Mistral's €3 Billion Round and AI Sovereignty

Picture a circle of children playing duck, duck, goose. Every so often a chair gets pulled from the ring, and by the end there are only one or two chairs left for a crowd of kids still running. That is the image SaaStr founder Jason Lemkin reached for on this week's 20VC, hosted by Harry Stebbings with Scale Venture Partners' Rory O'Driscoll, to describe what is happening to the software companies that raised money at giant valuations in 2021. "I felt like there's only like one or two chairs left from the pre-AI era," he said, and by the time the discussion turned to whiteboard app Miro's sale, he was ready to sit down. "I've ended the game for me of duck, duck, goose. I'm not running around the chairs anymore" (46:21).

The Chairs Run Out

Miro was one of the hailed names of the 2021 boom, a collaborative whiteboard tool for remote teams that raised at a $17.5 billion valuation (40:02). This week it sold to the Italian acquisition machine Bending Spoons for $1.35 billion (40:02), even though Miro is still growing at high single digits, still cash flow positive, and still pulling in $600 million a year in recurring revenue (46:44). O'Driscoll had watched this coming for a while: his firm tracks every unicorn by valuation and headcount, and Miro's 2021 price tag had simply gone stale while everyone around it kept raising new rounds at higher numbers. A tweet from Sequoia's Andrew Reed captured the mood, he said: a little cartoon of death knocking on doors, Evernote, then Airtable, then Miro, except instead of a scythe it carried a bending spoon. "Death comes for us all in SaaS land, right? And it was exactly right. It was just an inevitable cleanup operation because it was just so far wrong in terms of pricing" (41:41), O'Driscoll said. Early investors and founders still made money on the deal. Later-stage backers mostly did not, though O'Driscoll pointed out that late-stage venture math can still work if every loser returns at least its money back through liquidation preference: "If your losers give you a 1X, then you'll die rich" (42:03).

The same repricing logic sits underneath a chunk of the episode. Rory O'Driscoll cited a framework from investor Gokul Rajaram: above 30% annual growth, a company gets valued on its revenue, because fast growth forgives almost anything. Fall below that line and the market switches to valuing the business on EBITDA, a measure of operating profit, instead. "Above 30% growth you can use a revenue multiple. Below 30% growth you have to use an EBITDA multiple" (70:02). The hosts applied that lens to Adobe, whose new AI usage metrics rose even as overall new recurring revenue fell, and to design tool Canva, whose growth has slowed from roughly 30% to 20% (69:26). Both are healthy businesses. Neither is exciting anymore, and the transition between those two worlds, O'Driscoll noted, can take three or four years of the stock going nowhere while cash flow slowly catches up to the new multiple.

The Killer App Nobody Can Name

Not every corner of the show was about companies running out of road. Meta shipped an AI assistant called Muse the week of recording, built by a reported 500 engineers pulled onto the project after the open-source coding assistant OpenClaw took off (21:51). Muse gives every user a free virtual machine with two CPUs, two GPUs, 8 gigabytes of RAM, and 100 gigabytes of storage (23:27), running on Meta's own large language model and its enormous existing compute footprint. Lemkin estimated that delivering that kind of infrastructure costs rival AI agent products roughly $3 to $4 per user in compute alone (23:48), a cost Meta can absorb in a way almost nobody else can.

The obvious competitor is a startup called Instinct, which is reportedly raising $1 billion at a $10 billion valuation despite having close to no revenue (28:37), just months after raising at a fraction of that number. Lemkin walked through the decision as if he were presenting it to his own investment committee, and concluded he personally would not do the round: Meta has the servers, the model, and the balance sheet, and "this is the threat that every VC worried about" once a frontier lab actually decides to build a consumer app instead of just an API (31:34). His case for Instinct anyway rested on one gap: Meta's products do not talk to WhatsApp rivals, other carriers, or outside services, while Instinct can work across all of them (29:20), and one Instagram reel about Instinct pulled over 1,000 direct-message requests for invite codes (38:08), a level of organic pull he called genuinely rare. Both hosts kept circling back to a harder question neither could answer: what is the one thing an AI assistant does that nothing else can, the equivalent of the spreadsheet program VisiCalc that made the personal computer worth buying. "What's the killer app? Another reservation at a Cheesecake Factory or TGI Fridays?" Lemkin asked (25:04). Nobody on the call had one.

That same appetite for risk showed up in the debate over Anthropic chief executive Dario Amodei's call to "pace the frontier," a proposal for third-party monitoring of AI labs that both hosts found mostly unconvincing. Amodei's own list of concerns was narrow: cyberattacks, economic disruption, and losing control of the systems being built (04:35). It was an unnamed Anthropic employee's claim of a 10% chance of human extinction within ten years that set off the week's news cycle (04:56), and O'Driscoll was blunt about what he thought that implied about how seriously anyone actually believes it.

"If the feds really thought that there was someone in downtown San Francisco building a technology that was gonna blow up the fucking world, had a 10% chance of blowing up the world in the next 10 years, they would move in with a SWAT team, kill everyone in the place, and close it down." (Rory O'Driscoll, [05:12])

Europe's answer to all of this arrived in the form of Mistral, the Paris-based AI lab that raised €3 billion this week, Europe's largest tech funding round on record, led in part by Samsung (61:02). O'Driscoll argued the round has little to do with Mistral actually competing head-to-head with OpenAI or Anthropic on frontier capability. It is about sovereignty: after the United States restricted access to some American models abroad, European governments and companies concluded they cannot depend entirely on American labs for a technology this important, the same logic that built Airbus as a European counterweight to Boeing decades ago (62:52). It took Airbus thirty years to become a serious rival. Mistral, O'Driscoll suggested, is unlikely to match OpenAI's scale anytime soon, but it does not need to. It just needs to exist.

Somewhere in the same hour, the hosts also noted that Sam Bankman-Fried, the former FTX chief executive serving a 30-year sentence for fraud, may get a Supreme Court hearing on an Eighth Amendment argument that his fine was unconstitutional given that customers were ultimately repaid through the bankruptcy process (47:44). And WordPress creator Matt Mullenweg won back control of Automattic's board days after being ousted, a fight both hosts described as a battle over an asset whose relevance is fading regardless of who runs it (52:21). Different stories, same undercurrent: money that arrived on one set of assumptions is now being repriced against a very different set of facts, and almost nobody in the conversation was confident about which chairs are left standing when the music stops.