In a recent opinion piece for The Information, renowned venture capitalist Vinod Khosla cautions that today’s AI startup valuations are often “bonkers” not because of irrational hype alone, but due to structural pressures driving overvaluation. He notes that many venture capital firms feel compelled to have AI in their portfolios, limited partners are demanding exposure, and capital is flowing more easily than ever.

Khosla argues that this influx of capital is pushing valuations upward, sometimes without sufficient fundamentals. While many businesses will fail, he believes the sector will follow a power-law distribution: a few extraordinary winners could deliver returns large enough to overshadow many losses. He points out that value on paper (TVPI) can be misleading actual cash returned (DPI) often tells a different story.

According to Khosla, the most promising segments are foundation models, computing infrastructure, robotics, and defence; other areas like enterprise search and coding may succeed but possibly not at current inflated expectations. Investors, he suggests, must be highly selective backing the potentially transformative rather than diffuser.