Mega AI seed rounds aren't venture gold—data shows 1% of $100M+ first rounds hit 10x returns

Venture headlines over the past months have celebrated a string of record-breaking seed rounds: Yann LeCun raising $1 billion（約1600億円） for a company that did not exist a week earlier, Project Prometheus launching with $6.2 billion（約9900億円）, and Unconventional AI closing $475 million（約760億円） two months after founding. These announcements have invited a narrative that the venture model has fundamentally shifted and that AI represents a once-in-a-generation capital opportunity requiring unprecedented first checks.

Ellie McDonald, a principal at Bison Ventures, examined this claim against 15 years of data on publicly disclosed $100 million（約160億円）-plus first rounds—roughly 200 deals in total. The findings contradict the mega-round narrative. Of those ~200 deals, only 20% had recorded exits. Of that 20%, only a few delivered what the analysis calls a "venture-like return"—10x MOIC or better for the first-round investor. The bottom line: approximately 1% of companies that publicly raised $100 million（約160億円） or more in their first financing round generated returns that justify the venture asset class. Capital intensity, the data suggests, actually worked against outcomes.

This pattern is not new. Biotech, the venture sector with the longest history of mega first rounds, exhibits the same distribution: large initial checks often lead to a "very long tail of modest" outcomes. Yet biotech mega-seeds are understandable because the underlying science—running a Phase 1 trial—cannot be done on $3 million（約4.8億円）. AI mega-rounds, by contrast, do not face the same structural capital requirement.

Where AI mega-seed outcomes do exist, the math is nuanced. OpenAI and Anthropic will substantially improve the historical dataset when they exit. At OpenAI's projected IPO valuations, first-round investors are looking at 30–40x returns—"a fantastic outcome," McDonald notes, but "a fraction of what early institutional investors made on the generational outcomes of prior eras." Sequoia Capital and Kleiner Perkins each turned roughly $12.5 million（約20億円） of their Google checks into around $4 billion（約6400億円）, or north of 300x. First Round Capital reportedly converted a roughly $500,000 Uber investment into $2.5 billion（約4000億円）—nearly 5,000x. The difference was not company quality but entry price: those investors entered at valuations that left exponentially more room for upside to compound.

Meanwhile, the companies now held up as AI winners started small. Cursor's first round was less than $10 million（約16億円）. ElevenLabs raised $2 million（約3.2億円）. Legora closed $11 million（約18億円）. Sierra took $25 million（約40億円）. Even at the frontier-model layer, Cohere's first round was $5 million（約8億円）. Today, all of these companies are valued north of $5 billion（約8000億円） and generating hundreds of millions in revenue. Cursor, at under $10 million（約16億円）, is "the more representative data point. Project Prometheus at $6.2 billion（約9900億円） is the exception." The number of $50 million（約80億円）-plus seed rounds has exploded since 2018, but traditionally sized first rounds are also growing—and the headline-grabbing mega-rounds remain "a small fraction of what's actually getting funded, and an even smaller fraction of what will return venture-scale capital."

The article challenges a widespread perception that the venture model has fundamentally changed in the AI era. While mega-seed rounds—including Yann LeCun's $1 billion（約1600億円） company, Project Prometheus at $6.2 billion（約9900億円）, and Unconventional AI at $475 million（約760億円）—dominate venture headlines, an empirical review of 15 years of $100 million（約160億円）-plus first rounds reveals a sobering pattern: only 20% have recorded exits, and of those, only a handful achieved venture-scale returns. This mirrors dynamics long observed in biotech, where capital-intensive science necessitates large first checks but does not guarantee strong outcomes for early investors.

The core insight is that entry price, not company quality or capital size, determines return potential. Sequoia Capital and Kleiner Perkins turned roughly $12.5 million（約20億円） into $4 billion（約6400億円） on Google (300x+), and First Round Capital converted ~$500,000 into $2.5 billion（約4000億円） on Uber (nearly 5,000x)—returns dwarfing the 30–40x first-round investors project from OpenAI at its estimated IPO valuation. In contrast, today's canonical AI winners—Cursor, ElevenLabs, Cohere, Sierra—launched with seed rounds under $25 million（約40億円） and now command valuations exceeding $5 billion（約8000億円） while generating hundreds of millions in revenue. The pattern is consistent: capital efficiency at entry, not mega-round size, correlates with venture-scale compounding.