Microsoft announced that GitHub Copilot will say goodbye to its $10 fixed subscription on June 1, shifting to Token-based billing (the billing unit for AI processing text). This marks the end of the all-you-can-eat era for AI tools, as cost risks are fully passed on to users.
What this is
We observed a clear polarization in the AI sector this week. On one hand, model capabilities and cash burn are skyrocketing: Sam Altman hinted that GPT-5.5 could autonomously plan a press conference; OpenAI projects losses to hit $14 billion by 2026; Meta laid off 10% of staff to buy GPUs; Apple shelled out $100 billion to revamp Siri. On the other hand, the application layer is starting to pinch pennies: Microsoft not only shifted Copilot to usage-based billing but also axed the free baseline model, launching Agent 365, a management tool for Agents (AI programs that autonomously execute tasks), to charge management fees. Meanwhile, it was revealed that 79.8% of Claude's active users have household incomes over $100,000, signaling a significant premiumization trend; China's DeepSeek-V4 slashed its computing costs by a third by adapting to Huawei's Ascend chips.
Industry view
We judge that the shift from fixed subscriptions to Token billing is an unavoidable inflection point in AI commercialization. The high cost of LLM inference makes usage-based charging an inevitable choice for giants to recoup costs. But the risks of this pivot are equally sharp: it will directly kill developers' willingness to experiment. When every code rewrite and full-library index incurs a fee, innovation in small and medium teams will be locked down by exorbitant bills. What concerns us more is that the "Matthew Effect" in AI is intensifying—Claude has become an exclusive tool for executives, while grassroots developers not only have to count Tokens but also face AI's "passive resistance" (e.g., Claude Code refusing to execute tasks with specific naming). This uncontrollable underlying policy is driving up the hidden costs of human-AI collaboration.
Impact on regular people
For enterprise IT: IT budgets are shifting from predictable SaaS subscription fees to volatile computing bills. Enterprises must establish internal AI usage limits and approval workflows.
For the individual workplace: Collaborating with AI adds a new "coddling" skill—not only must you know how to write prompts, but also how to circumvent AI's passive resistance and optimize Token consumption.
For the consumer market: Top-tier AI tools are becoming exclusive productivity levers for high-income individuals; the barrier for ordinary people to access AI dividends is rising instead of falling.