Who is winning the AI coding market: Cursor, Claude Code, or Copilot?

Who is actually winning the AI coding market right now?

Claude Code is winning the AI coding market today if we judge by which specialist tool professional developers are actually choosing for serious coding work.

That answer changes depending on what we measure, which is why the race looks confusing from the outside. GitHub Copilot still has vastly more distribution. Cursor has the strongest publicly reported standalone revenue number. Claude Code, meanwhile, has pulled ahead in recent professional-developer usage.

The latest large JetBrains Developer Ecosystem Survey makes the split unusually clear. Based on more than 15,000 professional developers worldwide surveyed from May through July 2026, around 39% were using Claude Code at work. GitHub Copilot was at 21% and Cursor at 12%. Claude Code was also the most-used AI coding tool for 31% of respondents.

Commercially, Cursor tells a different story. Forbes reported that Cursor passed $4 billion in annualized revenue in early June after reaching $2 billion only four months earlier. Roughly three-quarters of that business was coming from companies.

Copilot dominates a third scoreboard. Microsoft said in its latest annual earnings call that GitHub Copilot had reached 50 million users, while GitHub itself had 225 million users.

So we get three different leaders depending on the question. Claude Code has the strongest current developer pull. Cursor has the strongest disclosed standalone monetization. Copilot has a distribution advantage nobody else comes close to matching.

If we have to pick the company winning the market rather than one particular metric, we put Claude Code first for now.

What we measure	Leader today	Best recent evidence
Professional developer adoption	Claude Code	~39% workplace usage
Primary-tool preference	Claude Code	~31% call it their most-used tool
Standalone coding-tool revenue	Cursor	>$4B annualized revenue
Overall user distribution	GitHub Copilot	50M users
Repository distribution	GitHub Copilot	GitHub has 225M users

If you want more recent data on this point, please see our latest AI code assistant market report.

Why did Claude Code suddenly overtake Cursor and GitHub Copilot?

Claude Code has grown so quickly because developers are moving from AI-assisted coding toward giving agents entire engineering tasks, and Claude Code fits that behavior extremely well.

The speed of the change is unusual. In JetBrains’ January 2026 survey, Copilot was still being used at work by 29% of professional developers. Cursor and Claude Code were tied at roughly 18%. By the May–July survey, Claude Code had jumped to around 39%, while Copilot had slipped to 21% and Cursor to 12%.

The wider coding market changed at the same time. JetBrains found that 90% of professional developers were using coding agents at least weekly and 68% were using them every day. Developers estimated that roughly 47% of the code they produced was now fully generated by agents. About one in five said they wrote no work code without AI help at all.

That environment rewards products designed around delegation. Claude Code can inspect a repository, modify several files, run commands, execute tests, investigate failures and continue iterating. The developer increasingly gives it an outcome rather than asking for the next few lines of code.

Anthropic’s own usage research supports that interpretation. Its analysis of roughly 400,000 Claude Code sessions found that debugging’s share of activity almost halved between late 2025 and spring 2026. Running software, analyzing data and handling broader end-to-end work all became more common. Anthropic estimated that the economic value of the typical task users gave Claude Code rose by roughly 25% during the period studied.

Are professional developers really using Claude Code more than Cursor and Copilot?

Yes. Claude Code currently has a substantial lead over both Cursor and GitHub Copilot in the strongest recent global survey of professional developers we found.

JetBrains surveyed more than 15,000 professional developers across multiple regions and languages and adjusted the sample to better reflect the global developer population. Roughly 39% said they used Claude Code at work. Copilot came in at approximately 21%, with Cursor around 12%.

The US gap was even larger. Claude Code reached roughly 47% workplace adoption there, meaning almost half of professional developers surveyed were using it.

These figures need one important qualification: they are adoption rates rather than mutually exclusive market shares. A developer can use Claude Code, Cursor and Copilot during the same week.

That gives us another useful measurement: which tool gets used most.

Claude Code was the primary AI coding tool for about 31% of developers. Compared with the 39% who used it at all, nearly four out of five Claude Code adopters were treating it as their main coding tool.

Cursor does not currently show that level of broad professional adoption. Neither does Copilot.

Is Claude Code becoming developers’ main coding tool or just something they use occasionally?

Claude Code is increasingly becoming the main AI tool developers reach for, which is more important than simply appearing somewhere in a crowded developer toolkit.

