There are ~25 companies with a ~$1B+ headline valuation in their first round of raised capital (TML, SSI, Ineffable, Ricursive, World Labs etc). Here are 11 reasons this happens, and the case for it: Outcomes are larger than ever before. Anthropic and OpenAI are ~$1T outcomes that are mostly liquid in private markets. Many of these startups are “neolabs”, big bold expensive research ideas pre-revenue and pre-product. If every single “neolab” raises at a $1B post at seed, if 1 out of a 100 “hit”, that could be a 10x return in 5-7yrs, an implied 40-60% IRR net of dilution.
Compute procurement. Today, capital is not much of a constraint as much as access to compute (GPUs) for many of these companies. This means getting, say, $100M, worth of GPUs can be a hard requirement. If you need $100M of GPUs and want low dilution (10%), you might easily back into a $1B valuation.
Talent procurement. Talent is also more of a constraint than capital. AI researchers are expensive, both on cash and equity. You need to raise to pay enough cash. High valuations also correlate with a high strike price / 409a, even though it might be 10-20% of the preferred price. If you’re valued at $1B and giving 0.25% to a researcher, they might still have to pay $250-500k to buy $2.5M of options / paper money. They might already have that money lying around but often could use cash to finance that and demand a commensurately higher base salary.
Limited competition increases success rate. p(success) = p(success | this idea works) x p(this idea works). It’s hard to measure p(this idea works). But, because AI is compute, talent and capital constrained, p(success | this thing works) is much higher because you might be competing with <5 other plays vs 1000s. If the idea works, you'll capture the value.
Preference stack is a perceived downside cushion. If the value of the talent and IP is significant (remember, Meta pays 9 figure packages for top talent), there’s a belief that this is easily acquired for at least more money than you put in. If you invest $100M at $1B, you might believe there’s no way the team doesn’t get bought for >$100M, as evidenced by many such acquisitions (Cursor, OpenClaw, Windsurf, Vercept, Astral, Bun, Coefficient Bio – not all are neolabs) so investors get their money back (>1x). That might lead VCs to believe the downside case isn't too bad.
Institutional investors often don’t pay the sticker price. Rounds are done in multiple closes to reduce the blended cost basis for the investor. Institutional players might get their ownership in the first unannounced round and put a token amount in the next, with strategics like NVIDIA piling on in the final round. Multiple rounds counted together is called one “seed”.