Khosla Ventures' Samir Kaul on leading Jeff Dean's Discovery Loop and why billion-dollar seed rounds are a losing bet

Khosla Ventures' Samir Kaul on Discovery Loop, NeoLab skepticism, and how the firm actually invests

Samir Kaul, a general partner at Khosla Ventures, co-led the seed investment in Discovery Loop, the new company founded by Jeff Dean after 27 years at Google. Kaul says there was no deck. Khosla and one co-lead simply backed Dean on the strength of his track record across Google Brain, TensorFlow, and TPUs, and the implicit signal that someone of his stature wouldn't put his legacy at risk unless the idea was genuinely meaningful.

Discovery Loop's thesis, as Kaul describes it, is applying AI to scientific experimentation: using models to predict the outcome of a given experiment, then running thousands or millions of parallel experiments to converge on answers faster. Target applications include new materials for solar cells, magnets for fusion reactors, and batteries. The analogy Kaul draws is to code generation, where short feedback loops let you verify results quickly. The same tight feedback cycle, he argues, is what makes research acceleration tractable rather than speculative.

“What Jeff and his team are doing is saying, look, we're going to take research, and we're going to figure out if you run this experiment, what do we think the outcome is? And based on that outcome, let's run thousands or millions of other parallel experiments and try to get to an answer. So it could be what's a new material for magnets for a fusion reactor? It could be what are new materials for a solar cell to make it more efficient? ... If a fund is taking 60% of their assets and putting in these large party rounds, these billionaire — I'd be shorting that all day.”

Why Khosla passed on NeoLabs

Khosla has passed on virtually every NeoLab that emerged post-OpenAI and Anthropic. The reason isn't talent, Kaul says. These are legitimately exceptional teams. The problem is that without a clear path to meaningful differentiation from frontier labs, there's no sustainable moat, and the competitive dynamic is brutal: a NeoLab has a breakthrough, and within months a frontier lab recreates it with superior scale and distribution.

Discovery Loop clears that bar, in Kaul's view, because it isn't competing to build a better general-purpose model. It's applying AI to a specific, verifiable problem domain where short-cycle proof points are achievable.

Against billion-dollar seeds

On the strategy of backing large party rounds at billion-dollar-plus seeds, betting on acquihire exits rather than long-run compounding, Kaul is blunt. Recent pseudo-acquisitions like Windsurf and Scale AI returned value to individuals but not to investors. Assuming an acquihire as the base case is a losing bet, structurally, because venture's asymmetry runs the other way: you can only lose 1x, but on something like OpenAI you can make 1,000x. Chasing capital preservation in a business with a 60-70% failure rate defeats the whole point.

Khosla's model, by contrast, is early conviction and large ownership. Rocket Lab is the clearest example Kaul cites: Khosla put in roughly $5 million for approximately a third of the company, owned 28% at IPO, and is watching a business now worth somewhere between $30-40 billion. Commonwealth Fusion is the same pattern. The fund makes concentrated early bets, not pro-rata reflexes.

Pro-rata discipline

On follow-on decisions, Kaul says defaulting to pro-rata is "scandalous." Every partner who walks into a meeting saying "we should do pro-rata because that's our right" gets pushed back. The firm's framework is simple: either pound the table for 3x pro-rata because the company is exceptional, or take a third to a quarter of pro-rata because conviction has softened. The only two legitimate reasons to match exactly are that the company is exceptional and pro-rata is the maximum you can get, or the company is decent and needs support to close the round.

Generalist over specialist

Kaul pushes back on the idea that specialist SaaS investors are structurally disadvantaged in hard tech. Across biotech, aerospace, fusion, and software, the core judgment calls are the same: team quality, market navigation, the ability to pivot. Specialist funds outperform during their boom cycle, he argues, then "suck wind" when the cycle turns. Crypto funds are the recent example. Thoma Bravo and Vista are the SaaS version. Khosla's consistency across 22 years comes from staying generalist, taking early technical risk, and avoiding market-risk bets where product alone has to carry the company.

Founder liquidity

On founders selling at later rounds, Kaul draws the line at roughly $10 million in cumulative secondary sales, evaluated against individual circumstances, whether it's a house, college savings, or releasing pressure so the founder swings for a larger outcome. Beyond $10 million, he says he'd need a serious explanation. On angels selling, he's largely indifferent unless the seller is someone with deep pockets who clearly doesn't need the capital.

The underlying philosophy across all of it is the same: Khosla is optimizing for the 1,000x, not the exit that gets capital back.