Thought Leadership
OpenAI CEO Sam Altman has acknowledged that soaring AI token costs are now a recurring concern among enterprise customers, while pledging that the company is working to make its models more efficient.
Speaking at the “Intelligence at Work” event, Altman said this was the first time customers had begun actively raising the issue of escalating AI spending. According to him, OpenAI is now focused on improving model efficiency to deliver more value per dollar spent, although he did not provide specific details on how this will be achieved.
“People are really saying it — it’s almost become a meme: My company has already used up its entire 2026 budget in the first quarter. Can you make this more efficient? We’re working on it. I think we’re going to find many ways to give people more value for less spending.”
Sam Altman, CEO of OpenAI
The Cost of “Too Much of a Good Thing”
The hype around so-called tokenmaxxing has played a role in driving costs higher. The idea behind it is simple: heavy use of AI models boosts productivity and justifies increased spending.
Jensen Huang, CEO of Nvidia, pushed this thinking to an extreme, suggesting developers should consume tokens worth at least half their annual salary. In his view, anything less signals underutilization of AI capabilities.
A striking example came from the startup OpenClaw, which reportedly processed 603 billion tokens in a single month — resulting in a $1.3 million bill (USD).
Growing Signs of Disillusionment
Early enthusiasm is now being tempered by real-world cost concerns. At Amazon, internal reports suggest AI agents were sometimes used less for productivity gains and more to climb internal AI usage rankings.
Meanwhile, Microsoft has reportedly scaled back its licensing of Claude Code due to cost pressures. A structural issue is also emerging: agent-based AI systems that autonomously complete multi-step tasks consume significantly more tokens than simple query-response models.
Altman Remains Bullish on Long-Term Growth
Despite the concerns, Altman remains confident in the long-term trajectory of AI adoption. He points to the Jevons paradox: as tokens become cheaper, overall demand rises disproportionately.
He illustrated this with internal OpenAI usage data. Six and a half years ago, the most active employee used around 100,000 tokens per month. Today, the global per-capita average inside the company sits at that level.
The current top user reportedly consumes 100 billion tokens per month, with Altman noting there is likely someone internally exceeding even that figure. If the trend continues, he suggests, the global average could eventually reach similar levels.
