GitHub’s Best Month Ever Was Bought by the Meter
GitHub Copilot’s shift to a usage-based billing model is the moment its AI coding tools stopped being a simple add-on and started behaving like core infrastructure, changing how enterprises think about cost, risk, and adoption of automated software development. On June 1, GitHub moved Copilot from flat-rate, per-user plans to billing based on how much customers use the tool, replacing a fixed number of requests with metered consumption. Within weeks, Chief Technology Officer Vladimir Fedorov told employees that June was “by far our best month ever,” directly linking the record customer usage to the new pricing approach. The message is blunt: when AI coding tools are priced like cloud compute, enterprise behavior changes fast. Instead of arguing about seats, teams can test agents, scale pilots, and let real workloads determine what they pay.
Inside the New GitHub Copilot Pricing Economics
The headline is not the price tag; it is the meter. GitHub replaced its old premium request units with GitHub AI Credits, which are consumed by token usage across input, output, and cached tokens, with different costs depending on the model a user runs. For Copilot Business customers, the list price is still USD 19 (approx. RM90) per user per month and includes USD 19 (approx. RM90) in monthly AI Credits, while Copilot Enterprise remains USD 39 (approx. RM185) per user per month with USD 39 (approx. RM185) in credits. Business customers receive an extra USD 30 (approx. RM140) in credits for June, July, and August, and Enterprise customers get USD 70 (approx. RM330), with usage pooled across the organization. That pooling matters: unused credits from one developer no longer sit idle while another team burns through their quota. The billing logic finally matches how software teams work—uneven, bursty, and highly skewed toward a subset of heavy users.
When AI Coding Tools Become Infrastructure, Flat Rates Break
GitHub’s surge shows what happens when AI coding tools become everyday infrastructure rather than novelty features. The platform had its “best month ever” thanks to growing demand for AI coding, with the usage spike arriving right after Copilot’s move to consumption-based billing. This was not a cosmetic change. Mario Rodriguez, GitHub’s chief product officer, pointed out that under the old setup a quick chat question and a multi-hour autonomous coding session could cost a user the same, which stops making sense once agents are doing serious work. Once AI coding is woven into daily workflows, flat-rate plans no longer match either cost or demand. Heavy users drive inference bills, and the vendor has to choose between swallowing that cost or rebalancing the model. GitHub chose to rebalance—and its June performance suggests enterprise teams were waiting for the product and the business model to align.
The User Backlash and the New Budget Conversation
Usage-based billing is not painless for developers. Once GitHub switched Copilot to the new meter, some customers started posting screenshots of projected monthly bills hundreds of dollars above their previous charges; one Reddit user saw an estimate of USD 847 (approx. RM3,990) for the next month. Heavy users are now confronted with the inference cost of their own workflows—proof that the subsidy phase for AI assistants is ending. A USD 19 (approx. RM90) seat is easy for a finance team to approve; a tool whose spend can swing with model choice, agent duration, and token volume becomes a real budget debate. Admin caps can limit exposure, but caps also reintroduce friction. Teams must think more carefully about when Copilot should drive multi-hour autonomous sessions versus quick completions. According to Business Insider, GitHub CTO Vladimir Fedorov still does not think the company needs to raise prices much, despite the spike in usage. That restraint will be tested as more customers push the limits of the meter.
Infrastructure Strain and the Blueprint for Enterprise AI Adoption
The real warning in GitHub’s record month is not about billing; it is about pipes. As Copilot usage jumped, GitHub ran into dozens of major outages this year, prompting sharp criticism from developers such as HashiCorp cofounder Mitchell Hashimoto, who argued the platform was no longer a place for serious work if it kept blocking developers for hours. Commit activity shows the scale of the load: GitHub’s COO Kyle Daigle wrote that commits were on pace to reach 14 billion in 2026, up from 1 billion in 2025. To keep up, Microsoft confirmed GitHub is using multiple cloud providers while continuing a planned migration to Azure aimed at 2027, with reports that it is turning to a major cloud rival for extra capacity. For any founder selling enterprise AI, the lesson is explicit: usage-based pricing can unlock spend from customers who get real value, but it will expose every weak assumption in your infrastructure plan. You cannot celebrate agent adoption and then act surprised when agent workloads behave nothing like autocomplete.
