations,_1996-2008.GIF" rel="nofollow">https://en.wikipedia.org/wiki/File:Subprime_mortgage_origina...</a>) looks to be about 3-5 years, I'd be 99% sure the crash would happen by then (with the 1% covering both the uncertainty in the cycle and something odd happening).
You're of course speaking of a world that doesn't manipulate the cost of memory.<p>Hard to run local modals when sam altman continues to drive up the cost of compute.
> The New Constructs report also notes that Anthropic’s assertion that AI could pose “a catastrophic or existential risk to humanity” is another reason why investors should avoid the IPO.<p>So, we go from an existential threat to basically unprofitable business. I can only see Amodei and Altman just as the next Holmes.
> In an interview with CNBC in 2021, Trainer acknowledged that “crazy stuff happens” and New Constructs doesn’t always get it right.<p>> “I can’t let that bother me,” he said at the time. “I have to stay true to what I think is right.”<p>Monkey remains true to darts, dartboard.
I posted other (different) research into the IPO yesterday: <a href="https://news.ycombinator.com/item?id=50033644">https://news.ycombinator.com/item?id=50033644</a>
“So, what you're telling me is that the music is about to stop and we're going to be left holding the biggest bag of odorous excrement ever assembled in the history of capitalism.”<p>A brilliant line from the movie Margin Call (a fictional portrait of the actual events of the 2008 crisis) that appears likely to get reused in upcoming AI bubble burst movies.
New Constructs takes a similar position on all mega IPOs. Sometimes they're right, but sometimes they're wrong. For example, they insisted Uber's IPO valuation was too high due to unsustainable losses, hidden liabilities, etc. Initially they were proved right when the post-IPO stock price plummeted, but in the long term Uber has performed well, become very profitable, and now trades well above its IPO valuation.<p>I'm expecting a similar story to play out with Anthropic: The stock price will come down significantly after the IPO, resulting in lots of headlines about the AI bubble bursting. But in the long term it'll climb up again and investors who stick around will do very well.
“Now trades well above the IPO valuation” does not mean something would have been a good investment at the IPO price given opportunity cost and the time value of money. Additionally, the current price could equally not represent good value.
Yeah. Look at SpaceX. Dropped to $108 from an Ipo price of $135, but it's currently at $162 which is... fine?
I suspect SpaceX is going to be the opposite of what will happen with Anthropic. With SpaceX I think the overwhelming majority of people didn't initially understand what Starship is, how far along it is, or what it even means in terms of its impact on the pathway of getting humans to be a multiplanetary species. So it was greatly undervalued and zoomed upward.<p>With Anthropic I think a lot of people haven't really realized the trend line of cost of "AI" (both in literal $ terms, and compute costs for local operation) or what open weight systems will do to the market in the "long run", where long run in this case means the next year or two. I don't think any LLM company is going to come out a winner in the longrun, because the price for LLM output is going to approach $0 with no chance for a moat whatsoever.
> the overwhelming majority of people didn't initially understand what Starship is, how far along it is, or what it even means ... So it was greatly undervalued<p>I don't think anyone was undervaluing the "space" segment of SpaceX. It's the clear leader in launch frequency and cost-to-orbit among the new wave of space firms, and shows no sign of losing that position.<p>Where the IPO raised a bunch of red flags was SpaceX suddenly acquiring Elon's AI firm, which had a significantly more nebulous basis for its valuation...
> the pathway of getting humans to be a multiplanetary species<p>I am afraid that you're suffering from Flavor-Aid poisoning here. We need a reality check about "multiplanetary colonization".<p>I've been following human spaceflight since I was born. I've actually participated in robotic spaceflight. And I'm here to say that it is robots who are destined to be a multiplanetary species. The robots have already successfully escaped our solar system and extensively explored the Moon, Mars, and elsewhere.<p>Many unscientific people believe that it's harder to colonize the Moon or Mars because they are farther than Africa or Rio de Janeiro is from Rome. But mere distance is not really your biggest obstacle.<p>I've also been following the ISS since its inception, and they have empirically disproven the feasibility of colonizing planets. If we succeed in establishing a crewed Moon base, it will be an ultra-Herculean effort, and it's on the edge of feasibility.<p>But Mars? Venus? Asteroids? Europa? You really think humans are going to live there?<p>The reasons that ISS has disproven this are manifold, but I'll point out two key points:<p>Their supply-chain is only about 8-48 hours, and they can stash large amounts of provisions up there. Their escape route is also only 8 hours and they're back on Earth in a jiffy. None of this is true for any body outside of Earth orbit. The Apollo missions arrived on the Moon and returned safely, by burning, jettisoning, consuming, and abandoning nearly everything they started with!!<p>A 15th-century expedition to Africa or Brazil means that you can fish in the ocean, breathe the air, drink rainwater along the way. An expedition to the Moon puts you in an airtight bubble with no stops, no rescue, no resupply, and most importantly, no breathing on the way, nor once you get there.<p>The ISS has also demonstrated the severe toll on the human physique in microgravity conditions. Watch astronauts who land or splash down. They are carted off in gurneys. They are literally unable to walk unassisted. NASA restricts returnees from driving or even jogging. The astronauts describe having to "relearn gravity" activities like brushing your teeth and showering. Any astronaut arriving on Mars would be as helpless as a toddler! Do you want to colonize Mars with toddlers? Because all the treadmills and workouts along the way won't make them fit for purpose on arrival.<p>So between here and Mars it's minimum six months, no escape, no rescues, no takey-backsies. We could send humans to Mars if we didn't want them to survive or return. I mean, no problem. There's no air to breathe on Mars; no water to drink; no fuel to burn; no fish to fish nor deer to hunt; no steel to build a shelter. You pack it all into the spacecraft, newbie?<p>But NASA and the rest of the world has amply demonstrated just how capable robots are where humans fail. Robots are going to inherit the solar system. Robots are in 2026, the <i>de facto</i> interplanetary species, <i>sine qua non</i>. My heart is warmed by these human achievements. I, for one, am happy to live and die on Earth. I think that every American should be grateful for Earth, and not permit our oligarchs to cheat and lie to us with fairytales of escaping our Earth.
