20VC: Instinct Raises $1B at $10B Valuation | AMD Buys Fei-Fei Li's World Labs for $8.2B | Meta Poaches MongoDB's CEO | Bessemer Raises $5.75B | Oura Pulls IPO & Nubank Eyes $8–12B Monzo Takeover

Summary

The conversation is a broad tour through the AI funding boom and its knock-on effects on talent, M&A, IPOs, and venture strategy. A major theme is that AI has created extreme dispersion: some companies now require hundreds of millions or billions to compete, while others can still ship with modest capital on top of frontier models. The hosts debate Anthropic’s leaked S-1, consumer agents like Instinct, the economics of inference, and whether public markets may ultimately be more patient than private ones. They also dig into talent migration toward hot AI firms, big exits like AMD’s World Labs acquisition, the difficulty of IPO timing after filing, and why a Nubank-led Monzo deal would reflect market structure more than just deal appetite.

Notes

• Anthropic's leaked S-1 was treated as mostly non-informative, except for customer concentration. Two customers accounted for 25% of revenue, implying one customer spent roughly half a billion dollars last year.
• Jack argued public-market investors may be more long-term oriented than private investors for AI names.
• Jason worried retail investors may react badly to safety disclosures, compute spend, and data-center risk after an AI IPO.
• Instinct’s $1B Series C at a $10B valuation became the episode’s core debate on consumer agents.
• Jack framed consumer agents as a new paradigm: agents that can act across the third-party internet, not just chat.
• Jason’s adoption bar was simple: the winner is the product people use all day long.
• Jack’s key test for Instinct was whether users could autonomously trust it to run their lives.
• Jack said the agent form factor will cannibalize chat time, even if it does not necessarily cannibalize the underlying labs.
• Rory described Benchmark’s Instinct bet as a very large, growth-style check.
• Jack said Benchmark first invested at $2.5B and then again at $10B, but still viewed it as early stage.
• Jack said the current market combines extraordinary valuation dispersion with unusually high upside.
• Rory invoked Tyler Cowen’s view that AI venture returns will be highly skewed and variance will rise, implying many failures.
• Harry said the real worry is not one round, but multiple rounds in quick succession without new data.
• Harry called AMD’s purchase of World Labs for $8.2B in stock the first big Neo Lab exit.
• Rory tied Fei-Fei Li’s ImageNet work to the deep-learning resurgence and framed the deal as deserved recognition.
• Jason said AMD’s 279% stock run and $1T valuation make the World Labs acquisition look cheap as talent insurance versus Nvidia.
• Rory said World Labs-style companies lacked the clear monetization path that made ChatGPT and coding tools obvious bets.
• Jack said there are only about 10 companies that can and want to do $10B acquisitions, which changes exit dynamics.
• Rory said credentialism matters for world-model M&A: technical pedigree helps create the desired acquisition candidate.
• Rory split the AI ecosystem into capital-intensive frontier labs and downstream startups that can ship on much less capital.
• Rory said some markets require hundreds of millions just to enter, while software can still start on $5M.
• Jack said the traditional seed round has become extreme or irrelevant in many market segments.
• Jack said some software companies still compound quietly over 10-20 years despite the market’s focus on fast-action deals.
• Rory said the market is mostly fast-action now after the 2022 discontinuity.
• Jason said the MongoDB CEO departure to Meta showed how much money can move a sitting public-company CEO.
• Jason claimed the departing MongoDB CEO had a $52M package and Meta offered $500M.
• Rory said the market is signaling that the only place to be is in hot AI companies, and people respond to price.
• Jack said talent is being pulled toward AI by money, attention, and the sense of being at the white-hot center of history.
• Jack added that talent liquidity is a positive because great people are moving to the most important opportunities.
• Harry said opportunity cost matters because capital could instead go to firms like Benchmark.
• Harry framed Jev as a $10B raise only a week after a seed around $200M.
• Jason said Jev already has meaningful routing share via OpenRouter and Vercel.
• Rory said the economics can work if the company captures a slice of the roughly $100B spent in the market.
• Jack said it was once rational to buy the labs, and recently the right answer has been to buy inference.
• Jack said many tasks have reached “intelligent saturation,” where enough intelligence is sufficient.
• Jason argued open source has likely peaked because frontier labs can price aggressively and enterprise buyers are wary of China-origin models.
• Jack said enterprises will increasingly post-train their own models on open weights and run them through inference vendors.
• Jack estimated inference clouds consume roughly a gigawatt, with OpenAI and Anthropic each in the high single digits of compute.
• Harry said OpenAI reopened its $200 plan after pausing it for compute shortage, but now halves what $200 buys.
• Jack said the critical unknowns are tokens per task and utility per token.
• Rory said token-cost forecasting is too noisy to trust, so customer spending behavior is the best signal.
• Jason said Sonnet 5.5 used 42% more input tokens and 44% more output tokens while cost fell about 10%.
• Rory said Aura’s pulled IPO surprised him because the company had top-tier banks and a profitable, consumer-known product.
• Rory said heavy secondary makes IPO pricing more sensitive for venture holders who are crystallizing gains.
• Rory compared pulling an IPO after filing to jumping into a bobsled with few exit ramps.
• Jason noted Aura employees had already received $534M from a tender offer, cushioning the blow of the IPO pullback.
• Rory said the IPO window is not obviously closed, citing the S&P near highs and Shiller P/E at an all-time high.
• Jack said AI-native companies have not yet gone public, but several could in 2027 if markets hold.
• Harry said Monzo may be acquired by Nubank for roughly $8B to $12B.
• Rory said it is more surprising that Nubank wants to buy Monzo than that Monzo wants to sell.
• Harry said Monzo wants to sell because it is too small for U.S. public markets and Europe’s public markets are unattractive.
• Rory contrasted U.S. neobank efficiency with weaker legacy-bank economics in Europe and Brazil.
• Harry cited Nubank’s 23% yearly drawdown, $47B market cap, and 12.6% weekly drop.
• Jason said the Monzo deal is effectively a way to buy time and enter the British market immediately.
• Rory said boardroom drama at Monzo made a sale more attractive after CEO and chairman churn.
• Harry said Bessemer raised $5.75B, contrasting a giant platform firm with a smaller seed fund and NFX’s GP-only model.
• Jason said a 2026 seed round could be $30M, making even a $1.7B fund feel small.
• Jack said firms like Bessemer have LP trust and scale, while GP-only investing changes the feel of the game.
• Jason said self-funding is attractive, but writing a $5M check may be too much risk without outside capital.
• Harry introduced Anthropic’s founders locking up 50.1% control.
• Rory said the voting-control mechanics likely need to be recreated post-IPO after common-stock conversion.
• Rory argued founder control is lower priority than existential model, cyber, and state-actor risks.