All-In Summit: In conversation with Vinod Khosla

Khosla said his initial OpenAI investment was twice the size of the largest initial bet he had made in 40 years of venture capital. He decided to back the company in late 2018, accepting uncertainty over timing because he believed practical AI value would arrive well before AGI.

Khosla thinks today’s major AI labs are too concentrated on the same recipe: more Nvidia GPUs, more parameters, and more data. His own bets include symbolic logic and other alternative learning approaches, which he expects are more likely to augment LLMs than replace them.

Commonwealth Fusion’s first investment case was built around a single technical milestone: proving a 20-tesla magnet could be built. Khosla said double-digit millions were enough to test that risk, with nuclear medicine and MRI as fallback markets even if fusion failed, giving the technology unusually valuable off-ramps.

Khosla backed an Australian AI-music startup after roughly half an hour and, by his account, without conventional diligence. The founder’s long-term target was a top-10 song untouched by humans; the product was later released as a “Midjourney for music” trained without YouTube or public music, reducing inherited music-IP constraints.

One host chose Khosla’s 2011 Series B offer even though it carried the lowest valuation among competing bids from Founders Fund and Andreessen Horowitz. The host said Khosla later proved worth the dilution trade-off through board members, introductions, mentorship, and pressure to expand the company’s ambition.

Khosla described a portfolio company that spent years chasing easy revenue instead of building the technology he had urged it to develop. He refused to support its recap unless it abandoned “bad revenue”; only after the crisis did management agree to invest in technology, lower burn, and stop optimizing for an acquisition.

Khosla’s main lesson from the venture boom-and-bust is that valuation cycles can collapse while the underlying technology keeps compounding. He pointed to uninterrupted internet-traffic growth after the dot-com bust and continued railroad construction after Britain’s railway bubble, arguing that long-term value creation matters more than timing sentiment.

Khosla expects AI to make paid work optional within roughly 20–25 years, but he thinks the transition will be politically and socially difficult. He expects some form of UBI to become necessary and says the deeper problem will be human meaning, while capitalism may have to account for inequality as well as efficiency.