Martin Shkreli on FTX's billion-dollar Cursor stake, Anthropic's EA mafia returns, and why more AI companies should go public

Summary

Martin Shkreli joins to walk through his project mapping estimated VC returns across the AI investment cycle — using public filings, cap table heuristics, and his network to reconstruct positions that Crunchbase and PitchBook have never tracked.

The headline number is FTX's position in Anysphere, the company behind Cursor. Shkreli estimates the bankruptcy estate holds roughly $1.2 billion in Anysphere stock today, based on what he believes was approximately $300,000 invested at a $4.4 million pre-money valuation — part of a $400,000 round. The bankruptcy lawyers, he notes, saw the name "Anysphere" and assumed it was worth nothing.

Other returns he surfaces from the same exercise: Jaan Tallinn, the Skype co-founder, turned $100 million into roughly $11 billion in Anthropic as an early backer alongside Reid Hoffman. Dustin Moskovitz put in $25 million and is sitting on approximately $4 billion. Thrive Capital remains the murkier case — Shkreli says nobody is quite sure how deep Thrive went into OpenAI, but their Anysphere position alone points to outsized gains. The broader Anthropic syndicate has grown so long, as Dan Primack has noted, that it's easier to list who isn't in it.

The EA network as alpha

The pattern behind several of these returns traces back to the Effective Altruism community. Shkreli's read is straightforward: a dense cluster of unusually smart people who substituted the EA network for a social life ended up in the same cap tables early. That network has since diffused — EA has become, in his framing, more like B2B SaaS than a tight-knit movement — but the early vintages were extraordinarily lucrative.

The IPO argument

Shkreli's most direct policy argument is that more AI companies should go public, and sooner. He sees the private markets as genuinely stretched — Anthropic's investor list now spans virtually every major fund, and he argues the capital requirements for frontier AI labs have outgrown what private markets can absorb. Public markets could realistically support $100 billion, $200 billion, or $300 billion raises in a way private rounds cannot. He uses Replit as a concrete example, arguing the company would likely surprise itself with the valuation it could command. Michael Grimes returning to Morgan Stanley, in his view, is a meaningfully bullish signal for the IPO window reopening.

The practical obstacle he highlights is almost bureaucratic: Replit can't list without first clearing the ticker conflict with Replimune, a drug company that already holds the closest equivalent. Elon Musk faced the same issue with X, waiting for United States Steel to be acquired before the ticker freed up.

Domain expertise as the new founder edge

As code generation gets cheaper, Shkreli argues the relevant moat shifts away from engineering headcount and toward product, sales, brand, and — most importantly — deep domain knowledge. His clearest example is Rogo, a finance-focused AI company built by people who actually did the job on Wall Street. He predicts a wave of founders emerging from incumbent industries — oil and gas, manufacturing, homebuilding — who understand the specific software problems their sectors have never solved, and who can now build the solution without a computer science background.

His ElevenLabs observation fits the same thesis. Shkreli's own company spent its first six to nine months trying to build a competitive voice product and couldn't match ElevenLabs' quality. His retrospective: ElevenLabs won not on technology but on aggressive sales execution, shipping early despite rough edges, and building distribution before the product was ready.

Finance adoption and Wall Street

AI adoption in finance is slow, and Shkreli thinks it will stay that way. The industry's contrarianism cuts both ways — it makes firms sticky customers once converted, but hard to move initially. Developers have embraced AI almost universally; traders and analysts remain more resistant, in part because the mystique around discretionary skill is hard to dislodge. Quant funds are a different story: he says many are quietly pulling new ideas from LLMs and using AI to implement strategies, which creates its own risk for firms whose edge has historically been proprietary.

On layoffs, his view is that productivity gains typically get reinvested into new work rather than headcount cuts — at least until there is genuinely nothing left for a person to do. In practice, he says that point rarely arrives at a growing company.

Neo-labs

Asked about the cluster of research-focused AI spinouts, Shkreli is cautiously bullish, citing the specific technical ambitions around continual learning and novel AGI architectures as genuinely hard problems worth funding. His concern mirrors the standard bear case: researchers starting companies rarely know how to run a business, and even a genuine technical breakthrough may be absorbed by a larger lab within months. The survivors, in his view, will be the ones who treat investor capital as a serious obligation and push through funding crunches rather than fold — he points to Inflection as the cautionary example of the alternative.