Cursor vs GitHub Copilot Market Share (2026 Survey)

Quick Answer (TL;DR)

GitHub Copilot reports 50 million total users but only 4.7 million paid seats. Cursor reached $4 billion annualized revenue and was bought by SpaceX. Claude Code now leads actual use at work at 39% of developers. Those three numbers are measured on different denominators, which is why most comparisons of this market are wrong.

Every figure on this page carries the date it was stated and who stated it. Where a number has not been disclosed, this page says so instead of estimating. See the changelog for what changed and when.

The short version, as of September 2026:

• GitHub Copilot has the largest installed base. Satya Nadella reported 50M total users on the FY26 Q4 earnings call (29 July 2026). The last paid figure Microsoft disclosed was 4.7M subscribers, up 75% year over year, on the FY26 Q2 call (28 January 2026). No newer paid number exists.
• Cursor has the highest disclosed revenue in the category, at roughly $4B annualized as of early June 2026. SpaceX closed its $60B all-stock acquisition on 14 August 2026.
• Claude Code leads on what developers actually reach for. In JetBrains research covering more than 15,000 professional developers fielded May to July 2026, 39% used Claude Code at work (47% in the US) against 21% for Copilot and 12% for Cursor.

Most teams stack tools rather than choosing one. In a Pragmatic Engineer survey of 906 developers fielded January to February 2026, 70% used two to four AI coding tools and 15% used five or more.

Cursor vs GitHub Copilot Market Share at a Glance

The "used at work" row is the only one measured the same way across all three, from JetBrains research fielded May to July 2026. The user and revenue rows come from different companies disclosing different things at different times, so they cannot be ranked against each other directly.

Note what is missing. Microsoft has never disclosed Copilot revenue. Neither Anysphere nor Anthropic discloses user counts for Cursor or Claude Code. Any table that fills every cell for this market is estimating, and usually not saying so.

Market size and growth

Independent research firms put the 2026 AI coding tools market at roughly $9.4B to $10.1B, with estimates from Mordor, Grand View, Precedence and Research and Markets clustering in that band. Global Market Insights projects $30.1B by 2032, at a 23.2% CAGR measured from a 2023 base of $4.8B in a report published September 2024.

The best measured figure, as opposed to projected, comes from Menlo Ventures. Surveying roughly 495 US enterprise AI decision makers in November 2025, it put enterprise spend on coding tools at $4.0B in 2025, up from $550M in 2024. In that same research, Anthropic took about 54% of enterprise coding spend against OpenAI's 21%. That is share of model-layer dollars spent on coding, not share of seats or tool revenue, and the two get conflated constantly.

Adoption is now close to saturation among professional developers. JetBrains found 90% use AI coding agents at work at least weekly and 68% daily (May to July 2026). Stack Overflow's 2025 survey, still the most recent edition, found 84% using or planning to use AI, but also that 46% actively distrust its accuracy against 33% who trust it and only 3% who highly trust it. High use and low trust are both true.

For a deeper look at AI tools across the broader software development lifecycle, see Best AI SDLC Tools 2026.

Market share by tool (2026)

GitHub Copilot: 50M users, 4.7M disclosed paid seats

The gap between those two numbers is the most misread thing in this market. 50M is total users, reported July 2026. 4.7M is paid subscribers, reported January 2026, and Microsoft has not updated it since. Comparisons that put Copilot's 50M against a rival's revenue are comparing a free-inclusive user count to a paying customer base.

What Microsoft has disclosed: nearly 140,000 organizations using Copilot with enterprise subscribers close to tripling year over year (April 2026), Copilot revenue accelerating over 60% quarter over quarter (July 2026), and one in three pull requests on GitHub now involving an agent (July 2026). Copilot revenue itself has never been broken out.

Pricing changed twice in 2026 and the direction matters. On 1 June all plans moved to usage-based billing, replacing premium requests with token-metered AI credits. On 1 September promotional Business and Enterprise credit allowances expired, cutting included usage by 37% to 44% at unchanged seat prices. A new Max tier at $100 per month now sits above Pro ($10) and Pro+ ($39).

For the backstory on how GitHub Copilot reached mainstream developer adoption, the distribution playbook is as instructive as the product.

