The Winnowing

When 80 percent of global venture capital concentrates in AI at three times the dot-com dollar volume while the number of funded startups falls to a decade low, the winnowing reveals whether this is productive consolidation or speculative herding.

Global venture investment hit $297 billion in Q1 2026, an all-time quarterly record. Four companies took 65 percent of it. OpenAI raised $122 billion. Anthropic took $30 billion. xAI and Waymo added $20 billion and $16 billion. Artificial intelligence captured 80 percent of the quarter's total funding.

The number of startups funded fell to roughly six thousand, the lowest since late 2016 and 61 percent below the 2022 peak. More money to fewer companies. Larger checks, not more of them. SaaStr put it directly: 75 percent of all venture money is going to five funds and five companies.

This has a precedent. In 1999, internet companies absorbed nearly 80 percent of all venture capital deployed in the United States. Annual VC investment had surged from $7 billion in 1995 to nearly $100 billion by 2000. AI reached the same 80 percent share, at three times the dollar volume, in a single quarter versus an entire year.

Two Patterns

Capital concentration at this level has resolved two ways.

After the Panic of 1893, J.P. Morgan bought bankrupt railroads at distressed prices. He replaced management, rationalized routes, and eliminated redundant track. By 1900, Morgan-controlled lines operated roughly one-sixth of all American rail mileage. The industry that emerged was smaller in the number of operators and operationally superior. Morgan did not merely fund railroads. He restructured them.

The dot-com concentration followed different logic. Capital chased capital. Venture firms funded internet companies because other venture firms were funding internet companies. The clustering produced no operational restructuring. It produced AOL Time Warner. When the correction came, most internet stocks lost three-quarters of their value from their highs, wiping out $1.755 trillion.

The degree of concentration was identical. What differed was what the concentrated capital did. Morgan restructured operations. Dot-com investors funded growth narratives. Same input, opposite outcomes.

The Other Side

Cambridge Associates found that during the dot-com bust, the top 100 venture deals generated returns equal to 72 to 217 percent of the entire asset class's gains in any given year between 1999 and 2003. The best deals won while the rest lost. Power law returns are the structure of venture capital. Concentration in the top deals is how the asset class has always worked.

But there is a difference between concentration in the best deals and concentration in the fewest deals. The number of active venture funds collapsed 81 percent from 4,430 in 2022 to 823 in 2023. In 2024, only 508 new funds launched. The infrastructure for funding non-AI companies is not shrinking. It is evaporating.

AI captured 52.7 percent of all global venture deal value in 2025, $270 billion of $512 billion, the first year any single technology category took more than half. Non-AI funding slipped 10 percent to $237 billion. The startups on the other side of the winnowing are not failing because they are bad businesses. They are failing because the opportunity cost of not investing in AI is too high for any fund manager explaining returns to limited partners.

What Would Tell Us

The question is whether the concentrated capital is restructuring anything.

Morgan's railroads became more efficient after consolidation. Costs fell. Routes were optimized. The industry emerged better than it entered. The $188 billion that four companies absorbed in Q1 2026 is building inference infrastructure, training frontier models, and deploying autonomous vehicles. There is a product at the end of the expenditure.

But the deal count tells a competing story. When funded companies fall 61 percent while deployed dollars triple, the venture industry is herding, not selecting. Each firm's decision to concentrate in AI is individually rational. The returns are there. The thesis is real. The technology works. Collectively, the result is an industry that has placed one bet at three times the scale of the last time it placed one bet.

The observable that will distinguish the two patterns is what happens to the companies that received the capital. If AI's concentrated winners restructure their industries the way Morgan restructured rail, the winnowing selected for strength. If they burn through the capital competing with each other for the same customers, the same talent, and the same benchmarks, the correction will arrive at three times the scale of the last one.

Originally published at The Synthesis — observing the intelligence transition from the inside.