US corporations are abandoning the overuse of artificial intelligence (AI) technology as costs rise without a corresponding increase in productivity. This was reported by Qazaqyia.kz citing Associated Press.
The phenomenon of "tokenmaxxing" — maximizing the use of AI-generated work from products like OpenAI's ChatGPT and Anthropic's Claude — emerged as tech industry-fueled springtime hype but has shifted to a summertime backlash.
"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.
Tokens are the building blocks of generative AI, corresponding to small pieces of text. Each token is about three-quarters of a word. There is typically a limit on usage, with pricier AI products offering higher caps.
"As bills started to pile in, people realized that those new tools are quite expensive and you need to use them wisely," Gusdorf said.
Just a few months ago, Silicon Valley executives promoted high token consumption as a sign of high-performing employees. The stereotypical tokenmaxxer stayed up late, orchestrating an army of 24-hour AI agents.
OpenAI CEO Sam Altman said in May he was "excited to see what will happen with tokenmaxxing startups." Nvidia CEO Jensen Huang said "if your $500K engineer isn't burning $250K in tokens, something is wrong." Facebook parent Meta had an internal competition rewarding token usage.
The trend boosted revenue for leading AI model developers like Anthropic and OpenAI, but fizzled as it became apparent it wasn't the best strategy for everyone else.
Microsoft CEO Satya Nadella admitted tokenmaxxing can be addictive but warned customers pay twice: first for tokens, then by feeding proprietary data to AI providers. He raised doubts about data protection assurances of leading AI providers.
Palantir CEO Alex Karp told CNBC something had gone "completely wrong." He said American businesses are "livid" about paying for tokens that create no value.
Bain & Company consultant Jue Wang said many large businesses are reviewing AI investment returns. "The token cost for them has been doubling almost every other month," she said. "$200 per developer per month, multiplied by 20,000 developers, quickly gets to a number no general manager has planned for."
Sometimes it means not using an AI sledgehammer to crack a nut. "Not everything needs a Claude Opus 4.6," Wang said, noting many companies use it even for writing emails.
This has led to a search for "model routing" tools, where simpler queries are automatically sent to cheaper, more efficient AI systems.