Sam Lessin: Google is winning AI, incumbents beat startups, and OnlyFans is a screaming buy

Summary

Sam Lessin, partner at Slow Ventures, makes the case that the AI wars are breaking in favor of incumbents — and that Google is the clearest winner so far.

Google is winning

After spending roughly 20 hours vibe coding across flights this week, Lessin switched from ChatGPT to Gemini around hour eleven. Gemini 2.5 Pro, he says, is meaningfully better at coding tasks — structuring, reasoning, a range of things. He's been arguing Google would win for two years, so he's unapologetic about the victory lap. His one critique of Google IO is that Google is terrible at marketing: they launched so many things that no one knows where to focus, even though the underlying technology is clearly strong.

Incumbents beat startups

Lessin's broader thesis holds that AI is a sustaining innovation, not a disruptive one, which means the companies with distribution, data, and defensible use cases compound their advantages rather than lose them. Meta is his cleanest example: he describes it as heads-you-win, tails-you-win. The most obvious high-value AI use case in the world is making ads more targeted, and Meta owns that surface. Llama's open-source distribution reinforces this — Meta doesn't much care whether it's their model or someone else's running the inference, because the ad dollars flow to them either way.

OpenAI hardware bet

Lessin is skeptical of the Johnny Ive / OpenAI hardware play. He acknowledges OpenAI needs a hardware strategy — being dependent on Apple and Android infrastructure is genuinely dangerous — but he puts the probability of successfully launching a new device category close to zero. The $6 billion figure he reads largely as a PR move, a number chosen to signal power in what he calls the AI pissing match. His view is that new form factors only work when they attach to something already on the body (glasses, watch) or replace something already in the hand (phone). A net-new device is a very hard sell.

OnlyFans is a screaming buy

Lessin says he has wanted to buy OnlyFans for years and still does. His argument: the asset is being mispriced because of two overlapping biases. First, mainstream tech investors won't touch it. Second, the investors who are aware of it assume AI girlfriends will collapse the creator economy and kill the business. Lessin thinks that's simplistic. OnlyFans runs at a roughly 10–15% take rate — well below YouTube's ~45% or most platforms' 30% — which is what makes the creator loyalty and cash generation so durable. He also argues the platform is actually a strong AI opportunity rather than an AI casualty: the high-volume messaging workflows that drive revenue are exactly the kind of human-in-the-loop operation that AI can enhance rather than replace. The buyer pool is small, the stigma is real, and that's precisely why the price is low relative to the earnings multiple of around 8x.

Model selection is a mess

One of Lessin's sharper observations from his week of vibe coding is that choosing which AI model to use for which task is genuinely unsolved. He compares it to walking into a room full of strangers and being asked who to trust, with new people appearing every two seconds. His working solution is to track what the smartest people he knows are using for specific task types — treating model selection as a human trust and identity problem rather than a technology problem.

AI won't eliminate accountability

Lessin pushes back on the idea that AI will straightforwardly eliminate jobs by automating tasks. His argument is that people hire employees not primarily to do work but to have someone accountable when the work goes wrong. Replacing a human with an AI queue that's 80% right still leaves the manager approving outputs — and the whole point of delegation is to avoid that. Until AI can carry genuine accountability, the productivity math is more complicated than it looks.

Seed fund economics

On emerging managers, Lessin is blunt: the business is bad for almost everyone in it. Funds under ~$50M can't generate meaningful economics even with a strong return. Funds over $200M face law-of-large-numbers problems that make generating top-quartile returns structurally difficult. His view is there's a sweet spot — which he believes Slow Ventures occupies — but most emerging managers will not survive. PitchBook data cited in the conversation shows emerging managers are on track to raise less than $20 billion this year, down from roughly $60 billion in 2021.