Jeffries said corporate spending on AI has surpassed the peak of the dot-com boom in the late 1990s relative to the U.S. economy, but a backlash against rising costs is starting to emerge as encouraging employees to freely use AI has proven much more expensive than expected.
U.S. investment in information technology and software reached a record 4.91% of gross domestic product in the first quarter of 2026, surpassing the previous peak of 4.46% at the height of the dot-com bubble in 2000. Additionally, AI-related spending contributed 1.34 percentage points to U.S. economic growth of 2% annually over the same period.
But in a note for Greed & Fear, Jeffries points to reports that Microsoft has begun revoking internal licenses for Claude Code, an AI-powered coding tool made by Anthropic, and reverting to its own GitHub Copilot product, pointing to growing signs that the company is reconsidering.
Uber’s chief technology officer reportedly warned internally that the company had burned through its entire 2026 AI budget in just four months.
The cause is a phenomenon that Jeffries dubbed “token maxing.” The term comes from the way AI models charge for output measured in units called tokens, similar to how printers charge per page.
As companies began to reward employees for high AI usage through internal leaderboards, employees began charging for AI for unnecessary tasks just to inflate their scores, incurring costs without increasing productivity.
Meanwhile, cloud service providers have increased prices by around 30% year-over-year, further increasing pressure on corporate budgets as the cost of AI output per token is falling much faster than the underlying cost of computing power.
Jefferies points out that Anthropic, the AI company behind Claude, has seen its annual revenue run rate jump from $9 billion at the end of 2025 to more than $44 billion by early May 2026, indicating that both the scale of enterprise AI deployments and the associated cost burden are being absorbed across the economy.
Despite cost pressures, the wave of widespread infrastructure spending shows no signs of slowing, with the Philadelphia Semiconductor Index, a gauge of semiconductor stocks, up 80% since the beginning of the year and trading 64% above its 200-day moving average.
Jefferies’ base scenario is that the current AI spending cycle ultimately destroys significant amounts of capital, drawing parallels to the dot-com boom bust and the 19th century railroad mania.
The bank also invoked Amara’s Law, which states that people tend to overestimate the short-term impact of technology while underestimating its long-term impact.
The biggest winners so far remain memory chip makers, with SK Hynix and Micron both having market capitalizations of more than $1 trillion this week.
