Nikhil Kamath and Coinbase CEO Brian Armstrong this week warned that valuations of premium AI companies like OpenAI and Anthropic face a structural threat from the rapid rise of open-source models, drawing comparisons to the dot-com bubble and crypto market cycles. The warnings come as a Chinese startup's latest model beat both Anthropic's and OpenAI's top offerings on coding benchmarks, underscoring the competitive pressure from cheaper, open alternatives.
Why open-source models are cheaper
Armstrong noted that open-source models cost up to 99% less for inference and are about six months behind frontier models. That cost gap, he said, poses a direct threat to the high valuations of proprietary AI companies. He expressed nervousness about fast-growing AI valuations, drawing parallels to crypto bubbles where corrections occur before real value emerges. Chamath Palihapitiya echoed the sentiment on X, tweeting that open-source models are the future and highlighting the cost difference between proprietary and open-source models.
The dot-com comparison
Kamath went further, saying shorting every private AI company today could make money in five years. He compared the current AI boom to the dot-com bubble, where many high-flying companies eventually collapsed. A tweet referencing a pattern of six bubbles and five crashes labeled 2026 as the AI bubble, adding to the bearish narrative. Kamath's blunt assessment suggests the market may be pricing in unrealistic expectations for companies like OpenAI and Anthropic.
AI's regional future
Kamath also predicted the AI industry will fragment from American giants to regional self-reliant economies building their own models through reverse-engineering. This aligns with the rise of open-source models that allow countries to develop AI without relying on U.S. tech giants. The trend could further pressure valuations as barriers to entry drop.
Kimi K3 beats the incumbents
Chinese startup Moonshot AI released Kimi K3, which beat Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol on coding tests. The model's performance demonstrates that open-source alternatives are catching up fast, validating Armstrong's point about the six-month lag. For now, the market is watching whether the open-source cost advantage will force a correction in private AI valuations, similar to what Armstrong described in crypto markets.
