Hotz argues that OpenAI and Anthropic's existential-risk messaging is structurally inseparable from their fundraising. The math only works if these companies are selling the first captured superintelligence rather than chat APIs — so the size of the claimed danger must map directly onto the size of the prize to justify $300B+ valuations against single-digit-billion revenue.
This camp pushes back on Hotz as overly cynical toward people who genuinely worry about misalignment. They argue that even if labs benefit financially from doom talk, that financial alignment doesn't automatically make the underlying technical concerns about superintelligence false or unworthy of attention.
This camp views Hotz's framing as a long-overdue articulation of an obvious structural problem: the same organizations publishing 'this could end humanity' white papers are also producing the Series F pitch decks, with no firewall between safety research and capital raises. They see this dual-use messaging as a credibility crisis the industry has refused to confront.
Hotz explicitly stops short of accusing doomers of lying. His more subtle claim is that market selection pressure rewards lab leaders who tell the biggest-stakes story, so the people who rise to run these companies are precisely those whose sincere beliefs happen to justify trillion-dollar valuations — no conspiracy required.
George Hotz — tinygrad founder, ex-comma.ai, the guy who jailbroke the iPhone at 17 — published "The Doom Justifies the Valuation" on his personal blog this weekend. It hit 95 on Hacker News within hours. The thesis is one sentence long and uncomfortable: the existential-risk talk coming out of OpenAI and Anthropic is not a separate stream from their fundraising — it is the fundraising.
The arithmetic is hard to argue with. OpenAI is reportedly raising at valuations north of $300B against revenue in the single-digit billions. Anthropic sits in a similar range. At conventional SaaS multiples — even generous ones for hypergrowth — these numbers are between 30x and 100x too high. The gap only closes under one assumption: that the product being sold is not chat completion APIs but the first commercially captured superintelligence. Hotz's read is that lab CEOs know this, and the discourse follows from it.
He isn't the first to notice the dissonance. What's new is the framing. Hotz argues the doom narrative isn't safety theater in the usual sense — performative caution to deflect regulators. It's something stranger: a coherent marketing position where the size of the danger maps directly onto the size of the prize. If your model could destabilize society, your model is worth a trillion dollars. If it can't, you're a wrapper on transformer math with a 70% gross margin.
The HN thread split predictably. One camp says Hotz is being cynical about people who genuinely worry about misalignment — and that motivated reasoning doesn't make a concern false. The other camp says this finally names the thing they've been watching for two years: a single org producing both the "this could end humanity" white papers and the Series F deck, with no firewall between them. Both are right in part.
The interesting move in Hotz's post is that he doesn't claim the doomers are lying. He claims the selection pressure is doing the work for them. A lab that internally believes its models are mundane raises at SaaS multiples and gets crushed by competitors who tell a bigger story. A lab that internally believes — or successfully convinces itself — that its models are world-historical raises at $300B and hires the best people. Over enough funding rounds, only the second kind survives. The doom isn't a lie; it's a filter.
This matters for how you read everything that comes out of these companies. The OpenAI safety team resignations last year, the Anthropic Responsible Scaling Policy, the constant drip of "we are scared of what we're building" — Hotz isn't saying these are fake. He's saying they are load-bearing for a $300B valuation in a way that makes them impossible to evaluate as pure safety signals. The people writing them have stock.
There's a useful contrast with the previous tech bubble. In 1999, Pets.com didn't claim it would end humanity. It claimed it would sell a lot of dog food. When the numbers didn't show up, the valuation collapsed because there was nothing to fall back on. The AI labs have built a valuation structure where the failure mode of "the technology turned out to be merely useful" looks identical to the failure mode of "we were wrong about everything" — both produce a 90% markdown. Hotz's quiet implication is that this is by design.
The other thing worth naming: Hotz has skin in this game on the opposite side. Tinygrad is an explicit bet that AI compute will commoditize, not centralize — that the future looks like cheap chips running open models, not three labs running god. He's not a neutral observer. But that's also why the post lands. He's running the only kind of AI company whose business model breaks if the doom story is true, and he's pointing out that everyone whose business model requires the doom story is the one telling it.
For anyone shipping on top of OpenAI or Anthropic, the practical read is mundane and important: treat lab roadmap announcements the way you'd treat any vendor pitch where the seller needs the technology to be more transformative than it currently is. When Sam Altman says GPT-6 will be "a country of geniuses in a datacenter," that's not a forecast — it's a fundraising input. Price your own product assumptions accordingly. If your moat depends on the next model being 10x better, your moat depends on a forecast made by someone with billions of dollars riding on the forecast.
This cuts both ways on capacity planning. The labs need the doom story for valuation, but they also need usage growth to justify ongoing rounds. That means API pricing will stay aggressive, free tiers will stay generous, and the squeeze that everyone has been predicting — "they'll jack up prices once you're locked in" — keeps not happening because it would collapse the growth narrative the valuation depends on. Good news for your gross margin. Less good for vendor independence: the longer the subsidy lasts, the more you build on it.
For architecture decisions: the post is a reminder that the open-source frontier (Llama, Qwen, DeepSeek, Mistral) is doing something structurally different from the labs. Meta and the Chinese labs don't need the doom story because they're not raising on it. Their models get released because the model is a feature in a larger business, not the business. If you're picking an inference path for a 3-year product, the calculus isn't just capability — it's which provider's economics still work if AI turns out to be very useful but not transformative. That's the case Hotz is implicitly making tinygrad and the open ecosystem are priced for.
The post will land differently inside the labs than outside. Researchers who took pay cuts to work on alignment didn't sign up to be a line item in a Series F deck, and the framing — even if half right — corrodes the thing it describes. But the structural point is hard to unsee once Hotz has named it: the more credible an AI lab's existential-risk messaging, the more it needs to be true to justify the cap table, and the more its cap table needs it to be true regardless of whether it is. The next twelve months of model releases, safety frameworks, and capability demos should be read through that lens. Not as lies. As load-bearing.
Somehow the Chinese phrases 内卷 https://en.wikipedia.org/wiki/Neijuan and 摆烂 ("let it rot" https://zh-yue.wikipedia.org/wiki/%E6%93%BA%E7%88%9B) are quite appropriately used, even in stylistic tone. Don't know how much Chinese geohot actually spe
Even if Dario and Altman originally believed what they were saying, their scaretactics worked wonders for investment. Their companies are now incredibly incentivized to keep the AI Apocalypse narrative going further and further. It is hard to imagine them stopping, as that may lose them investment.T
I came to the same conclusion.Specifically Anthropic's whole PR has all been about danger, safety, doomerism all to make themselves indirectly more important and central to the debate.Calling meetings in Washington DC in order to let everyone know they made a cyberweapon is part of those PR mov
What's the evidence that the doom narrative is connected to valuations? It seems more like a marketing/recruiting strategy. As the post points out, institutional investors generally think the idea of mass unemployment is BS, and they are investing accordingly.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
I have never understood this line of thinking. Why would a company warn that their technology is dangerous if it wasn't? Looking at the ongoing Mythos export control issue, it's obviously not good for profits. If you assume AGI is possible, you must agree that it would be at the very least