MW Wasted AI budgets at Uber, Microsoft and Nvidia trigger hiring - because human workers are cheaper
By Hardika Singh
Uber blew its entire 2026 AI budget by April. Here's why replacing workers with bots backfired.
Unproductive AI spending is catching employers off guard.
Corporate executives are realizing that it's more expensive to replace employees with artificial intelligence, contrary to what was previously believed. That's good news for the U.S. labor market.
What's causing this about-face? Blame the tokens. When an employee asks AI for help, the request consumes digital tokens, which are the currency of LLMs and a cost of doing business.
Workers at tech companies in particular have been encouraged to ramp up their token consumption - known as "tokenmaxxing" - by employers conflating AI tools with productivity. It has become somewhat of a status symbol amongst employees to be tokenmaxxing to prove that you're going above and beyond to meet key performance indicators.
But tokenmaxxing is expensive. For text-based inquiries, the math is simple, with 750 words costing 1,000 tokens. But generating video, images, code and more can cost a lot more - and it's not until after the task has been carried out that the bill comes due.
These costs have been catching employers off guard, especially with the increased use of agentic AI sucking up tokens. As a result, employers are frequently blowing through their AI budgets, putting them in a situation where it's no longer clear that using AI is cheaper than hiring people.
For example, Uber Technologies' (UBER) operations chief recently raised concerns about the cost of tokenmaxxing, especially because by April, the ride-hailing company had already blown through its entire 2026 AI budget. Microsoft $(MSFT)$ has canceled its Claude Code licenses and asked employees to use its own GitHub Copilot CLI - with many speculating that high costs drove the decision. A team at Nvidia (NVDA) for months has reported higher costs for AI than humans. Amazon.com (AMZN) axed its internal AI leaderboard. One company even accidentally spent $500 million in a month recently on Claude, according to Axios.
While Big Tech stocks may still be flying high, some companies are realizing they can't sustain this level of spending on tokens.
The widespread pullback shows that companies are dissatisfied with the return on investment on AI. Big Tech companies have poured billions of dollars into building out AI infrastructure over the past few years in hopes that it's the future. Enterprises have adopted AI into their workflows to fare better against competitors. While Big Tech stocks may still be flying high, some companies are realizing they can't sustain this level of spending on tokens.
Companies have tapped AI to make work teams leaner, faster and more efficient. Roughly 117,000 workers have been laid off so far this year at 164 tech companies, according to Layoffs.fyi, while close to 125,000 employees at 275 companies lost their jobs in 2025.
But the backlash against tokenmaxxing suggests that the worst is over for the labor market.
Token prices have risen about 60% since the end of February, driven by more users turning to agentic AI to independently carry out tasks and a crippling shortage of memory and storage. That token-price increase has tested some companies' threshold for what they're willing to pay for AI versus human capital.
The bigger problem than the cost is that AI doesn't automatically lead to improved productivity and actionable results on the bottom line, making it seem more like a fun perk to have rather than an absolute must - especially when it costs so much. Uber's operations chief pointed out that there's no proof yet that tokenmaxxing leads to improved consumer features.
Just as it's hard to prove a direct link between the job-market slowdown and AI, it's a stretch to credit AI for the U.S. economy's increase in productivity. There's currently no good way to measure a relationship between AI and productivity.
Read: The economy is creating more jobs, but these groups are having trouble finding work
Amazon tried tracking AI usage and displaying scores on a companywide leaderboard, but the technique failed because workers started inflating their usage scores, meaning they were likely not being very productive. (Of course, that's not the case everywhere, but generally speaking, when workers are pressured to carry out a task that they don't necessarily want to, it rarely leads to a positive impact on the bottom line.)
Another concern is that token consumption will keep increasing even if token prices decline, especially once OpenAI and Anthropic go public. Neither of these companies is profitable at the moment, but the tide can quickly turn. Anthropic is expected to post an operating profit of $559 million in the June quarter, thanks to its popular coding tools. In that scenario, it won't take long before executives spin "powered by real people" into the new, hot trend over "powered by AI."
Still, workers clearly have some time before AI relegates them to the permanent underclass.
Hardika Singh is an economic strategist at Fundstrat Global Advisors, an investment research firm. Read her disclosures here: https://fundstratdirect.com/disclosures/hardika-singh/.
More: Jobless claims jump to a 4-month high, but don't be fooled: Layoffs aren't on the rise
Also read: Amazon, Microsoft and Google are quietly morphing their businesses - and Wall Street is missing the big picture
-Hardika Singh
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
June 04, 2026 16:09 ET (20:09 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments