The artificial intelligence spending spree that dominated Silicon Valley just months ago is collapsing under the weight of runaway costs and disappointing returns. What started as a celebrated strategy of maximizing AI usage has shifted into a corporate backlash, with enterprises discovering that burning through tokens,the building blocks of generative AI systems,doesn't automatically translate into business value.

OpenAI CEO Sam Altman was among the most vocal champions of this trend. In May, Altman declared he was "excited to see what will happen with tokenmaxxing startups, both for how they work internally and the products they can build." Nvidia CEO Jensen Huang went further, suggesting that if a $500,000 engineer wasn't spending $250,000 in tokens monthly, "something is wrong." Meta even ran internal competitions rewarding high token consumption.

In May, Altman

But as bills arrived in the summer of 2026, the reality became impossible to ignore. Token costs for large enterprises are doubling almost every other month, according to management consultants advising Fortune 500 companies. For a company with 20,000 developers each spending $200 monthly on AI tokens, that's a $4.8 million monthly bill that most general managers never anticipated.

What Exactly Is "Tokenmaxxing" and Why Did It Fail?

Tokens are the fundamental units that AI systems process. Each token represents roughly three-quarters of a word, and AI providers typically charge based on how many tokens users consume. The strategy of "tokenmaxxing" involved using AI for virtually every task, regardless of whether it was the right tool for the job.

The appeal was seductive: executives could claim their companies were AI-forward, and employees could position themselves as productivity champions by orchestrating armies of AI agents working around the clock. The problem emerged when companies realized they were paying premium prices for AI models to handle tasks that cheaper, simpler tools could accomplish just as well.

"It's very easy to create something you don't need with AI," said Vincent Gusdorf, head of AI analytics at Moody's Ratings and author of a new report recommending a more disciplined approach.

Vincent Gusdorf, Head of AI Analytics at Moody's Ratings

The backlash has been swift. Palantir CEO Alex Karp told CNBC that something had gone "completely wrong" with the tokenmaxxing approach. He described American businesses as privately "livid" about paying enormous sums for tokens that create no measurable value.

How Are Companies Rethinking Their AI Strategy?

Rather than abandoning AI entirely, enterprises are adopting a more surgical approach focused on return on investment. The shift involves three key strategies:

• Model Routing: Automatically directing simple queries to cheaper, efficient AI systems while reserving expensive, powerful models like Anthropic's Claude Opus 4.6 for genuinely complex tasks such as software engineering or deep research.
• Task-Specific Evaluation: Asking whether each use case actually requires a cutting-edge large language model (LLM), or whether a simpler tool would suffice, such as generating routine emails.
• Open-Source Alternatives: Exploring less expensive open-source AI models from Chinese startups like Moonshot's Kimi or Zhipu's GLM, which offer comparable capabilities to leading U.S. models at a fraction of the price.

Bain & Company consultant Jue Wang explained the shift: "The token cost for them has been doubling, almost every other month. Let's say $200 per developer per month. Multiply that by 20,000 developers, which is often what we're dealing with at these companies, and that quickly gets you to a number that is not a line item that any general manager has planned for".

Microsoft CEO Satya Nadella has also raised concerns about the tokenmaxxing model, warning that customers are "paying twice for AI." They pay once in token costs and again by feeding proprietary company data to external AI providers, creating potential data security risks.

What Does This Mean for AI Companies Like OpenAI?

The pullback in tokenmaxxing spending could significantly impact revenue for leading AI providers. OpenAI and Anthropic built their business models partly on the assumption that token consumption would continue climbing. The shift toward more disciplined, cost-conscious usage threatens that growth trajectory.

Meanwhile, Altman is heading to Washington for meetings with top economic officials, including Treasury Secretary Scott Bessent, where he is expected to preview new OpenAI models while facing questions about recent security breaches and intensifying competition with China.

The broader lesson mirrors a pattern from software development history. Raffi Krikorian, chief technology officer at Mozilla, drew a parallel to an earlier era when the number of lines of code a programmer wrote was considered a productivity metric. That approach eventually fell out of favor as companies realized that more code didn't necessarily mean better software.

"I think tokenmaxxing is moving through the exact same pattern. I think this is going to be an interesting blip that we're all going to look back on to laugh at in a year," said Krikorian.

Raffi Krikorian, Chief Technology Officer at Mozilla

The tokenmaxxing moment reveals a fundamental truth about AI adoption: hype and discipline rarely coexist. As enterprises mature in their AI usage, they're learning that the most valuable applications aren't always the ones that consume the most tokens, but rather the ones that deliver measurable business results at a sustainable cost.