Cheap AI Could Threaten the IPOs of OpenAI and Anthropic

As Chinese AI labs deliver performance comparable to US frontier models at far lower prices, the premise behind OpenAI and Anthropic's expected IPO valuations of over $800 billion is being shaken. OpenAI is expected to file a confidential listing application as early as this week.

The price gap is stark. Cost comparison for the same task from AI benchmarking firm Artificial Analysis:

• Anthropic Claude: $4,811
• OpenAI ChatGPT: $3,357
• DeepSeek: $1,071
• Kimi: $948
• Zhipu GLM: $544

Claude is nearly 9 times more expensive than the cheapest Chinese alternative. With 45% of enterprises spending over $100,000 per month on AI in 2025, this price difference matters a great deal.

Enterprise adoption is shifting rapidly. On the AI model marketplace OpenRouter, the share of Chinese model usage surged from about 1% in 2024 to over 60% in May 2025. Enterprises use an "advisor model" strategy: using cheap open-source models by default, and calling on frontier models like OpenAI and Anthropic only for tasks the cheaper models can't solve.

Databricks CEO Ali Ghodsi said, "You can really cut costs well this way." Google CEO Sundar Pichai also suggested cheaper Flash models at the I/O conference, noting that "many companies have already burned through their annual token budget and it's only May." He said that if the biggest customers moved 80% of their workloads to Gemini 3.5 Flash, they could save over $1 billion annually.

The performance gap is also disappearing. DeepSeek unveiled a preview of its next-generation model last month, showing performance on par with or close to the latest models from OpenAI, Anthropic, and Google on coding, agentic, and knowledge benchmarks. Other Chinese labs including Moonshot, Xiaomi, and Zhipu have also released comparable-level models over the past 4 months.

The US response is also taking shape. Nvidia has released its own AI system for free, allowing companies to run it on their own servers. Reflection AI raised funding at a valuation of several billion dollars with the goal of building US-made open-source models for companies that don't want Chinese alternatives. Cohere CEO Aidan Gomez said regulated industries like banking and defense won't use Chinese models regardless of price, and that the company's revenue grew 6x last year.

In a May policy report, Anthropic warned that US models are "only months ahead" of Chinese models, and that Beijing is "winning the race for global adoption" on cost. An OpenAI insider countered that enterprise demand surges like a "vertical wall" every time a new frontier model like GPT-5.5 launches, arguing that open source isn't eating into the core business. However, one enterprise AI CEO assessed that "the growth is real, but if this (advisor model) technique hadn't been used, frontier models would have scaled much faster."