On March 9, 2026, Microsoft announced Copilot Cowork inside the new Microsoft 365 E7 bundle, at $30 per user per month. The product is Anthropic's agentic technology, white-labeled. Ben Thompson at Stratechery read the bundle as a 65 percent premium over the prior E5 tier for AI capabilities customers can already get cheaper elsewhere, in exchange for the governance wrapper that comes with a Microsoft contract.

Nine weeks later, Tom Warren broke the other half of the story in his Notepad newsletter at The Verge. According to Warren's reporting, Microsoft is starting to cancel internal Claude Code licenses for the Experiences + Devices group, the org that ships Windows, Microsoft 365, Outlook, Teams, and Surface. The deadline to migrate to GitHub Copilot CLI is June 30. Most of the affected developers favored Claude Code.

These look like opposite calls. They are not. They are the same calculation, applied at two different layers of the stack.

Microsoft sells the substrate at a premium. Microsoft consumes the surface in-house. Both decisions reward the same owner.

What Microsoft Actually Did

Read Warren's reporting carefully. The cancellation is most Claude Code licenses in Experiences + Devices, not company-wide. Azure, the Developer Division, and AI Platform are not named in the memo as reported by Warren and corroborated by Windows Central and The Decoder. The framing line, attributed to Microsoft's Experiences + Devices VP Rajesh Jha via Warren, sets it up as benchmark-then-converge: Microsoft offered both tools to learn quickly, and is now collapsing onto Copilot CLI because it is the product "we can help shape directly with GitHub."

Microsoft opened Claude Code to employees in December 2025. Within months, thousands of staff, including non-developers in product, design, and program management, had picked it up, per Windows Central. Warren describes Microsoft as one of Anthropic's top customers via these seats. That ends on June 30 for the affected org.

What does not end is Anthropic's flow through Microsoft. Claude Opus 4.6, Claude Sonnet 4.6, and Claude Haiku 4.5 all sit inside Copilot CLI's multi-model lineup alongside GPT-5.3-Codex and Gemini 3 Pro, per the GitHub Changelog GA notice. Claude is also live across Microsoft Foundry, billed through Azure. Anthropic's $30 billion compute commitment to Azure, announced in November 2025, runs as planned. Microsoft's own Azure sales force is commissioned on Anthropic model sales the same way they are on OpenAI's, PYMNTS reported.

So the headline is wrong if you read it as "Microsoft cancels Claude Code." What Microsoft canceled is the wrapper. The substrate flows on.

Substrate Versus Surface

This is the cleanest test case yet of a frame that will define every 2026 enterprise AI procurement decision.

The model is substrate. The CLI, the IDE plug-in, the chat surface, the bundle inside the productivity suite, all of that is surface. Surfaces are where users live, where data flows, where habits form, where the bill gets sent. Substrate is interchangeable. Surface is sticky.

Microsoft has spent fifty years owning surfaces. The company started life as a BASIC interpreter business in 1975 and has been monetizing developer surfaces ever since: MS-DOS, Windows, Visual Basic, Visual Studio, C#, TypeScript, VS Code, GitHub, npm. The build-versus-buy pattern is well documented. Microsoft builds the surface and either buys or rents the runtime underneath.

Copilot Cowork is the same pattern, dressed in 2026 fashion. Anthropic built the agentic technology. Microsoft wrapped it in governance, identity, compliance, M365 integration, and a $30 per user per month subscription. We covered the AWS version of this play in our piece on how Anthropic is renting AWS's procurement layer, which makes the same structural argument from the other side of the table. Anthropic rents the procurement distribution. Microsoft rents the substrate. The wrapper is the product.

Cancelling Claude Code inside Experiences + Devices is the second move in the same playbook. Microsoft would rather have its own developers using Microsoft's wrapper, with Microsoft's billing, on Microsoft's roadmap, even if the wrapper underneath calls a Claude model. That is not a defection from Anthropic. It is a reclamation of the surface.

Microsoft owns surfaces. Microsoft consumes substrate. The Claude Code cancellation is consistent with selling Cowork at a premium, not contradictory.

What the Developers Wanted

The awkward part of the story is that Microsoft's own engineers did not pick Copilot CLI. They picked Claude Code.

The JetBrains April 2026 developer survey is the cleanest external read. Claude Code was the "most loved" agentic coding tool at 46 percent. Cursor came in at 19 percent. GitHub Copilot sat at 9 percent. Awareness of Claude Code went from 31 percent in mid-2025 to 57 percent in January 2026. Work-use adoption inside that window grew six-fold, from 3 percent to 18 percent. The product gained mindshare faster than any agentic coding tool in the survey's history.

Inside Microsoft, the pattern looked similar. Warren reports that Microsoft engineers favored Claude Code. Yahoo's headline captured the sentiment cleanly: "Microsoft is ditching Claude Code for Copilot CLI, but its own devs aren't happy." Engineers cite specific Claude Code advantages: large-codebase ingestion, multi-step refactor coherence, the ability to hold the shape of a complex change across many files without losing the plot. Copilot CLI, generally available since February 25, 2026, is faster at completion and quick-fix but still closing the gap on the harder agentic work.

