The spread of mass-market AI models and companies' moves to cut costs could become a stumbling block for OpenAI and Anthropic as they pursue IPOs at high valuations, a report said.

As of May 20 local time, both OpenAI and Anthropic are valued at more than $800 billion. Observers say the growing importance of cost efficiency, alongside performance, in the enterprise AI market could be a negative factor.

According to the report, this earnings season began to show AI cost burdens affecting corporate results. Meta, Shopify, Spotify and Pinterest said rising AI and inference costs were pressuring margins.

A CloudZero survey found the share of companies spending at least $100,000 a month on AI rose to 45% in 2025, from 20% in the same period a year earlier. An Artificial Analysis assessment put the cost for top-performing models at $4,811 for Anthropic Claude and $3,357 for OpenAI ChatGPT. By comparison, DeepSeek cost $1,071, Kimi $948 and Zhipu GLM $544. For the same task, Claude was almost 9 times more expensive than the cheapest Chinese model.

Low-cost models are also rapidly catching up to frontier models on performance. DeepSeek's next-generation model V4 Preview, released last month, showed performance similar to or close to the latest models from OpenAI, Anthropic and Google on coding, agent and knowledge benchmarks. Over the past 4 months, Moonshot, Xiaomi and Zhipu have also released models at a similar level.

Companies are also changing how they use AI. Databricks CEO Ali Ghodsi (알리 고드시) said the "advisor model" approach is spreading, in which firms use low-cost open-source models as a default and call OpenAI or Anthropic frontier models only for tasks they cannot solve.

On OpenRouter, the share of Chinese models rose from about 1% in 2024 to more than 60% in May this year. Figma is selling a feature that cuts customers' token usage by 20 to 30%.