For some, playtime is over.<\/p>\n
The tech world is having a big debate about whether tokenmaxxing, the idea of using lots of AI tokens to boost productivity<\/a>, has gotten out of control. It has also got executives wondering when the return on investment will start to show.<\/p>\n Last week, Amazon closed an internal dashboard<\/a> that tracked AI use after some staff performed tasks just to climb the leaderboard, the Financial Times reported.<\/p>\n “Please don’t use AI just for the sake of using AI,” Dave Treadwell, an Amazon senior vice president, told staff. “Use AI to help you solve customer problems, to help you solve business problems, to innovate.”<\/p>\n An Amazon spokesperson told Business Insider that employees set up the unofficial dashboard a few weeks ago to “drive awareness” of how AI can accelerate work and that it “was never intended to promote the use of AI for usage’s sake.”<\/p>\n Amazon isn’t alone in recalibrating on AI use. Uber’s COO, Andrew Macdonald<\/a>, said in an interview released in late May that he has yet to see improvements directly linked to increased AI spending<\/a>.<\/p>\n Some AI skeptics see these as signs of a bubble as OpenAI and Anthropic look to go public this year.<\/p>\n “If enough other companies report the same, the bubble pops,” Gary Marcus, an AI researcher and professor emeritus at New York University, wrote on X last week about Uber’s experience.<\/p>\n Tech giants want to reduce their AI bills<\/p>\n For execs worried about AI costs, the tech industry has mixed signs about how it might play out.<\/p>\n This week, Microsoft’s GitHub Copilot, an AI-powered coding assistant, is moving from a fixed monthly payment to usage-based billing. In an April blog post, GitHub said it had so far absorbed much of the rising cost of its technology, and the fixed pricing model is “no longer sustainable.”<\/p>\n It joins others, including Anthropic and OpenAI<\/a>, in moving away from flat-rate seats to usage-based billing for business customers.<\/p>\n While price changes have irked some developers, investors say it’s a necessary move and a sign of the market maturing.<\/p>\n “For the past year, a lot of AI products were effectively subsidizing usage in the race for growth. Now the bill is showing up,” Barry Downes, the managing partner at investment firm Sure Valley Ventures, told Business Insider.<\/p>\n “That’s a healthy transition, not a warning sign,” he added.<\/p>\n Uber COO Andrew Macdonald’s comments sparked a debate over tokenmaxxing.\u00a0<\/p>\n Bloomberg\/Getty Images<\/p>\n On the other side of the coin, costs could come down over time as AI companies develop more efficient models and use them to gain a competitive edge.<\/p>\n Google, for example, says its latest Gemini 3.5 Flash model rivals frontier offerings at a lower price, as does Anthropic’s latest Opus 4.8 model<\/a>. As Business Insider’s Hugh Langley reports, Google has a strong position<\/a> to reduce costs because it owns the full stack \u2014 chips, data centers, cloud, models, and many big applications.<\/p>\n After years of AI labs competing on intelligence, they’re now competing on intelligence per dollar, with OpenAI, Anthropic, and others offering smaller, more efficient models.<\/p>\n Tokenmaxxing faces a ‘necessary reality check’<\/p>\n Oded Tahori, the founder and CEO of Jeen.ai, has been following the shift away from tokenmaxxing, as his own startup helps organizations deploy AI and manage token spend.<\/p>\n He told Business Insider that the tokenmaxxing craze stemmed from “FOMO” and companies not understanding the full challenge of building with AI. Now, he said, the focus is shifting to connecting spend to outcomes.<\/p>\n “I think that they should have goals about budgets,” he said. “And a reason why to spend the money: if you want to spend a million on tokens, you should do something good with that in your business.”<\/p>\n He suggested rewarding employees who use AI to build things that benefit the company with a larger token budget and cutting the budget for those who might be “harming” it.<\/p>\n<\/p>\n