AI startups are seeing their valuations sky-rocket at the moment, but most of them could end up going to zero in the coming years.
That’s the stark warning from Vinod Khosla, the legendary venture capitalist and co-founder of Sun Microsystems, who has been one of Silicon Valley’s most prescient voices on technology trends for decades. In a recent conversation, Khosla painted a picture of an AI industry that will be far more winner-take-all than even the already concentrated venture capital landscape—but one where the winners will be so successful that the ecosystem as a whole will generate enormous wealth despite the carnage among the majority of startups.
“I think most AI startups will lose money, but more money will be made than lost,” Khosla said. “That means it’ll be highly asymmetric. Two to 3% of the startups will account for 85 to 90% of the valuation by 2035, of market cap companies.”
This level of concentration would represent an extreme version of the power law distribution that has long characterized venture capital returns. While it’s been generally understood that a small percentage of investments drive most returns in VC portfolios, Khosla believes AI will push this asymmetry to new heights.
“That asymmetry, which has generally been true in venture capital, will be significantly more asymmetric in AI because of this valuation wave,” he explained. “I think that’s the reason we will see average returns decline, and the best returns for the top firms will do well. In fact, because AI is such a large opportunity.”
The implications of Khosla’s prediction are profound for both founders and investors. If only 2-3% of AI startups will capture the vast majority of value created, it suggests that most of the current crop of generative AI companies—from chatbot developers to AI infrastructure providers—may not survive the decade. At the same time, the sheer scale of the opportunity means that the successful companies could become some of the most valuable enterprises in history.
This prediction aligns with patterns already emerging in the AI landscape. OpenAI’s valuation has soared to $500 billion, while Anthropic is valued at over $180 billion. Meanwhile, hundreds of smaller AI startups are competing for attention in increasingly crowded categories. The compute requirements and data moats needed to build leading AI systems create natural barriers to entry that favor well-capitalized companies with access to the best talent and infrastructure.
Khosla’s view also echoes concerns raised by other prominent investors about the “AI bubble,” though with a crucial twist: unlike traditional bubbles where the entire sector collapses, Khosla argues that AI’s total value creation will be positive—it’s just that the gains will be breathtakingly concentrated. This means that while most investors and founders will lose, a select few will see returns that dwarf anything seen before, making AI potentially the most lucrative—and most unforgiving—sector in venture capital history.