Microsoft exposes AI cost problem

Fortune reported on May 22 that a growing number of companies are discovering an awkward part of enterprise AI adoption: wider use can push bills above the cost of the labor the software is meant to replace. The article said Microsoft has started canceling most of its direct Claude Code licenses and steering engineers toward GitHub Copilot CLI after a rapid internal rollout drew heavy usage. Fortune also cited Uber as another example of adoption outrunning budgets, as companies press employees to use more AI tools while trying to contain the resulting costs. Reuters separately reported this week that Starbucks terminated an AI inventory program across North America nine months after launch because the system made basic counting and labeling mistakes. Together, the two reports point to the same operational problem: enterprise buyers are not only asking whether AI works, but whether it works reliably enough, and cheaply enough, to justify production deployment. (finance.yahoo.com) ### Why are AI bills rising even as model prices fall? Fortune said the pressure is coming from usage, not just sticker price. The publication described an “AI paradox” in which cheaper tokens still lead to larger total bills as companies increase prompts, expand seats and layer on agent-based workflows that run more tasks per employee. (cnbc.com) CRN reported after Microsoft’s third-quarter earnings that Chief Executive Satya Nadella described an AI business model shifting toward a mix of per-seat and consumption-based pricing, especially for agents. That means enterprise spending can rise with adoption even when unit costs fall. ### What did Microsoft and Uber show about enterprise usage? (finance.yahoo.com) Fortune said Microsoft had opened access to Claude Code to thousands of developers, project managers and designers before later pulling back most direct licenses. Microsoft did not comment to Fortune, and Anthropic did not immediately respond to the magazine’s request for comment, according to the report. (crn.com) Uber’s example was more explicit. Fortune cited Uber CTO Praveen Neppalli Naga as saying in April that the company had used up its entire 2026 budget for AI coding tools in four months, after internal efforts to encourage adoption. That is the kind of number finance teams and sales-operations teams watch closely because it separates curiosity from controlled spend. (finance.yahoo.com) ### What went wrong at Starbucks? Reuters reported that Starbucks ended the worker-facing inventory program this week after deploying it across North American stores about nine months earlier. The system, part of CEO Brian Niccol’s effort to fix product shortages, was designed to automate stock counts for store workers. (vuink.com) Reuters said the tool struggled with basic identification tasks, including confusing products during counts, and Starbucks reverted to manual inventory checks. Reports from other outlets described the errors in more detail, including problems distinguishing similar milk types, but Reuters’ account established the central point: the system did not perform reliably enough in store operations. (cnbc.com) ### Why does this matter for pipeline qualification? Microsoft’s pullback and Starbucks’ rollback show two different failure modes. One is economic: usage expands faster than budgets. The other is operational: the tool reaches production but does not perform consistently enough to stay there. Both complicate enterprise buying decisions. (cnbc.com) For sales and finance teams, that means AI opportunities cannot be treated as one budget bucket. Exploration spending, pilot spending and validated production spending behave differently, and recent reporting suggests buyers are starting to scrutinize those categories more closely as token and agent costs become recurring line items. CNBC reported this month that many large companies are now tracking AI usage because costs are becoming standard business expenses, even though measurable earnings impact remains less common. (vuink.com) ### What should readers watch next? Microsoft’s next public readout on AI monetization is likely to come through future earnings commentary and product-pricing disclosures, especially around Copilot and agents. Starbucks’ next visible step will be whether it replaces the withdrawn inventory system with a revised internal tool or another vendor in North America. Fortune’s May 22 report and Reuters’ May 21 account are the clearest markers so far of how quickly AI enthusiasm can run into budget controls and operational tests. (cnbc.com) (crn.com)