After AI Frenzy, Tech Giants Now Curb Employee Use to Slash Soaring Costs
Earlier this year, major technology companies urged employees to incorporate artificial intelligence into their work as much as possible. Employees referred to this practice as 'tokenmaxxing,' where a token is a unit of AI output roughly equivalent to a word fragment. At companies like Meta and Amazon, workers competed on leaderboards that tracked token usage.
However, the costs of providing AI tools—supplied by companies such as OpenAI and Anthropic—have proven substantial. In response, many firms are now limiting AI usage, marking a shift toward what some call 'tokenminning,' or minimizing token consumption.
Meta announced last week that it would soon restrict AI use after observing an 'exponential increase' in expenses. Uber revealed in May that it had exhausted its projected AI budget for the year in just four months and has since implemented monthly caps on AI coding tools. Walmart has also set limits on various AI applications, and both Amazon and Meta have removed the tokenmaxxing leaderboards.
The rapid reversal highlights the ongoing uncertainty around AI integration. 'The biggest problem is this is all changing so fast, people and companies don’t know what to do,' said Rob May, CEO of Neurometric, a startup that advises businesses on AI efficiency. He noted that measuring employee AI savviness by token volume encouraged usage over effectiveness.
AI companies like OpenAI and Anthropic offer tiered subscriptions ranging from $10 to $200 per month, with usage limits. However, the bulk of their revenue comes from enterprise clients—including Meta, Shopify, and Amazon—who pay subscription fees plus per-token costs for thousands of employees. Simple tasks, such as summarizing a meeting transcript, may use a few hundred tokens; complex requests like writing code can consume tens of thousands.
As AI models grow more powerful, costs have escalated. Anthropic’s latest model, Fable, is twice as expensive as its predecessor, Opus. Many employees default to using the most advanced models for all tasks, driving up expenses. Additionally, the rise of AI 'agents'—systems that work autonomously for extended periods—has led some engineers to use tens of thousands of dollars’ worth of tokens per month.
Companies are now reassessing returns on AI investments. Andrew Macdonald, Uber’s chief operating officer, said in a podcast that linking AI spending directly to shipped features remains difficult. 'That link is not there yet,' he stated.
Despite cutbacks, large-scale AI investment continues. Meta plans to spend billions on AI this year but aims to identify areas for savings. Salesforce CEO Marc Benioff announced that his company will spend hundreds of millions on AI, but now tracks 'agentic work units' rather than tokens to measure output rather than mere usage.
The long-term impact on AI providers remains uncertain. During the peak of tokenmaxxing earlier this year, AI companies reported record revenues.
