Silicon Valley technology firms are rapidly increasing consumption of AI tokens, driving sharp growth in budgets with unclear benefits to productivity, according to multiple industry sources [1, 2]. Uber exhausted its entire 2026 AI budget within the first four months, driven primarily by widespread internal adoption of Anthropic’s Claude Code AI programming tool [1, 3, 4, 2]. Uber CTO Praveen Neppalli stated that individual engineers spend between $150 and $2,000 monthly on Claude Code, with some sessions costing as much as $1,200 in two hours [3, 4]. The company also introduced an internal leaderboard incentivizing Claude usage, prompting competition that increased token consumption [4].

Despite high usage, Uber COO Andrew Macdonald said, “We have not established a direct correlation, and while it may indirectly improve product ship rates, it is hard to equate these stats to ‘we indeed produced 25% more useful consumer features.’” [1] Silicon Valley’s growing AI spending has raised concerns over “tokenmaxxing,” a term describing excessive AI token consumption without corresponding return on investment [1, 5, 2]. Investor Michael Burry called tokenmaxxing possibly “a crazy, rushed, and transient phase.” [1]

Microsoft, another major user of Claude Code, began revoking internal licenses in mid-May 2026 due to ballooning costs and planned to terminate most access by June 30 [6, 3]. Still, internal Microsoft reports claim AI tools can deliver productivity improvements up to 80%, suggesting some firms see value despite cost issues [3]. Meanwhile, some companies like Meta, Visa, and Morgan Stanley closely monitor employee AI usage to assess cost and benefit balances [1, 4].

Industry leaders emphasize the learning curve. Nvidia CEO Jensen Huang remarked, “You will not be replaced by AI but by someone who masters AI.” [1] Google CEO Sundar Pichai warned at 2026 I/O that companies overspending on AI tokens would face worsening budget challenges. “I believe this problem will get worse for the rest of this year,” he said [1, 2].

In April 2026, Anthropic announced a shift to full API pricing for Claude’s agent tools, raising costs for enterprise users and triggering pushback [4]. Despite criticisms over rising costs and perceptions that AI product quality or capability have become more restrictive, Anthropic expects Q2 2026 revenue to reach $10.9 billion and potentially post its first profitability [6, 3, 5].

Top tech firms reported galloping AI capital expenditures exceeding $650 billion in Q1 2026, with 85% of companies misforecasting AI spending by more than 10%, leading to gross margin drops near 6 percentage points [3].

The rapid adoption in companies like Uber, where 84% of engineers were active Claude Code users by March 2026, illustrates the intensity of AI integration but also raises flags over cost efficiency [4]. Microsoft’s rollback signals growing financial controls in response to rising bills [6, 3]. By the end of June, Microsoft plans to end most Claude Code access for its Experiences and Devices division [3].