- Corporate AI use has driven revenue growth from $9B in 2025 to $44B in early 2026
- Tokenmaxxing involves employees inflating AI usage metrics by doing unnecessary tasks
- Microsoft and Uber are cutting AI spending due to excessive internal consumption costs
Big corporations have spent the better part of two years encouraging employees to experiment freely with artificial intelligence tools. The bill is now arriving and it is eye-watering. Anthropic's annualised revenue run rate has surged from $9 billion at the end of 2025 to over $44 billion as of early 2026, according to data cited by Jefferies' Christopher Wood in his weekly Greed & Fear note dated May 28.
The leap is breathtaking. However, Wood's larger argument is that the very engine driving that growth — unchecked corporate AI usage — is beginning to stall and the phenomenon has a name: tokenmaxxing.
Tokenmaxxing refers to employees gaming internal AI usage leaderboards by deploying AI tools for unnecessary tasks, simply to inflate their activity scores. The Financial Times reported the practice at Amazon and the Greed & Fear suggests it is hardly unique.
A speaker at an AI conference attended by Wood recently put it plainly: measuring AI tool usage as a KPI makes little sense because more activity does not equal more productivity. As the speaker noted, "time saved is not money saved."
The corporate backlash is already visible. Microsoft has reportedly begun cancelling internal Claude Code licences, moving teams back to GitHub Copilot. Uber's chief technology officer reportedly warned internally that AI usage had burned through the company's entire annual budget in just four months. In response, AI model providers are reportedly pivoting to token-based, pay-as-you-use pricing for enterprise customers, a structural shift that could meaningfully slow consumption.
ALSO READ: Uber And Nvidia Executives Unsure Of Returns On Investments
AI Capex Is Now Bigger Than The Dot-Com Boom
The macro backdrop makes this inflection point all the more consequential. US investment in information processing equipment and software has hit 4.91% of nominal GDP in Q1 2026 — surpassing the previous peak of 4.46% recorded at the height of the Dot-com bubble in Q4 2000. Wood invokes Amara's Law — the principle that people overestimate technology's short-run impact and underestimate its long-run effect — as a caution against both excessive euphoria and excessive pessimism.
Yet the Philadelphia Semiconductor Index, up 80% year-to-date and trading 64% above its 200-day moving average, suggests markets are not in a cautious mood. Memory chipmakers Hynix and Micron have both crossed $1 trillion in market capitalisation this week.
Wood's base case: a lot of capital will be destroyed in this cycle. The question is no longer whether AI is transformative. It is who pays for the transformation and whether the productivity gains will ever match the invoice.
ALSO READ: 'Big Short' Michael Bury Turns Bullish On Ignored Fintech Stocks Amid AI Bubble Warning
Essential Business Intelligence, Continuous LIVE TV, Sharp Market Insights, Practical Personal Finance Advice and Latest Stories — On NDTV Profit.