Microsoft cancelled the majority of internal Claude Code licences across its Experiences and Devices division, the team behind Windows, Microsoft 365, Outlook, Teams, and Surface, effective 30 June 2026, redirecting engineers toward GitHub Copilot CLI in a decision exposing an uncomfortable truth: enterprise AI tools deliver productivity gains that companies cannot afford to sustain at current token pricing structures.

The withdrawal follows a six-month deployment that saw usage rates climb to 84–95 per cent amongst thousands of engineers, with per-user API costs reaching $500–$2,000 monthly as continuous usage patterns emerged once genuinely useful tools became available.

The contradiction defines enterprise AI’s current crisis. The technology works sufficiently well that adoption accelerates until expenditure becomes unsustainable, forcing retreat regardless of productivity benefits.

Microsoft represents a visible manifestation of a broader corporate reality.

Uber’s CTO, Praveen Neppalli Naga, disclosed that the company exhausted its entire planned 2026 AI coding budget by April, just four months into the fiscal year, with individual engineers generating token costs comparable to junior developer salaries whilst 70 per cent of committed code now originates from AI assistance.

Even Nvidia, the company building chips powering AI infrastructure, faces internal economic questions.

An Nvidia vice president publicly acknowledged that running AI costs his division more than employing human staff, a striking admission from a firm whose business model depends on accelerating AI deployment to validate continued chip demand growth.

The challenge stems from token-based pricing models that charge per API request.

Short interactions remain affordable. However, AI coding assistants require continuous context, reading codebases, generating suggestions, explaining logic, debugging errors, and creating persistent API calls throughout engineering workflows.

When tools prove genuinely useful, usage becomes constant rather than occasional, transforming manageable pilot costs into unsustainable production expenses.

AI software prices across the United States climbed 20–37 per cent according to industry tracking, whilst GitHub responded to margin pressure by shifting all Copilot plans toward usage-based billing through AI Credits, launching on 1 June 2026.

The pricing adjustments acknowledge that current flat-rate structures cannot accommodate actual consumption patterns emerging as adoption scales.

MIT research found AI automation proves economically viable in only 23 per cent of jobs, with humans remaining cheaper across the remaining 77 per cent despite the technology demonstrating the capability to perform those tasks.

The gap between technical capability and economic viability creates a strategic dilemma for enterprises committed to AI transformation narratives whilst confronting budgets incapable of supporting sustained deployment.

Gartner positioned generative AI squarely in the “trough of disillusionment”, predicting that 25 per cent of planned 2026 AI budgets will slip into 2027 as proofs of concept fail during procurement review.

A separate Gartner analysis found that only 28 per cent of AI infrastructure projects fully deliver against their business cases, suggesting a market repricing rather than a technology experiencing an awkward adolescence.

Microsoft’s retreat toward GitHub Copilot CLI, a tool it owns rather than licenses, signals vertical integration as a potential solution.

When token costs from external providers become prohibitive, controlling the entire stack from model through deployment enables improvements in unit economics that are impossible whilst paying per-token retail rates.

For enterprises lacking Microsoft’s resources to build proprietary alternatives, the calculation becomes stark: curtail AI tool usage despite productivity benefits, or accept technology expenditure that may exceed headcount costs for affected roles.

Neither option aligns neatly with transformation narratives that dominated corporate AI strategy documents throughout 2024 and 2025.

Big Tech collectively committed approximately $740 billion toward AI-related expenses in 2026, a 69 per cent increase from 2025, suggesting the industry believes current economics represent temporary pricing inefficiencies rather than a fundamental constraint.

Whether costs decline sufficiently to enable mass adoption, or whether AI tools remain expensive enough to limit deployment to the highest-value use cases, will determine whether 2026 represents a correction or an inflexion point in enterprise AI’s trajectory.

ALSO WATCH:MARKETING EDGE ONTV