Cursor: $4B annualized, now owned by SpaceX

Cursor's revenue curve is the steepest publicly reported in the category: $100M in January 2025, $500M by June, $1B by November, $2B in February 2026, $3B in late April, and roughly $4B annualized by early June 2026. Those figures come from Forbes and Dealroom reporting sourced to people familiar with the company, not from Anysphere directly.

Cursor's own enterprise page claims 64% of the Fortune 500 and more than 50,000 enterprises, without published methodology. Its paying-user count has never been disclosed by the company or credibly reported, so this page does not state one.

SpaceX announced the acquisition on 16 June 2026 and closed it on 14 August in an all-stock deal valued at $60B. That follows SpaceX's acquisition of xAI in February 2026. Cursor's pricing page now advertises Grok access on the $20 Individual tier, which is the clearest signal of what the combination is for.

On models: Composer 2 shipped 18 March 2026 built on Moonshot AI's Kimi K2.5, a detail Cursor initially omitted and later conceded was a miss. Composer 2.5 followed on 18 May 2026.

Claude Code: first on usage, last to disclose numbers

Anthropic stated Claude Code was at over $2.5B run rate on 12 February 2026, more than doubled since the start of that year, with enterprise representing over half of Claude Code revenue and 4% of all public GitHub commits authored by it. Weekly active users had doubled since 1 January, though Anthropic has never published an absolute user number.

There has been no company-stated Claude Code revenue breakout since February 2026. Figures circulating for May and August trace only to aggregators and a forecasting site, so they are not repeated here.

The usage picture is clearer than the revenue picture. JetBrains put Claude Code at 39% at-work use (47% in the US) in May to July 2026, up from 18% in its January wave and about 3% in mid-2025.

The widely quoted "46% most-loved" figure needs care. It comes from The Pragmatic Engineer's survey of 906 developers published 3 March 2026, and it is share of mentions in an open-ended free-text question, not a most-loved poll: 46% of respondents named Claude Code unprompted, against 19% for Cursor and 9% for Copilot. That measures salience among a self-selected, senior-skewing newsletter audience. It is a real signal and it is not a satisfaction score.

For scale, Anthropic's overall run rate went from $14B in February to $47B in May to $65B by the end of July 2026, with Claude Code an acknowledged driver.

The long tail, and what left it

Three of the four tools this section used to list are gone or renamed:

• Tabnine was acquired by Tricentis on 30 July 2026, at a price estimated in the tens of millions.
• Amazon Q Developer is being retired. New signups closed 15 May 2026 and IDE plugins and paid subscriptions reach end of support on 30 April 2027. AWS is pointing users to Kiro.
• Windsurf, formerly Codeium, became Devin Desktop on 2 June 2026 under Cognition, which reported a $900M annualized run rate and a $48B valuation in September 2026.
• Google is winding Gemini Code Assist down rather than repricing it. Gemini CLI and the individual, Pro and Ultra tiers stopped serving requests on 18 June 2026, and new console purchases were blocked from 4 September 2026. The successor is Antigravity, which Google reports at 2.4 million weekly active users.
• Cline passes 5.3M VS Code Marketplace installs and is actively developed. Its "11 million users" claim is derived from install counts across two marketplaces, not distinct people. Aider has not had a code push since 22 May 2026, with 493 pull requests unmerged. The repository is not archived and no maintainer has said anything, so treat it as dormant rather than dead.
• Replit reports 60 million total registered users (September 2026). The last hard revenue figure it published was $150M annualized, in September 2025.

This section used to claim the long tail was "roughly 15 to 20% of the market by usage." That has been removed because it cannot be supported, and the reason is worth stating plainly: no survey with a published denominator breaks out any of these tools. The two JetBrains studies and Stack Overflow's are the only ones with credible methodology, and none of them measures Devin Desktop, Amazon Q Developer, Kiro, Replit, Tabnine, Cline or Aider. Every per-tool market share percentage circulating for these products traces back to listicles rather than research.

Where a number does not exist, the useful thing a reference page can do is say so.