Microsoft made the call anyway. That is the part to sit with. The most loved tool, on the broadest survey, with the steepest adoption curve, used internally by a workforce that prefers it, got the cancellation notice. Preference loses to procurement. Every time.

There is a reasonable rebuttal. The procurement logic is real: one bill, one security review, one MCP control plane, one set of governance levers, a vendor Microsoft can shape directly. Microsoft cannot evangelize Copilot CLI to enterprise customers while its own developers privately use a competitor. The optics matter. The roadmap matters more. None of that is wrong. It also does not soften what just happened. Developer love is not the same as developer hours.

Anthropic's Exposure, Read Honestly

The "Microsoft defection" framing misses the actual risk profile.

Claude Code hit $2.5 billion in run-rate revenue by February 2026, per MindStudio's read of Anthropic's enterprise progression. Over half of that is enterprise. Business subscriptions quadrupled in the first months of 2026, per Context Studios. The Microsoft seats inside that book are real revenue but, on the math, not the load-bearing piece. Anthropic loses a high-profile internal reference customer. It does not lose the business.

The harder number is concentration. Cursor and GitHub Copilot together account for roughly $1.2 billion of Anthropic's ~$4 billion run-rate, per VentureBeat. That is a third of revenue dependent on two wrappers Anthropic does not own. GitHub Copilot is now in the act of converting Microsoft's internal Claude Code seats onto a different wrapper while still routing some of those queries to Claude models. The substrate revenue continues. The wrapper economics shift toward Microsoft. Anthropic gets the inference dollars; Microsoft gets the seat margin.

The Ramp data from this week sharpens the picture. Anthropic now has more business customers than OpenAI, 34.4 percent to 32.3 percent, the first time Ramp has measured a crossover. Twelve months ago, Anthropic was at 9 percent. Enterprise momentum is the loud signal. Concentration in two wrappers is the quiet one. As we noted in our coverage of the OpenAI checkout pivot, the model companies are no longer fighting for the transaction. They are fighting for the surfaces around it. The Microsoft Claude Code call says the surfaces are not for sale.

The Billing Model Is the Real Tell

Read the calendar.

Copilot CLI went generally available on February 25, 2026. The cancellation deadline for Claude Code inside Experiences + Devices is June 30, 2026. One day before that, on June 1, GitHub Copilot moves to usage-based billing with monthly AI credit allotments. The fiscal year ends June 30 too, which is the easy explanation: a cost cut booked under a CFO's pen before the books close.

Take the easy explanation off the table for a moment. The billing change is the bigger fact. Microsoft is forcing convergence onto a wrapper that has been generally available for 12 weeks, on a billing model it owns, in the fiscal quarter it controls. The same window in which Anthropic ejected bundled tokens from enterprise seat deals, moving customers toward separately metered API usage and triggering enterprise pushback. Anthropic restructured its billing. Microsoft is restructuring its consumption. They are not coordinated. They are reading the same market.

Surface owners want usage-based revenue because usage growth is the story enterprise software now tells. Anthropic, OpenAI, GitHub, all of them want the meter running. The wrapper that owns the meter owns the customer. Cancelling Claude Code routes those queries through GitHub's meter instead of Anthropic's direct seat meter, even when the underlying model is still Claude. The Anthropic relationship continues. The Anthropic billing relationship does not.

That is the procurement reality every enterprise will face. We touched on it in our coverage of OpenAI's FedRAMP clearance, where the question was not which model is best but which model is procurement-ready. The Microsoft Claude Code call answers the corollary: where there is a wrapper option a buyer controls, the buyer takes the wrapper, and the model goes back to being a line item underneath.

The fiscal-year framing is real but small. The billing-model alignment is the systemic story. Microsoft is converging onto a wrapper it owns, on the billing model it owns, in the quarter it controls.

What To Watch

A few things will tell us whether this is a one-off or a template.

First, whether other large enterprises with internal Copilot bundles run the same migration. If they do, Anthropic's direct enterprise channel for Claude Code stalls and the company doubles down on wrappers it does own. If they do not, the Microsoft move reads as Microsoft-specific.

Second, whether Anthropic's Claude Code adoption curve flattens. Six-fold work-use growth in nine months is a number that will be hard to repeat. The base rate matters now, not the slope.

Third, whether Microsoft's internal engineers actually stick on Copilot CLI through June 30 or whether attrition and frustration force a rollback. Anthropic's biggest single-customer reference would re-light. Microsoft's roadmap pitch to enterprise gets harder. Neither outcome is priced into the current story.

The closing question is the one Anthropic has not had to answer in public yet. Anthropic has captured developer love. Microsoft has captured developer hours. In every quarter where those two metrics diverge, which one pays the bill?

Sources

Anthropic captured developer love. Microsoft captured developer hours. Which one pays the bill in 2027?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.