Microsoft PullBack, Uber Budget Blowout Put AI Coding Costs in Focus
Rapid adoption of AI coding tools is exposing cost and reliability pressures, forcing companies to rethink how they deploy them.
Topics
News
- The Vatican Enters the AI Governance Debate
- Microsoft PullBack, Uber Budget Blowout Put AI Coding Costs in Focus
- Shastra VC Launches $100 Million Deeptech Fund
- Persistent, IIM Ahmedabad Launch AI Value Compass for Enterprise AI
- India Launches National AI Doctors Mission to Train Doctors in AI
- HCLTech Says 43% of Enterprise AI Projects May Fail
Microsoft Corp. is scaling back internal use of Anthropic’s Claude Code and moving many developers toward its own GitHub Copilot CLI, The Verge reported earlier this month.
The move comes about five months after Microsoft opened Claude Code access to thousands of employees in December 2025, including engineers, product managers and designers.
The tool gained traction inside the company, but The Verge reported that Microsoft now plans to remove most licences, with its Experiences + Devices team winding down use by the end of June.
Microsoft has framed the shift as an effort to converge on Copilot CLI as its main agentic command-line coding tool. The timing also falls at the end of Microsoft’s financial year on 30 June, making licence cuts a way to reduce operating costs, The Verge report said.
The decision does not end Microsoft’s broader relationship with Anthropic. Claude models remain available through Microsoft Foundry, the company’s Azure-based AI platform, under a partnership announced in November last year. As part of that arrangement, Anthropic committed to buying $30 billion of Azure compute capacity, while Microsoft made Claude available to enterprise customers alongside OpenAI models.
The shift comes as Microsoft’s developer platform faces reliability and security pressure.
GitHub, which Microsoft acquired in 2018, has reported repeated service disruptions this year.
On 11 March, GitHub CTO Vlad Fedorov wrote in blog post that the company had “not met our own availability standards,” citing rapid usage growth and scaling limits across parts of its architecture.
GitHub’s own availability report, published on 14 May, said the platform experienced 10 incidents in April that degraded services, including code search, Copilot, webhooks, Git operations, GitHub Actions and deployments.
GitHub also confirmed a security incident on 20 May in which about 3,800 internal repositories were affected after an employee installed a malicious Visual Studio Code extension, SecurityWeek reported. GitHub said its current assessment was that the activity involved internal repositories only.
Cost pressure is showing up elsewhere, too. Uber exhausted its full-year 2026 Claude Code budget within months, The Information reported on 14 April.
Business Insider, citing that report, said on 25 May that Uber CTO Praveen Neppalli Naga’s comments had triggered internal discussions about AI token consumption and its trade-offs.
Uber COO Andrew Macdonald said in a podcast interview released on 23 May that the company had not yet drawn a clear link between higher AI token use and more useful consumer features.
The strain reflects a broader problem for enterprise AI deployment. Unlike traditional software subscriptions, many AI coding tools are priced by usage, meaning costs rise with token consumption and compute demand.
That model can make spending harder to forecast once AI coding tools move from pilots to daily engineering workflows. Internal leaderboards, aggressive adoption targets and broad access can push usage higher before companies have clear measures of productivity or return.
Competition is also intensifying. Claude Code and Cursor have gained developer attention, while GitHub Copilot is under pressure to improve reliability and keep pace with rival tools.
