The world’s biggest tech companies promised that AI would make everything cheaper, faster and more efficient. That pitch still gets repeated at every conference and earnings call. But behind closed doors in 2026, many of those same companies are quietly pulling back because the bills have become impossible to ignore. Meta, Microsoft and Uber are all wrestling with AI costs that spiralled well beyond what anyone planned for, and a workplace habit called tokenmaxxing is sitting right at the centre of it.
The problem did not come from ambitious consumer products or large public-facing AI launches. It came from inside their own offices. Employees using AI tools every day, chasing spots on internal leaderboards, ran up costs that nobody at the top fully anticipated until it was already too late.
What Tokenmaxxing Actually Means
The word itself tells you a lot. Tokenmaxxing puts together “token,” the unit that AI systems charge for, and “maxxing,” which basically means pushing something as far as it will go. A token is roughly four characters of text. Every question you ask an AI and every answer it gives back uses them, and companies pay for every single one.
What started as a way to track how much employees were using AI tools slowly turned into something else entirely. Usage became a performance signal. Workers stopped being judged only on what they produced and started being judged on how much AI they were using while producing it.
Nvidia CEO Jensen Huang pushed this thinking hard, saying engineers should be using AI tokens worth at least half their annual salary just to be considered fully productive. That kind of statement lands differently when it comes from one of the most powerful figures in the industry. At Amazon, employees admitted to using AI for pointless tasks simply to boost their internal scores. The same thing happened at Microsoft and Meta. When you reward people for using more AI, they use more AI. Whether it actually helps is a separate question entirely.
The cost side became even worse when companies started deploying agentic AI, the kind that works through a problem step by step rather than answering a single question. That approach burns through tokens at a rate a regular chatbot never could. Some companies burned through their entire annual AI budget in three months.
The Numbers That Made Executives Stop and Stare
Uber’s CTO Praveen Neppalli Naga told The Information in April that the budget he had mapped out for 2026 was already gone, and it was barely month four of the year. Uber’s COO Andrew MacDonald called it a “head-exploding moment” and said the company had no choice but to start having serious conversations about what it was actually spending on AI, because it clearly was not free.
Microsoft cancelled most of its internal Claude Code licences roughly six months after rolling them out. Microsoft’s AI chief Mustafa Suleiman told Bloomberg that Anthropic was extremely expensive and that the company was actively trying to reduce and eventually eliminate that dependency. One unnamed company, reported by Axios, somehow spent half a billion dollars on AI tokens in a single month after nobody thought to put a cap on usage. Amazon and Meta quietly pulled their internal AI leaderboards once the true cost of those rankings became clear.
Where Things Go From Here
For the people actually doing the work, the pressure has not gone away. Some employees still feel like they have to use AI for everything because someone somewhere is watching the numbers. Some developers have started paying out of their own pockets for AI tools just to meet expectations set from above.
The broader tokenmaxxing trend is running out of steam. Politicians are starting to ask questions. Data centre expansion is facing pushback. Even people who work in AI are calling token counts a metric that measures activity rather than value.
The honest reality is that most companies are still in the experiment phase. Only one in four organisations has reached anything close to full AI adoption, while nearly half are still running limited tests. The money keeps going out while the measurable return stays hard to pin down.
At some point, the question stops being how many tokens your employees are burning through. It becomes whether any of it is actually moving the business forward.
Also Read: Mark Zuckerberg Admits Mistakes During Meta’s AI Restructuring, Says No More Mass Layoffs In 2026
Syed Ziyauddin is a media and international relations enthusiast with a strong academic and professional foundation. He holds a Bachelor’s degree in Mass Media from Jamia Millia Islamia and a Master’s in International Relations (West Asia) from the same institution.
He has work with organizations like ANN Media, TV9 Bharatvarsh, NDTV and Centre for Discourse, Fusion, and Analysis (CDFA) his core interest includes Tech, Auto and global affairs.
Tweets @ZiyaIbnHameed
