How Cheap AI Could Derail OpenAI And Anthropic's IPOs

TL;DR

OpenAI and Anthropic's trillion-dollar IPO valuations are threatened by cheap Chinese open-source AI eroding their pricing power before they go public.

Key Points

• 1.OpenAI and Anthropic are pursuing IPOs at valuations above $800 billion. Their pitch is that they are the next Microsoft or Google, with durable pricing power lasting decades — but that moat is already eroding.
• 2.Chinese open-source models have surged from 1% to over 40% of AI traffic in one year. On Open Router, the largest AI traffic aggregator, three of the top five models are now Chinese, with labs like Deepseek, Moonshot, and ZHIPU shipping competitive models in just four months.
• 3.Anthropic's Claude costs nine times more than the cheapest Chinese alternative for the same work. A $10 million AI budget on Claude Opus burns out in weeks; the same budget on Deepseek stretches across most of a year.
• 4.Chinese labs turned export-chip restrictions into an algorithmic advantage. Cut off from Nvidia's best chips, they optimized for smaller, cheaper, more efficient models — producing what Aidan Gomez called 'extremely smart algorithms' that are not an inconsequential advance.
• 5.The one stronghold for premium pricing is high-trust regulated industries. Banks, grid operators, defense, and healthcare won't use Chinese models regardless of cost, creating a niche where Western labs can still charge a premium for 'democratically aligned technology.'
• 6.American competitors are targeting that exact regulated-industry gap. Cohere (run by Attention paper co-author Aidan Gomez) reported 6x revenue growth last year building secure, on-prem models deployable on just 2–4 GPUs; Nvidia's open-source Nemo Tron models are also being adopted by Palantir, Salesforce, ServiceNow, and CrowdStrike.
• 7.Even Elon Musk hedged by merging xAI into SpaceX rather than keeping a standalone AI lab. xAI was burning roughly $1 billion per month before the merger; backstopping it with SpaceX's revenue signals that pure-play frontier AI economics are difficult — an option OpenAI and Anthropic don't have.
• 8.Aidan Gomez warns the industry has an image problem alongside its business problem. Public skepticism around AI slop, energy costs, and artist compensation is growing, and he argues spokespeople need more empathy rather than steamrolling critics — while still predicting insatiable enterprise demand for AI.