Everything you’re saying is true but it’s nevertheless going to happen anyway.
I fully agree with your points: the hard, unyielding reality which most Trekkie fanatics refuse to acknowledge, except for one: "severe toll on the human physique in microgravity conditions."<p>That's easily fixed!<p>Spinning a space station for gravity is straightforward, it's just that the ISS wasn't designed for it, and NASA didn't have the budget to build a second, bigger space station with "extra engineering challenges" on top.<p>It's definitely practical though, it's just a matter of launching enough mass to orbit.<p>Like you said, Starship is going to change a lot of things, starting with space station design.
> spinning ... for gravity<p>Yeah, I mean, I get that the spinning has long been a viable proposition. Perhaps if we send a spinning expedition to Mars, the astronauts won't be toddlers when they arrive on a barren, hostile, resource-free world. Yet, that is a lot of extra mass; that is a lot of extra fuel; that is a lot of extra "stuff goes wrong" risk inherent in that design.<p>But it's unprecedented, untried, and not currently in any pipeline, AFAIK. I believe there were a few experimental spinning crewed flights. But does Starship spin and simulate gravity? Is there any human spinning spacecraft extant, in production, or even being designed?<p>They don't need to spin their way to the Moon, so it would be unlikely that any Artemis or future crewed Moon mission spins at all. But if spinning our way to Mars will solve an otherwise intractable set of problems, I suppose we'd better get started on proving it.
While I think SpaceX is still horribly over-valued they at least have a diverse line of businesses that are very hard to copy.<p>Anthropic is a one trick pony and the world is discovering lots of other cheap ponies that do the same trick.
Maybe the portion of Anthropic that produces Opus/Sonnet can be spun out for us devs at a reasonable valuation?<p>Keep all the Mythos/Projects/etc. hype/feature-creep crap in some other entity and value that at $5 tn or something if you wish.
Just spin the golden-egg production part out of the goose, then we can cook and eat the rest.
Anthropic's small models are almost certainly distilled from their larger ones. And their biggest training expense are going to be experimental models that aren't even intended for release, just to test out some idea.
This reminds me of all the people asking for AWS to be spun out of Amazon.
[dead]
The model is engine, the harness is the car Anthropic are selling.<p>They are a service company, selling home-grown, mass-produced, conveniently packaged intelligence at a well established price for consumer. Their API pricing might just be for value proposition and tax deduction purposes.<p>Every tech company, tech department, startup that I know is using claude code to deliver. The boutique deepseeks, qwens, glms and kimis are nice, but ain't production ready in 98% of cases. All the pi's, opencodes, etc are nice, but much less enterprise friendly. And google just outright sucks in comparison. There's a use case for each, but not at the scale Anthropic is doing.<p>Only real competition - OpenAI - is close, but they're burning money to stay there, which imo is unsustainable long term.<p>Is Anthropic worth $2T? Can't say. But boy they have a good opportunity to test this, and they may not have a similar chance in the future.<p>I think they have a fair chance at this.
I've tried a bunch of LLMs, and I agree that Claude is currently the best at coding, but the gap isn't huge. Gemini 3.8 Flash is surprisingly good and I'm looking forward to trying Gemini 4. The open weights models like Kimi, GLM, etc. are a little behind in quality I think but not massively.<p>With harnesses I see little difference between what Zed or Pi gives me and what Claude Code gives me. Enterprise features? The AI can help code those in a week.<p>AI itself is a great breaker of moats.<p>The question isn't whether they lead today. It's whether there's any kind of moat. Without a moat, frontier model development is an industry like chip fabrication or large volume manufacturing. Those industries are a red queen's race where if you stop innovating for even a short while you sink. Ask Intel what happened when they decided to coast for a few years.<p>Chip fabs and bulk manufacturers make a lot of money, sure, but they don't command such a multiple. Anthropic is priced like they're going to be to AI what Google was to web search or Apple is to (US at least) mobile and higher-end PCs.<p>I don't see it. I think they'll be a big player but <i>that</i> big?<p>Of course... maybe if they suddenly have that much cash they'll be able to carpet bomb the entire world with AI doom Rationalist ideology and buy their regulatory capture and outlaw their competition.