Developers rarely use only one AI product anymore. Someone might keep Copilot active for inline completion, open Cursor for one repository and run Claude Code in a terminal for a larger task. Counting every installed or occasionally used product therefore tells us less than it used to.

The more interesting number is the relationship between Claude Code adoption and primary usage. Around 39% of professional developers in the latest JetBrains survey used Claude Code for work, while 31% named it their most-used AI coding product.

That is an unusually high conversion from trying or using a product to making it the default.

We can also see the depth of the behavior in Anthropic’s own session analysis. Claude Code users were running the tool for an average of about 20 hours a week. In a typical session, Anthropic found that humans made most of the planning decisions while Claude made most of the execution decisions.

The broad adoption numbers already favor Claude Code. The usage pattern makes that lead more convincing.

Has GitHub Copilot lost its lead?

GitHub Copilot has lost its old lead among actively chosen AI coding tools, although Microsoft still controls the biggest distribution channel in the market by a huge margin.

Copilot was the product that turned AI coding into a mainstream software category. Its original advantage was almost impossible to miss: install it inside a developer’s existing workflow and offer useful completions without asking the developer to change how they worked.

The recent developer data shows that advantage fading. JetBrains measured Copilot workplace adoption at roughly 29% at the beginning of 2026 and around 21% several months later, while agent-first products gained ground.

Microsoft’s own numbers are much stronger when we look at reach rather than preference. The company said in its latest annual results that GitHub Copilot now has 50 million users. GitHub has 225 million users overall, and more than 90% of the Fortune 500 use the platform.

Microsoft also said Copilot revenue accelerated by more than 60% quarter over quarter after GitHub introduced more usage-based billing. Earlier in the year, Microsoft had reported 4.7 million paid Copilot subscribers, up 75% year over year.

Copilot’s business has hardly collapsed. What changed is its place in the developer’s mind.

If you want more recent data on this point, please see our latest AI code assistant market report.

Is Cursor really losing if its revenue is still exploding?

Cursor is losing some broad developer share right now, but its commercial performance is far too strong to call the company a loser.

This is probably the strangest contradiction in the entire AI coding market.

Cursor’s awareness among developers increased from roughly 69% to 75% between JetBrains’ January and later 2026 surveys, yet workplace adoption fell from about 18% to 12%. People increasingly know the product, but a smaller percentage of the overall professional developer population reports using it at work.

At almost exactly the same time, Cursor’s revenue went vertical.

Forbes reported that annualized revenue reached $2 billion in February, $3 billion in late April and more than $4 billion in early June. Cursor therefore added roughly $2 billion of annualized revenue in about four months. Around 75% of the run rate reportedly came from business customers.

Large enterprise deployments help explain how those two trends can coexist.

NVIDIA says more than 30,000 developers use Cursor daily and that developers using it commit more than three times as much code. Coinbase says more than 2,400 developers use Cursor regularly; 75% of its pull requests are now created by agents, and the average engineer has increased merged pull requests by 55% since the beginning of the year. Box says more than 85% of its engineers use Cursor daily, with roadmap throughput improving by 30% to 50%.

Cursor increasingly looks like an enterprise-heavy power-user business rather than the universal developer default it once appeared capable of becoming.

Who is making more money from AI coding: Cursor, Claude Code or Copilot?

Cursor has the largest clean standalone coding-product revenue figure we can verify today, although Claude Code is growing fast enough that the gap may already be narrower than the disclosed numbers suggest.

Cursor exceeded $4 billion in annualized revenue according to Forbes, up from $2 billion only a few months earlier.

Anthropic officially said earlier in 2026 that Claude Code had passed $2.5 billion in run-rate revenue. That figure had more than doubled since the start of the year. Weekly active Claude Code users had also doubled, business subscriptions had quadrupled, and enterprise customers represented more than half of Claude Code revenue.

Those Anthropic numbers are older than Cursor’s latest revenue milestone, so comparing $4 billion with $2.5 billion as though they were measured on the same day would be misleading. Claude Code has continued gaining users since Anthropic published the figure.

Microsoft gives us even less precision. Copilot has tens of millions of users and millions of paid subscribers, and Microsoft says revenue has recently accelerated sharply, but it does not publish a standalone annual revenue number comparable with Cursor’s.

