Bison Ventures analysis questions the billion-dollar AI seed round

Bison Ventures principal Ellie McDonald says roughly 200 publicly disclosed first financings of $100 million or more over the past 15 years produced exits in only 20% of cases, with about 1% generating 10x or better MOIC for first-round backers. The analysis lands as AI companies are raising first rounds once associated with late-stage scaleups, including Yann LeCun’s $1 billion raise, Project Prometheus’ $6.2 billion launch financing and Unconventional AI’s $475 million round two months after its founding.

McDonald’s argument is that the headline seed round has become a poor proxy for the health of venture formation. The numbers she cites suggest that very large entry prices can cap investor upside even when the underlying company is strong, a point that matters for funds underwriting AI infrastructure, model labs and other capital-heavy startups at prices that assume fast category dominance.

Biotech is the comparison case

Bison Ventures, where McDonald is a principal, has experience in biotech, a sector that has long used large first rounds because clinical development costs can be high from the start. McDonald writes that the biology-driven need for capital has produced some strong outcomes, but also a long set of more limited returns for investors who entered at the first financing.

The firm’s dataset covered every publicly available $100 million-plus first round it could identify over a 15-year period, totaling about 200 deals. McDonald says only a small number of the exited companies met the firm’s threshold for a venture-style return, defined as at least 10x MOIC for the first-round investor.

AI may improve that record, according to McDonald, but she frames the likely gains as concentrated. She says eventual exits by OpenAI and Anthropic would roughly double the number of outlier returns in the dataset. Even then, she cites reports that OpenAI’s first investors may see 30x to 40x returns at projected IPO valuations, a strong result but below some earlier venture outcomes.

For comparison, McDonald points to reported returns from prior technology cycles: Sequoia Capital and Kleiner Perkins each turning about $12.5 million in Google into roughly $4 billion, and First Round Capital’s roughly $500,000 Uber investment reportedly becoming $2.5 billion. Her conclusion is that entry price, not just company quality, drove the difference.

AI winners did not all start with mega-rounds

The analysis also pushes back on the idea that AI company formation now requires a giant seed round. McDonald says the number of seed financings above $50 million has risen sharply since 2018, while more conventional first rounds have also continued to grow.

She cites several AI companies now held up as category leaders that began with smaller first rounds: Cursor at less than $10 million, ElevenLabs at $2 million, Legora at $11 million, Sierra at $25 million and Cohere at $5 million. McDonald says each is now valued above $5 billion and producing hundreds of millions in revenue.

That distinction matters for seed investors deciding whether a billion-dollar first check is a new template or an edge case. McDonald’s view is that the largest rounds may be necessary for some companies, particularly at the frontier-model layer, but the disclosed history of mega-seeds does not show that more capital at inception reliably improves venture returns.

The claim is narrower than a bearish view on AI. McDonald says some of the current AI mega-seed companies will produce 10x-plus outcomes. Her warning is about portfolio construction: a strategy built around rare exceptions has had a weak record across the public mega-seed data Bison reviewed.

This story draws on original reporting from Crunchbase News.