The venture headlines have been hard to ignore. Yann LeCun raised $1 billion for a company that didn't exist a week earlier. Project Prometheus launched with $6.2 billion. Unconventional AI closed $475 million just two months after founding. Reading the news, you'd be forgiven for thinking the seed round has been permanently reinvented.

But according to Ellie McDonald, a principal at Bison Ventures, the data tells a very different story.

What 15 Years of Mega-Rounds Actually Show

Bison Ventures compiled a dataset of every publicly available $100 million-plus first round over the past 15 years — roughly 200 deals. The findings are striking:

  • Only 20% of those companies had recorded exits
  • Of those, only a handful delivered 10x MOIC or better for the first-round investor
  • That works out to approximately 1% of companies in the dataset generating returns that justify the asset class

"Capital intensity, as it turns out, actually worked against venture outcomes."

McDonald draws on Bison's deep background in biotech — the sector with the longest history of mega first rounds — to frame the argument. In biotech, large first rounds are often scientifically necessary (you can't run a Phase I trial on $3 million), but they've consistently produced a long tail of modest returns for early investors.

The Return Math Is Nuanced, Even for the Winners

The dataset will improve as OpenAI and Anthropic eventually exit. McDonald acknowledges that those two companies alone will roughly double the number of outlier returns in the sample. But even there, the math is instructive.

First-round OpenAI investors are reportedly looking at 30–40x returns at projected IPO valuations — a genuinely strong outcome. But compare that to the prior era's benchmarks:

  • Sequoia Capital and Kleiner Perkins each turned roughly $12.5 million in Google into approximately $4 billion — over 300x
  • First Round Capital reportedly converted a ~$500,000 Uber investment into $2.5 billion — nearly 5,000x

The difference wasn't company quality. It was entry price. Early investors got in at valuations that left meaningful room for compounding.

The Real AI Winners Started Small

Perhaps the most compelling part of McDonald's argument is the empirical record of today's most celebrated AI companies:

  • Cursor — first round under $10 million; now valued above $5 billion
  • ElevenLabs — first round of $2 million; now valued above $5 billion
  • Legora — first round of $11 million
  • Sierra — first round of $25 million
  • Cohere — even at the frontier-model layer, raised just $5 million in its first round

Every one of those companies is now generating hundreds of millions in revenue. The pattern directly challenges the assumption that massive upfront capital is a prerequisite for AI-era success.

What This Means for Founders and Investors

For startup founders, the pressure to raise a splashy seed at eye-watering valuations may actually be working against their investors' interests — and by extension, future fundraising dynamics. High entry prices leave less room for upside to accrue, compressing returns for everyone downstream.

For early-stage investors, the implication is clear: the playbook that generated the industry's greatest outcomes — buying meaningful ownership in capital-efficient companies at prices that leave room to compound — hasn't changed, even if the headlines suggest otherwise.

Mega AI seed rounds are real, and they're growing. The number of $50 million-plus seed rounds has exploded since 2018. But they remain a small fraction of what actually gets funded, and an even smaller fraction of what will return venture-scale capital.

A few of today's billion-dollar seeds will produce genuine outliers. They always do. But building a portfolio strategy around the exceptions rather than the pattern has a 15-year losing track record — and the data, from Google to Uber to Cursor, consistently vindicates the contrarian case.