
According to a Wall Street See report on May 23, Microsoft has canceled its internal licensing of Anthropic's Claude Code due to excessive token-based costs. This decision carries profound symbolic weight: despite being OpenAI's largest investor, a cloud service provider for Anthropic, and the owner of its own flagship AI coding product GitHub Copilot, Microsoft still chose to "part ways" when confronted with the real costs of pay-as-you-go billing.
Cost Pressure: The "Invisible Killer" of AI Coding Tools
The stated reason for Microsoft's cancellation of Claude Code licensing is "excessive token-based costs," but the deeper rationale is more complex:
First, imbalanced cost structure. As an agent-level coding tool, Claude Code may involve hundreds of thousands or even millions of tokens in context windows per task execution, with a single use costing several dollars. At the scale of Microsoft's tens of thousands of engineers, the monthly bill could reach millions of dollars.
Second, product competition conflict. Claude Code's internal use at Microsoft directly cannibalizes GitHub Copilot's market share. Microsoft acquired GitHub for US$7.5 billion in 2022, and Copilot's subscription revenue is its core asset. If engineers become accustomed to using Claude Code, Copilot's renewal rates will suffer.
Third, the disconnect between strategic investment and commercial reality. Although Microsoft invested US$5 billion in Anthropic and committed to Azure cloud service collaboration, this does not mean internal usage comes "for free." Anthropic's token-based pricing applies equally to Microsoft, and Azure's internal settlement system cannot absorb these costs.
GitHub's "Triple Crisis"
The timing of Microsoft's cancellation of Claude Code licensing coincides with GitHub's unprecedented trust crisis:
Frequent outages: Multiple service disruptions in recent months have impacted developers worldwide.
Security breach: The TeamPCP supply chain attack resulted in the theft of approximately 3,800 internal GitHub repositories.
Talent drain: Core engineers are defecting to competitors like Cursor.
More critically, AI-native coding tools like Cursor are eating away at market share at an alarming rate. Cursor has surpassed one million monthly active users, and its Composer 2.5 integration, powered by Kimi K2.5, has further intensified competition.
Industry Trend: AI Coding Tools Shift from "Subscription Model" to "Pay-for-Outcome"
Microsoft's cancellation of Claude Code licensing reveals a deeper transformation in the AI coding tool business model:
| Model | Representative Product | Advantage | Disadvantage |
|---|---|---|---|
| Subscription | GitHub Copilot ($19/month) | Predictable revenue | Low perceived user value |
| Pay-per-token | Claude Code | Pay for actual usage | Uncontrollable costs |
| Hybrid | Cursor | Subscription + excess tokens | Balances risks for both parties |
Microsoft's choice demonstrates that even tech giants cannot bear the unlimited costs of agent-level AI usage. This signals that the industry will accelerate its transition toward a "pay-for-outcome" model—where users pay not for tokens, but for "tasks completed."
Conclusion
Microsoft's decision to "part ways" with Claude Code is a landmark event in AI commercialization. It proves two things: first, the cost of agent-level AI far exceeds market expectations—even a giant like Microsoft must watch its spending; second, competition in AI coding tools has shifted from a "feature race" to a "cost-efficiency race." When Cursor can deliver comparable functionality at lower cost, GitHub Copilot's subscription model faces a fundamental challenge. Microsoft's dilemma lies in this: it must protect GitHub Copilot's cash flow while simultaneously hosting Anthropic's model services on Azure. This "left-hand-right-hand" strategic contradiction may force Microsoft to accelerate Copilot's agent-level upgrade—or risk being left behind entirely in the AI coding race.