How developers actually use AI coding tools

The defining behavior is tool stacking. In The Pragmatic Engineer's survey of 906 developers, 70% used two to four AI coding tools and 15% used five or more. That survey is a self-selected newsletter audience skewed senior, so read it as directional rather than representative.

The implication for product builders: the assumption that one tool wins the developer is dead. Developers route different problems to different tools. Build for that reality, not the older model of one IDE and one assistant.

A caution on all survey data in this market. JetBrains surveys its own user base, which skews toward JetBrains IDE users. The Pragmatic Engineer surveys its subscribers. Neither is a random sample of developers, and the two disagree on ordering in ways that sampling alone probably explains.

For PMs evaluating which AI coding tools their team should adopt, the AI tools SDLC guide covers the full stack including testing, code review, and deployment tooling.

Enterprise vs startup adoption

This page previously published adoption rates by company-size segment. Those figures could not be traced to any source and have been removed rather than restated.

What is sourced: Anthropic reported enterprise as over half of Claude Code revenue (February 2026). Microsoft reported nearly 140,000 organizations on Copilot with enterprise subscribers close to tripling year over year (April 2026). Cursor reported roughly 75% of its run rate as enterprise, about $2.6B annualized (June 2026).

The pattern those three disclosures share is that every vendor in this market is moving upmarket at once, and none of them publishes a segment breakdown that can be compared against the others.

Five product opportunities in this market

The market is growing quickly but specific gaps remain unaddressed. Five opportunity areas for builders:

1. AI code review bots specialized by framework. Generic code review works, but framework-specific review consistently outperforms on quality. Whitespace exists per major framework.
2. AI-powered technical debt detection. Most tools generate new code; few systematically identify and quantify existing debt. The buyer is the engineering manager, not the IC.
3. Code migration tools for legacy-to-modern stack moves. Java to Kotlin, Angular 1 to React, Python 2 to 3. Each migration is a multi-million-dollar enterprise project today.
4. AI test generation for untested codebases. Coverage tools exist; AI-driven coverage that writes meaningful tests for legacy code is still early.
5. Review capacity tooling. DORA measured median pull request review time rising about 91% as AI adoption rose. Generation got cheap and review did not, so the bottleneck moved. Almost every tool in this market still sells generation.

For more context on how these category opportunities map to broader trends, see the AI Coding Assistants trend page.

Where the market goes next

1. The denominator war gets worse before it gets better. OpenAI stopped breaking out Codex from ChatGPT Work after July 2026, reporting a combined agent number without defining the activity window. Expect more vendors to report the flattering aggregate.

2. Consolidation continues. Tabnine, Windsurf and Cursor all changed hands, and Google and AWS both retired a product line, inside twelve months. Cognition raising at $48B on a $900M run rate suggests investors do not read this as winner-take-all.

3. Pricing moves from seats to metered usage. Copilot, Kiro and Claude all now meter. The per-seat comparison that made this market easy to analyze is ending, and procurement has not caught up.

For PMs building in this space, the AI ROI calculator and AI build vs buy decision tool are useful for evaluating whether to build, partner, or buy AI tooling for your team.

Changelog

• 2026-09-13 Full source audit. Corrected the 46/19/9 figures, which were attributed here to a JetBrains April 2026 survey that contains no such ranking. They come from The Pragmatic Engineer, n=906, fielded January to February 2026, and they measure share of mentions in an open-ended question rather than a most-loved poll. JetBrains' own April figures were Copilot 29%, Cursor 18%, Claude Code 18% on a use-at-work metric. Removed a "generates 46% of all code" claim sourced to a GitHub post from February 2023. Replaced the $12.8B market size and 27% CAGR, neither of which traced to a source, with the $9.4B to $10.1B independent range and Menlo's measured $4.0B of 2025 enterprise spend. Updated Copilot to 50M users on 4.7M disclosed paid seats, Cursor to ~$4B annualized and the closed SpaceX acquisition, and Claude Code to first place on at-work usage at 39%. Removed unsourced segment adoption rates. Marked Tabnine as acquired, Amazon Q Developer as retiring, and Windsurf as renamed to Devin Desktop.
• 2026-04-29 First published.

Related reading: AI Code Review Tools Market Share 2026: Data + Trends.