Expand menu

On September 2, Judge Leonie Brinkema of the Eastern District of Virginia closed the remedies phase of the Justice Department's ad-tech antitrust case against Google. The headline everyone ran was the predictable one: Google stays in one piece. I kept the other half, which strikes me as considerably more interesting for anyone running a software company.

Quick recap. In April 2025 the same court found that Google had illegally monopolized two markets within open-web display advertising and had illegally tied its publisher ad server to its exchange. What remained was what to do about it. The government asked for three structural remedies: divest AdX, open-source DFP's final auction logic, and, failing that, divest the rest of DFP as well. The judge rejected all three. Her reasoning, per coverage of the hearing, was that dismantling Google Ad Manager would do more damage than the conduct it was meant to correct: small publishers use DFP for free, a third-party buyer could end up owning AdX, and a divestiture opens years of appeals while the market stays exactly as it is.

So far, the familiar story: the third consecutive Big Tech monopolization finding, and the third time a court declines to split the company.

What follows is the new part. Instead of a corporate remedy, the judge imposed a package of behavioral remedies that reads like a technical requirements document:

AdX real-time bid data must be made available to rival ad servers.
Unified Pricing Rules are deprecated: a publisher can set different price floors for individual bidders.
First look and last look are banned on open-web display, with a commitment never to rebuild them.
Differential revenue sharing based on which ad server a publisher chooses is prohibited.
Google must deliver equivalent latency and data signals across all paths.

All of it under six years of monitoring. The written opinion was sealed for fourteen days for redactions, and the parties have thirty days to submit a jointly proposed final judgment. I say that because it matters: the list above comes from the hearing and from specialist coverage, not from the signed document. When it lands, the fine print will be worth reading.

Why this isn't an antitrust story

Read it again without the word remedy and tell me what it is. It's a ticket. A federal court just wrote the functional spec for an auction system that runs at planetary scale, in milliseconds, and put an audit date on it.

Latency parity is not a legal concept. It's a metric. Someone at Google has to define what gets measured, where it gets instrumented, at what percentile, and then prove to a monitor for six years that p99 down the in-house path is no better than p99 down the rival's. Real-time bid data available to third parties is an API, with a contract, versioning, quotas and an SLA. Different price floors per bidder is a data-model change in a system that today assumes a unified rule, with everything that drags upward into reporting and downward into the auction engine.

That's the difference between a remedy you sign and a remedy you deploy. The first one ends when it's signed. The second one starts there.

What I take from this as someone running a software company

Three things, and none of them are about Google.

One: compliance is moving into the backlog. For years the corporate answer to a regulation was a policy, a contract annex and a training session. Increasingly, the answer is a user story. If a regulator can demand latency parity between two routes through your system, it can demand any observable property of your product. And an observable property has to be built, tested and monitored. If legal and engineering in your company only speak to each other once there's already a problem, this is the year to fix that.

Two: observability stopped being an internal tool. Nobody builds traces and metrics with an adversarial third party in mind. But a six-year decree turns your telemetry into evidence. Systems that can't explain why they made a decision, and in which millisecond, are expensive systems to defend. This holds just as much, and probably more, for products with models in the loop: the argument being applied to an auction today will be applied to a ranking, a score or an agent router tomorrow.

Three: auction design is the product. The case is about ad tech, but the pattern is general. In any marketplace, the matching rules, who sees what, in what order, with what information and with how much of a head start, are the part of the system that creates or destroys the value. Google didn't lose for having bad software; it lost for how those rules were written. If you operate a marketplace, however small, you already have an auction policy. The question is whether you wrote it on purpose or it just ended up that way.

My read

I think the judge made the pragmatic call and that it will age poorly. A behavioral remedy asks a court to supervise, for six years, the behavior of a system that the defendant itself designs, measures and reports on. The information asymmetry is total. And the question the coverage itself left open is still unanswered: what does a publisher do if it wants a different ad server but can't walk away from Google's demand? The crudest levers were taken away. The gravity of the system didn't move.

But for us, the people who build and sell software, the conclusion holds even if the remedy fails. The state has stopped asking companies to split themselves apart. It's asking their systems to behave a certain way and to be able to prove it. That's engineering, not law. And it's a capability you build before you need it, not the day the letter arrives.