Enterprise AI used to come at a predictable price. Amnesty International rewrote the draft law.
Uber The 2026 AI budget was completely exhausted by April. Technology Manager Praveen is a Nepali Naga He said the company “went back to the drawing board.” according to Information. coo Andrew MacdonaldThe productivity issue is far from closed, he said, speaking on the Rapid Response podcast, Fortune I mentioned. “This connection does not exist yet,” he said. “It’s very difficult to draw a line between one of those statistics and, ‘Okay, now we’re actually producing 25% more useful consumer features.'” Uber will weigh nominal costs directly against the cost of hiring engineers, McDonald said.
On the company’s earnings call, according to the same Fortune report, CEO Dara Khosrowshahi Independent customers have built roughly 10% of the committed code, he said. Uber’s R&D spending reached $3.4 billion in 2025, up 9% year over year. The tools work. Math is not like that.
Microsoft is backing down
Microsoft Hit the same ceiling. The company began revoking most internal Claude Code licenses in mid-May, and redirected engineers across its Experiences and Hardware division to the GitHub Copilot CLI by June 30, the end of its fiscal year, according to a separate Fortune magazine. a report. Six months ago, Microsoft made Claude Code accessible to thousands of employees across engineering, product, and design. Fortune noted that the cancellation does not affect Microsoft’s broader Foundry deal with Anthropic, which includes an investment of up to $5 billion.
The financial dispute between the two companies stems from a structural mismatch between how they price AI tools and how their enterprise finance teams are built. AS PYMNTS I mentionedannual licenses and seat-based pricing gave CFOs a consistent cost structure they could predict. Token-based consumption, where fees accrue based on the volume of text being processed and generated, has broken this model.
An increase in internal trials, a new product feature, or a poorly optimized claim can cause costs to rise in ways that are difficult to anticipate. Engineering decisions now have direct consequences on the balance sheet, which finance teams are not structured to track.
Advertisement: Scroll to continue
Symbols as a defective agent
The problem goes deeper than bill fluctuations. Symbols measure volume, not score. piments I mentioned Enterprises are increasingly using token consumption as a primary measure of AI adoption and workflow density, but the signal has limits. A poorly structured prompt that forces the model to iterate and regenerate consumes more tokens than a directed, summary query, yet does not necessarily produce useful output.
Nvidia CEO Jensen Huang framed the risks at GTC in March. “I can fully imagine that in the future every engineer in our company will need a nominal annual budget,” Huang said, estimating that such allocations could be as much as half the value of the base salary. according to Computer world.
The billing problem is compounded by agents
Proxy coding tools double the cost exposure compared to standard chatbot interactions. A one-turn conversation creates one inference call. A proxy session, where the model plans, executes, verifies, and self-corrects through multiple steps, generates more steps.
AS PYMNTS I mentionedAnthropic has moved to usage-based billing for enterprise customers, and SaaS companies including Salesforce and HubSpot are preparing to adopt results-based pricing. Adobe Announce Results-based pricing for new AI product suite in April. GitHub is transitioning all Copilot plans to usage-based billing through AI credits starting June 1, replacing a fixed-rate license that obscures actual consumption.
Google CEO Sundar Pichai put the total volume into context at this month’s I/O conference, revealing that the company now processes 3.2 quadrillion tokens per month, a seven-fold increase from 480 trillion a year earlier. Record I mentioned Pichai acknowledged the dynamic of consumption: “Now some might call this tokenism and maybe there is some truth to that.”