For that reason, Cursor wins the revenue comparison on disclosed evidence.

Product	Best disclosed standalone revenue marker	What we can safely say
Cursor	>$4B annualized revenue	Largest recent clean product figure
Claude Code	>$2.5B run rate	Older figure, still growing very quickly
GitHub Copilot	Not separately disclosed	Large and accelerating, but impossible to rank precisely

Who is winning enterprise AI coding?

Cursor currently has the strongest evidence of deep company-wide deployment, even though Claude Code is catching up quickly and Copilot remains easier to roll out inside many existing GitHub customers.

Cursor says more than 50,000 enterprises use the product and 64% of Fortune 500 companies have adopted it. Those are company-reported figures, so the named deployments are more useful than the headline percentage.

NVIDIA has more than 30,000 daily Cursor users. Coinbase has more than 2,400. Box reports more than 800 daily users and 85% engineering adoption. Cursor has also published deployments at Stripe, Salesforce, PayPal, Rippling and other large companies.

The important part is how deeply some of these customers are changing their engineering process.

Coinbase says certain teams cut the time from idea to production from roughly 20 days to less than two. Box says major migrations now take 80% to 90% less effort. NVIDIA has pushed Cursor across code generation, testing, debugging and deployment rather than restricting it to autocomplete.

Claude Code has a strong counterargument. Anthropic says business subscriptions quadrupled from the beginning of the year and enterprise customers already represent more than half of Claude Code revenue. Companies are clearly moving it from individual developer subscriptions into centrally managed deployments.

Copilot has another advantage entirely. If a large company already runs its repositories, permissions and engineering processes through GitHub Enterprise, adding Copilot creates far less organizational friction than changing the team’s main development environment.

Still, when we look specifically for evidence that companies have reorganized engineering work around one AI coding product, Cursor has the deepest public case studies today.

Why are developers choosing Claude Code for harder coding work?

Claude Code has become especially popular for difficult coding jobs because it behaves more like an engineer executing a task than an autocomplete system waiting for the next prompt.

Anthropic’s research gives us a rare look at what people actually do with the product. It analyzed around 400,000 Claude Code sessions from approximately 235,000 users.

Debugging accounted for about one-third of sessions early in the observation period. Several months later, it had fallen to roughly one-fifth. Meanwhile, running and operating software increased from about 14% to 21%, while writing and data-analysis work expanded substantially.

The typical task also became more valuable. Using comparable freelance job postings as a proxy, Anthropic estimated that the economic value of Claude Code tasks rose by roughly a quarter during the seven months studied.

JetBrains found another interesting pattern. Around 32% of developers who named Claude Code their primary coding tool said agents generated more than 80% of their code.

That does not mean Claude Code is independently doing 80% of software engineering. Writing code is only one part of engineering, and humans still decide architecture, requirements, reviews and production trade-offs.

It does show how far user expectations have moved. Advanced developers increasingly judge an agent on whether they can hand it a messy task and come back to working code.

If you want more recent data on this point, please see our latest AI code assistant market report.

Is Cursor still the best AI-native code editor?

Cursor still looks like the strongest dedicated AI-native editor, even while Claude Code leads the broader coding-agent race.

Cursor’s strength comes from controlling the environment around the model. The company can combine repository context, editing, agents, review, model routing, organizational rules and cloud execution in one place.

That gives Cursor more surfaces on which to improve the experience than a simple terminal agent has.

The product has also moved well beyond its early reputation as “VS Code with good AI.” Cursor now offers background agents, automated pull-request review through Bugbot, enterprise model controls, team-level rules and its own Composer model family.

Large customers are using those capabilities as an engineering platform. Coinbase describes Cursor as a foundation for an “agent-first” development model. Box calls it its core IDE and agent platform. NVIDIA has extended Cursor into several parts of its software-development lifecycle.

Claude Code takes a lighter approach. Developers can run it from a terminal and keep using their preferred editor around it. That portability is part of the appeal.

Someone who wants AI deeply woven into an editor may still prefer Cursor. Someone comfortable delegating work through an agent and checking the results elsewhere may find Claude Code simpler.

Does GitHub give Copilot an advantage that Claude Code and Cursor cannot copy?

GitHub gives Copilot a structural advantage neither Anthropic nor Cursor can realistically recreate: Microsoft already controls where a huge amount of the world’s software is stored, reviewed and merged.

