AI costs are rising, and OpenAI’s CEO, Sam Altman, has highlighted a sudden shift in corporate attitudes toward AI spending. Altman noted that skyrocketing token costs, which were barely a concern once, have now rapidly emerged as a major worry for businesses aggressively adopting generative AI tools.
Speaking at an enterprise event earlier this week, Altman described the change as abrupt. “It’s kind of a meme now, ‘My company spent my entire 2026 budget in Q1, can you make this more efficient?’” he said.
Just months earlier, Altman said that companies showed little worry about their AI expenditures. “The issue never came up. People were totally happy with the amount they were spending… All of a sudden [AI costs] are a huge issue.”
For those seeking context, tokens serve as the basic unit for measuring AI usage in large language models (LLMs). Most companies typically buy licenses with usage limits and incur additional charges when employees exceed them through heavy interaction with tools like ChatGPT, Claude, or similar systems.
Unprecedented growth in AI usage
At the moment, Altman provided striking figures to illustrate the scale of adoption. Six and a half years ago, OpenAI’s top internal token user consumed about 100,000 tokens per month, which was likely the global leader at the time. Today, though, that volume represents roughly the per capita average worldwide.
OpenAI’s current top internal user now burns through approximately 100 billion tokens a month. Altman, however, mentioned an external user who consumes even more, calling it a personal “embarrassment.”
Recently, companies like Uber reported exhausting their full 2026 AI budgets within the first few months of the year. Reports have surfaced of individual firms spending hundreds of millions on tools like Anthropic’s Claude in a single month to get work done. In response, organisations including Uber, Walmart, Microsoft, Meta, and Amazon have begun imposing caps on employee AI usage, dismantling internal “token leaderboards,” and encouraging more efficient practices.
But why is everyone alarmed at AI usage now?
Altman expressed uncertainty about the precise trigger for the sudden cost sensitivity, despite OpenAI working on efficiency improvements and delivering more value per token. Some of the factors likely contributing include:
– Rapid employee adoption following initial encouragement to experiment with AI. Some firms even mandated using AI tools to accelerate workflows.
– Agentic AI workflows, which can consume significantly more tokens than simple chat interactions (potentially 5–30x or higher in some cases), may have led to cost concerns.
– Scaling infrastructure costs for providers, even as per-token pricing has decreased over time.
With Anthropic confidentially filing for a trillion-dollar IPO this year, the discussion on AI expenditures comes at a critical time for OpenAI, which is also rumoured to file for an IPO later this year.
Altman reassured at the event that OpenAI aims to make AI “great and affordable” so users “never worry about it.” Despite companies grappling with AI usage bills that have outpaced expectations – prompting urgent calls for efficiency – OpenAI hopes to reduce usage costs as more data centres go live.