AI's biggest valuation risk may come from open-source models

The AI boom has produced extraordinary valuations, but cheaper competition could make them harder to justify.

Chinese open-source models cost a fraction of what OpenAI, Anthropic, and Google charge. According to OpenRouter data, Anthropic's Claude Fable 5 costs roughly $50 per million output tokens, more than 50 times as much as competing models from DeepSeek and Xiaomi.

Even Google's Gemini 3.1, the cheapest proprietary model in the comparison, costs roughly four times as much as Moonshot AI's Kimi K2.6.

AI's valuation problem

That price gap matters for AI companies whose valuations have extraordinary revenue growth priced in.

OpenAI has reached $50 billion in annualized revenue, well below the previously reported $70 billion. Meanwhile, Anthropic reported $16.8 billion in third-quarter revenue, nearly four times its total for 2025.

But strong revenue growth hasn't stopped Anthropic from hemorrhaging money.

The company recorded a staggering $80.7 billion net loss in the first half of 2026. And cheaper alternatives from DeepSeek, Alibaba, and Moonshot give businesses more options to cut AI spending.

For now, restrictions on Chinese models in the U.S. and other Western markets remain the biggest moat of American AI labs.

📌 Bottom line: OpenAI and Anthropic are counting on enormous revenue growth to justify their valuations, but cheaper Chinese models could force them to charge less for the same services.