Microsoft says GitHub now has 225 million users, including more than 90% of the Fortune 500.

That position becomes more valuable as coding agents become autonomous. An agent needs access to repositories, branches, issues, tests, deployment workflows, security policies and human approvals. GitHub already sits in the middle of all of those things for millions of teams.

Microsoft says one in three GitHub pull requests now involves an agent. It also says every major coding agent runs on GitHub.

That last point reveals both the opportunity and the problem for Copilot.

GitHub can benefit even when Claude Code or Cursor creates the code. The work still frequently ends up in a GitHub repository. Microsoft therefore has a way to remain strategically important even when developers prefer somebody else’s coding agent.

But GitHub being the destination does not automatically make Copilot the preferred agent.

Did SpaceX buying Cursor make Cursor stronger?

The SpaceX acquisition makes Cursor much stronger in compute and financing, while creating new problems around its neutrality as a multi-model platform.

SpaceX completed the acquisition of Cursor parent Anysphere at an implied $60 billion equity value. Cursor is now a wholly owned SpaceX subsidiary.

The obvious benefit is resources. Cursor says access to SpaceXAI infrastructure gives it much more compute for training and serving its own models. That could lower its dependence on companies such as Anthropic and OpenAI while improving the economics of heavy agent usage.

The timing matters because one of Cursor’s biggest vulnerabilities has already become real.

OpenAI has decided to wind down its relationship with Cursor following the SpaceX acquisition. Wired reported recently that OpenAI expected the Cursor partnership to generate more than $1 billion of annual revenue, yet still chose to end it because of concerns tied to Elon Musk and SpaceX.

Anthropic, by contrast, is continuing to supply Claude models to Cursor.

That leaves Cursor in an unusual position. One major model provider has become a direct competitor and supplier that no longer wants the relationship. Another major competitor is still supplying Cursor. SpaceX now gives Cursor the capital and infrastructure to reduce that exposure by developing more models internally.

The acquisition makes Cursor’s own model development much more important than it was a few months ago.

Does Cursor have a real moat, or can Claude Code copy it?

Cursor has a real enterprise and workflow moat, although most individual product features can eventually be copied.

A feature checklist understates what Cursor has built. Companies with thousands of developers have configured access rules, model permissions, security policies, budgets, coding conventions and internal workflows around the product.

Cursor recently introduced organization-level management so large customers can control separate business units, budgets, models and sandboxes from one place. That kind of administrative layer sounds boring compared with a new agent demo, but it becomes important once an AI coding product spreads through a multinational company.

The customer evidence is harder to replicate quickly as well. NVIDIA, Coinbase and Box have already built substantial workflows around Cursor. Switching several thousand engineers to another product would require more than proving that another agent writes code slightly better on a benchmark.

Cursor is also trying to own more of its model stack. SpaceX gives it far more compute to pursue that strategy, while the OpenAI breakup gives it a very obvious reason to do so.

Cursor’s moat looks meaningful but permeable. It buys the company time; it does not guarantee permanent leadership.

If you want more recent data on this point, please see our latest AI code assistant market report.

Who controls the most valuable part of AI coding: Anthropic, Cursor or Microsoft?

Microsoft controls the broadest strategic layer through GitHub, Anthropic currently controls the strongest developer-facing agent, and Cursor has built the most valuable independent AI-native development environment.

Those are different businesses, and they can all make a lot of money at the same time.

Anthropic owns both Claude and Claude Code. When a developer uses its first-party agent, Anthropic captures the model usage and the application relationship. That vertical integration can become economically powerful if Claude continues to rank among the best coding models.

There is one catch: Anthropic also sells Claude intelligence to competitors. Cursor supports Claude alongside models from several other providers, and GitHub Copilot also offers Anthropic models. Developers can therefore benefit from Claude’s coding ability without using Claude Code itself.

That arrangement lets Anthropic make money twice when Claude is preferred: directly through Claude Code and indirectly through API usage inside competing products. But it weakens the exclusivity of Claude Code’s model advantage.

Cursor owns the environment around the models. It can decide how context gets assembled, how agents interact with repositories, how teams manage rules and budgets, and which model should handle a particular request. This lets Cursor benefit from improvements produced by several model companies rather than betting everything on one model family.

Microsoft owns GitHub, the layer where much of the resulting software is stored, reviewed, secured and merged. That makes GitHub valuable regardless of which agent generated the code.

