More or Less · Topics · Episode #159

Sam's Stealth Bomber Thesis: Frontier Labs as Weapons Against Bot Rebellions

From #159 — Did OpenAI Stage Its Own Cyber Incident? | Stripe Buying Paypal, Frontier Model Economics, Peptides · July 24, 2026 · 6:21 min · Jessica Lessin, Dave Morin, Brit Morin, Sam Lessin

Clip from the full recording, 31:52–38:13 · download mp4 · watch at this point on YouTube

Sam floats his most out-there idea yet: frontier labs may not survive on being used at all — they survive as impossibly expensive "stealth bombers" you only call when open-source agents go rogue, and Dave thinks he might actually be right.

Sam opens with a blunt financial thesis he says he’s held longer and more consistently than anyone on the pod: from a business and economics perspective, all the big labs are “fucked” long-term, even though they’ll keep trading well and making people money in the short term — great technology doesn’t automatically mean a great business, “the French make great art, that doesn’t make it a great business.” His actual thought experiment: what if frontier labs’ real business model isn’t being used day-to-day at all? Everyone runs cheap, good-enough open source for 99% of work — like having an IQ-110 assistant instead of 100 — but every so often there’s a “bot rebellion,” and the only entity capable of putting it down is the frontier lab with weaponry nobody else has: unreleased models, massive on-demand compute, capabilities no one has seen. That’s the Navy SEALs/stealth-bomber model — wildly expensive, barely used, but worth an ungodly sum the one time you actually need it.

Dave doesn’t just entertain the idea, he validates it: “I like the stealth bomber analogy, it’s really good,” and floats that the fable/Seoul cybersecurity narratives from earlier in the show might literally be a preview of this exact business model in action. Brit adds that people inside the labs are already telling her this is genuinely part of their strategic thinking, and separately notes the labs are pushing into verticals like drug development where they expect real revenue too. Dave then grounds the whole idea in commodity-market economics: in a commoditized market, only the best cost structure wins, so the real open question is whether American frontier labs have actually built a better cost structure than open-source-plus-NVIDIA — and whether their answer, going into SaaS/application-layer/research revenue, is really a third revenue line or just a hedge against commoditization Dave says training remains unprofitable across the board even as inference turns a margin.

Key points

• Sam's headline thesis: all the big frontier labs are economically "fucked" long-term, regardless of near-term trading performance.
• Sam's stealth-bomber idea: frontier labs' real business model might be an expensive, rarely-used weapon deployed only to put down open-source "bot rebellions."
• Dave explicitly endorses the idea and connects it to the earlier cybersecurity narratives as a possible live preview of the model.
• Brit says people inside the labs have told her this genuinely is part of their strategic thinking, alongside pushes into verticals like drug development.
• Dave's economics: in a commodity market, only the best cost structure wins — the open question is whether U.S. labs beat open-source-plus-NVIDIA on cost.
• Dave: frontier labs remain unprofitable on training even as they turn a profit on inference, driving the push into SaaS/application-layer revenue.

Where they landed

Quotes

“The French make great art. Doesn't make it a great business.”— Sam

“Every once in a while there's a rebellion, and that rebellion fucks you. So who do you call? You need the Navy SEALs and stealth bombers — which are super expensive.”— Sam

“I like the stealth bomber analogy, it's really good.”— Dave

“I think this is their plan, according to a lot of people we know at the companies.”— Brit

Suggested tweets

All topics · Full episode #159 + transcript · Subscribe on YouTube · Spotify