GitHub Copilot Hit 4.7 Million Paid Seats, and 46% of New Code on GitHub Is Now AI-Written
GitHub Copilot has grown to an estimated 4.7 million paid seats, up 75% year-over-year, with annual recurring revenue estimated between $0.9 and $1.1 billion. Perhaps more striking than the subscriber count: GitHub says 46% of all new code pushed to the platform is now AI-generated.
The growth is coming alongside a wave of feature expansion in August 2026: a Slack integration for agentic coding help, enterprise rollout with data residency controls, a new speed-focused model (Grok Code Fast 1) in preview, and code review now generally available for Xcode/Swift projects.
GitHub Copilot has grown to an estimated 4.7 million paid seats, up 75% year-over-year, with annual recurring revenue estimated at $0.9 to $1.1 billion. The more striking number sits alongside it: GitHub says 46% of all new code pushed to the platform is now AI-generated, up from a much smaller share just a couple of years ago.
The growth has come with a burst of feature releases through August 2026. Copilot in Slack entered public preview, bringing agentic coding help, planning changes, investigating issues, triaging bugs, drafting pull requests, directly into team chat channels, with human oversight and existing permissions preserved. Copilot's coding agent also became available in GitHub Enterprise Cloud with data residency controls, aimed at regulated industries that need to keep code and data within specific regions.
On the model side, Grok Code Fast 1 rolled out in public preview as a speed-focused option for routine completions, and Copilot code review reached general availability for Xcode and Swift projects, alongside new admin controls letting organizations manage the rollout centrally. Separately, as of August 5, Microsoft set OpenAI's GPT-5.6 Sol as the default model for its own staff's Copilot usage.
Why It Matters
Copilot's growth curve matters beyond GitHub's own business, it's one of the clearest large-scale data points on how fast AI coding assistance is actually being adopted inside real engineering teams, not just experimented with. A 75% year-over-year seat increase on a product already past 4 million users is a sign the adoption curve hasn't plateaued.
The 46% AI-generated code figure is the number worth sitting with longest. If accurate and sustained, it means AI-assisted coding has crossed from "a tool some developers use" to "close to half of all new code," a shift with real implications for how software gets reviewed, tested, and maintained going forward.
Key Details
- Company: GitHub (Microsoft)
- Product: GitHub Copilot
- Paid seats: 4.7 million, up 75% year-over-year
- Estimated ARR: $0.9-1.1 billion
- Share of new code that is AI-generated: 46%, per GitHub's own data
- New in August 2026: Copilot in Slack (public preview), Enterprise Cloud data residency, Grok Code Fast 1 model preview, Xcode/Swift code review (GA)
- Default model for Microsoft staff: OpenAI GPT-5.6 Sol, as of August 5, 2026
Technical Analysis
The feature list this month points to two different growth strategies running in parallel. One is going deeper into existing developer workflows: sub-issue creation, Xcode code review, data residency for regulated enterprises. The other is going wider, into spaces where code discussion already happens but Copilot hasn't lived before, like Slack, where the agent can now triage bugs and draft pull requests inside a team channel rather than requiring a dedicated IDE session.
The Grok Code Fast 1 preview is worth watching separately: it's explicitly positioned as a speed-optimized alternative rather than a capability upgrade, suggesting GitHub is hearing demand for faster, cheaper completions for routine tasks, not just smarter ones for hard problems.
Competitive Comparison
| Product | Better At | Weakness |
|---|---|---|
| GitHub Copilot | Deepest GitHub/IDE integration, largest installed base, multi-model choice (GPT, Grok options) | Tied to GitHub's ecosystem; less useful for teams on other code hosts |
| Cursor / other AI-native IDEs | Built ground-up around AI-assisted editing, often faster iteration on agentic features | Smaller installed base, less enterprise compliance tooling than Copilot Enterprise |
| Standalone model APIs (direct Claude/GPT use) | Maximum flexibility, no platform lock-in | Requires teams to build their own tooling around the raw model, no packaged IDE/PR/review integration |
Industry Impact
Developers get faster boilerplate and routine-task completion, but the 46% figure also raises the practical question of who's reviewing that much AI-written code carefully versus rubber-stamping it.
Engineering managers and enterprises now have a mainstream, board-level number (46% of new code) to reference when setting AI-usage policy, code review standards, and security review processes.
Competing AI coding tools face a market where the default, bundled-with-your-code-host option has crossed into near-majority usage, raising the bar for what a standalone competitor needs to offer to win developers away.
Future Outlook
The following is analysis and prediction, not confirmed fact.
If AI-generated code continues climbing toward and past 50% of new commits, expect the conversation to shift from "should teams adopt AI coding tools" to "how do code review and testing practices need to change now that they're the default," a genuinely different and harder problem than adoption itself.
The Slack integration also hints at where agentic coding tools are heading generally: less a separate destination developers visit, more an agent embedded in the tools teams already use for communication and planning.
Key Takeaways
- GitHub Copilot has reached an estimated 4.7 million paid seats, up 75% year-over-year, with $0.9-1.1B in estimated annual recurring revenue.
- 46% of all new code pushed to GitHub is now AI-generated, according to GitHub's own data.
- August 2026 additions include a Slack integration, Enterprise Cloud data residency, a speed-focused Grok Code Fast 1 model preview, and general availability of Xcode code review.
- Microsoft set OpenAI's GPT-5.6 Sol as the default model for its own staff's Copilot usage as of August 5, 2026.
Source: AI Business
