For thirty years, the most dangerous sentence in Microsoft’s history was a question about Windows. The 1998 antitrust case that nearly split the company in two was about a browser bolted onto an operating system. So when news broke that the Federal Trade Commission has opened a formal antitrust investigation into Microsoft — issuing civil investigative demands to the company’s rivals — the muscle memory in Silicon Valley was to ask which version of Windows was the problem this time.

It’s the wrong question. Microsoft barely cares about Windows anymore, and neither does the FTC. The civil demands going out to competitors aren’t about desktops. They’re about the part of Microsoft that prints money in the dark: Azure, enterprise licensing, and the way Copilot and Teams get welded onto contracts companies already can’t escape. This is an investigation into the cloud empire that grew up while everyone was still watching the Start menu.

The quiet part: civil demands go to rivals, not Microsoft

Here’s the detail that tells you how serious this is. The FTC didn’t just send Microsoft a letter. It issued civil investigative demands — CIDs — to Microsoft’s competitors. That’s the regulatory equivalent of a prosecutor interviewing witnesses before anyone gets arrested. The agency isn’t fishing. It’s building a record, and it’s building it from the testimony of the companies that have spent years complaining about how Microsoft does business.

Those complaints are not a mystery. Cloud rivals have argued for years that Microsoft makes its software dramatically more expensive to run on Amazon Web Services or Google Cloud than on Azure — that the licensing terms for Windows Server and SQL Server are structured to punish you for leaving. Security vendors have argued that bundling defensive tools into the E5 enterprise license is a way to suffocate companies like CrowdStrike without ever competing on quality. And everyone has watched Microsoft jam Teams, and now Copilot, into agreements where the customer never asked for either.

That’s the playbook the FTC is now mapping. Not a single product. A pattern.

Follow the money: bundling is the whole business model

To understand why this terrifies Redmond more than any browser case ever did, you have to understand what Microsoft actually sells in 2026. It does not sell software. It sells the impossibility of leaving.

A large enterprise that runs on Microsoft 365, Azure, Active Directory, and now Copilot is not a customer in any meaningful sense — it’s a hostage with a renewal date. Every additional product Microsoft bundles into that relationship raises the cost of escape by an order of magnitude. The genius of the model is that no single product has to win on merit. Teams didn’t beat Slack because it was better. It beat Slack because it was already there, included, pre-installed in a license the company had no choice but to buy. The European Commission already forced Microsoft to unbundle Teams for exactly this reason. The FTC is now asking whether the same logic applies to the cloud and to AI.

If the answer is yes, the threat isn’t a fine. Microsoft can pay any fine the way you pay for parking. The threat is a structural remedy — a rule that says Microsoft cannot bundle Copilot into enterprise agreements, cannot tie Azure discounts to its own software, cannot use the licensing terms that make AWS the expensive option. That doesn’t dent a quarter. It dismantles the flywheel.

The OpenAI problem in the basement

And then there’s the thing nobody at Microsoft wants the FTC to look at too closely: OpenAI. Microsoft poured roughly $13 billion into the most important AI company on earth, got preferential access to its models, and wired them into every product it ships. That arrangement has already started fraying in public — but to an antitrust regulator, a strained marriage is more interesting than a happy one, because the unwind is where the leverage lives.

The uncomfortable question for Microsoft is whether its OpenAI stake plus its Azure dominance plus its enterprise distribution adds up to something a regulator can call a monopoly on the deployment of AI — not the models themselves, but the pipes that carry them to every business in America. If the FTC decides that’s the real prize, this investigation stops being about Microsoft’s past and becomes about who controls the next decade.

Who gets hurt — and who’s quietly cheering

The obvious losers, if this goes badly for Microsoft, are Microsoft’s shareholders — though they’ve survived worse, and the stock has a way of treating antitrust drama as background noise until a judge says otherwise. The obvious winners are the companies whose lawyers are right now drafting the most damaging possible answers to those civil demands: the cloud rivals, the security vendors, the collaboration tools that lost not because they were worse but because they were unbundled.

But the real beneficiary, if the FTC actually follows through, is the enterprise buyer who has spent a decade being told that “included” is the same as “free.” It is not. The cost of every bundled product is paid in the freedom to choose a better one. An antitrust remedy that forces Microsoft to sell its products à la carte would, for the first time in years, make those products compete for the contract instead of inheriting it.

The verdict

Formal investigations are not lawsuits, and civil demands are not verdicts. The FTC could spend two years on this and walk away with a consent decree that changes a footnote. Microsoft has the best antitrust lawyers money can rent and a quarter-century of practice surviving exactly this. The smart money says it survives this one too.

But the framing matters more than the odds. The 1998 case taught Washington that you can’t regulate a tech monopoly by attacking one product, because the monopoly isn’t in the product — it’s in the bundle. The FTC appears to have finally learned that lesson. It’s not asking what Microsoft sells. It’s asking why you can never stop buying. That’s the right question. And it’s the one Microsoft has spent thirty years making sure nobody would ask.