MicrosoftのGitHub、コスト危機を背景にAI利用の従量課金制へ移行

Microsoft is closing the AI buffet offered to GitHub Copilot customers, acknowledging that it can’t sell AI like Red Lobster's Endless Shrimp.

The US seafood restaurant's all-you-can-eat shrimp promotion led the company to bankruptcy in 2024 and while Microsoft is nowhere near so financially overextended, the software giant's code hosting biz has decided it no longer wants Copilot to operate at a loss.

GitHub is therefore shifting Copilot from request-based billing to usage-based billing on June 1, 2026.

GitHub absorbed much of the escalating inference cost, but the current premium request model is no longer sustainable

Under request-based billing, GitHub Copilot subscribers will be allowed to submit a set number of premium requests, with certain models priced at a higher request rate but without any consideration for the complexity of the request. So complex prompts that require a lot of "thinking" often cost GitHub more than the company earned in subscription fees.

"Today, a quick chat question and a multi-hour autonomous coding session can cost the user the same amount," explained Mario Rodriguez, chief product officer on the GitHub product team, in a blog post. "GitHub has absorbed much of the escalating inference cost behind that usage, but the current premium request model is no longer sustainable."

Under usage-based billing, there's a more direct correlation with metered tokens – sets of three or four characters that represent the basic economic unit for selling AI services.

It's not quite as simple as $X for X tokens – different models meter tokens at different rates – so GitHub has devised a virtual currency unit called GitHub AI Credits that's worth $0.01.

Copilot customers consume input tokens, output tokens, and cached tokens, each priced based on the model used. Microsoft converts that to a cost measured in AI Credits.

"Instead of counting premium requests, every Copilot plan will include a monthly allotment of GitHub AI Credits, with the option for paid plans to purchase additional usage," said Rodriguez. "Usage will be calculated based on token consumption, including input, output, and cached tokens, using the listed API rates for each model."

Knowing the outcome of this calculation in advance will be difficult – usage-based billing is non-deterministic, so users can never be sure how much time, and how many tokens, a model will consume to respond to a specific input. Different prompts may involve tools that complicate token consumption calculations.

GitHub at least intends to try to give customers a hint of what's coming. Rodriguez said the company will introduce "a preview bill experience in early May, giving users and admins visibility into projected costs before the June 1 transition."

AI companies were taken by surprise when OpenClaw attracted widespread attention in February, prompting a surge of experimentation with AI agents running 24/7 on various tasks. And the increasing competency of AI models around this time also encouraged more developers to explore AI coding.

As a result, companies offering subsidized access to AI services through subscription plans faced more demand than they could satisfy with their inferencing infrastructure. The price correction that followed has been rippling across the industry.

• Microsoft updates the Windows Update Experience: You can hit pause now

• SpaceX dusts off Falcon Heavy for first flight in 18 months

• Cursor-Opus agent snuffs out startup's production database

• AI reality check: Here's what three companies learned building wallets, homes, and games

GitHub last week signaled its intent to stanch the red ink by suspending the creation of new Copilot, Pro, Pro+, and Student plans.

Before that, Anthropic and Google took steps to limit some uses of its services. OpenAI responded by debuting a more expensive $100 subscription tier in an effort to boost usage of its Codex model, even as the company is contemplating an end to unlimited usage under subscription plans. Cloud providers like AWS and Azure have been dealing with capacity challenges too.

GitHub's subscription rates will remain the same: Copilot Pro is $10/month, Pro+ is $39/month, Business is $19/user/month, and Enterprise is $39/user/month.

At the $0.01 GitHub AI Credit rate, Copilot Pro subscribers get 1,000 AI Credits per month. Copilot Pro+ subscribers get 3,900. Once users exhaust the usage allowed under a plan, they can define an overflow budget – or just stop using AI until the next monthly billing cycle resets their AI Credit balance.

Organizations and enterprises will receive 1,900 and 3,900 API Credits per user per month respectively. However, existing Copilot Business and Copilot Enterprise customers get a higher number of API Credits from June 1 through September 1, 2026, at 3,000 and 7,000 respectively.

Users on annual subscription plans have the option to cancel and receive a pro-rated refund or to be downgraded to Copilot Free upon subscription expiration – those plans will not be renewable. Regardless, those riding their annual subscription plans to the bitter end will see prices skyrocket for premium models. For example, Anthropic's Opus 4.7, subject to a 7.5x multiple under request-based billing, will see its multiplier jump to 27 going forward. OpenAI's GPT-5.4 will see its multiplier rise from 1x to 6x.

The new regime isn't entirely metered. Subscribers who reach their AI Credit limit can continue to access Copilot for code completions and Next Edit Suggestions – these services are unlimited on paid plans.

That's more than you can say about Endless Shrimp, recently revived for a limited time only. ®