The economics differ as well. Anthropic can keep more of the stack when Claude Code uses Claude. Cursor historically had to pay external model providers heavily, although its investment in proprietary models could change that. Microsoft can combine seat subscriptions, usage charges, GitHub Enterprise and broader cloud relationships.

Anthropic’s challenge is therefore bigger than keeping Claude technically strong. Claude Code has to remain the preferred first-party environment even when the same underlying models are available inside competing products.

What would Copilot have to do to become number one again?

GitHub Copilot can become number one again if Microsoft turns GitHub’s huge installed base into a genuinely better agent experience rather than relying on distribution alone.

The opportunity is enormous. Copilot already has 50 million users, and GitHub has 225 million. Microsoft does not need to convince developers to move their repositories or adopt an unfamiliar company.

It mainly needs to stop developers from leaving the GitHub ecosystem when they have difficult work to do.

Microsoft is clearly moving in that direction. GitHub has shifted its product language toward agent-first and multi-agent development. It has introduced consumption-based billing for heavier agent workloads and increasingly treats coding agents as workers that can open pull requests and operate against repositories rather than simply suggest text inside an editor.

The company also has the freedom to offer several frontier models. Copilot does not need a Microsoft model to beat Claude on every benchmark if it can combine strong outside models with superior GitHub context and workflow integration.

The weak point today is preference. Claude Code’s recent professional usage shows developers voluntarily choosing another product even though Copilot often comes with the platform they already use.

What would Cursor have to do to beat Claude Code again?

Cursor can overtake Claude Code if it turns its huge enterprise business into the best place to manage many coding agents while making its own models strong enough to reduce dependence on rivals.

Cursor already has most of the difficult ingredients: billions of dollars in annualized revenue, large enterprise deployments, a widely known developer brand and control of an AI-native editor.

The next step is harder.

As coding becomes more autonomous, developers may spend less time manually editing files. Cursor therefore has to make sure the value of its environment survives even when agents produce most of the code.

Its recent products point in that direction. Cursor is expanding background agents, automated code review, cloud execution, organizational management and proprietary models. The company increasingly talks about developers orchestrating agents rather than typing beside an autocomplete tool.

SpaceX adds huge resources to that effort. Losing OpenAI as a model supplier adds urgency.

Cursor could become the neutral operating system for software agents even if Claude remains one of the best individual models.

So who is winning: Cursor, Claude Code or GitHub Copilot?

Claude Code is winning the AI coding market right now because professional developers are choosing it more often for the kind of agentic work that increasingly defines the category.

Our final ranking is Claude Code first, Cursor second and GitHub Copilot third, but the gap between those positions is very different depending on what we measure.

Claude Code has the clearest current advantage in developer preference. The strongest recent global professional-developer research puts it far ahead of Cursor and Copilot in workplace adoption, while an unusually large share of its users also call it their primary coding tool. Anthropic’s own behavioral research shows people giving Claude Code longer, broader and more valuable tasks over time.

Cursor deserves second place, and not by much. Its reported annualized revenue doubled from $2 billion to more than $4 billion in roughly four months, and companies such as NVIDIA, Coinbase and Box are using it deeply enough to redesign parts of their engineering workflow. The market-share surveys probably understate Cursor’s economic position because some of its strongest customers are extremely large and intensive users.

Its risk has also become clearer lately. Cursor is now owned by SpaceX, OpenAI is ending a supplier relationship reportedly worth more than $1 billion a year to OpenAI, and the company increasingly needs its own models if it wants to remain independent from competitors that also provide the intelligence inside its product.

Copilot comes third on current product pull but remains the easiest company to imagine retaking first place. Fifty million Copilot users and 225 million GitHub users are an extraordinary distribution base. Microsoft also owns the repository and collaboration layer where work from Claude Code, Cursor and almost every other coding agent frequently ends up.

AI coding is moving away from helping developers type faster and toward agents that can take a meaningful engineering objective and execute much of it themselves.

Claude Code is currently the product developers are choosing most aggressively for that new way of working.

Cursor has built the strongest independent commercial challenger, while Copilot still has the distribution to reverse the ranking if GitHub turns that reach into stronger agent preference.

Today, though, Claude Code is the one setting the pace.

If you want more recent data on this point, please see our latest AI code assistant market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.