NYT Targets Microsoft Supercomputer in OpenAI Suit

In the original complaint, the Times described Microsoft systems as if they were generic cloud services. The amended version argues the machine was tailor-made. The Times alleges Microsoft designed an unusually complex supercomputer specifically to use essentially the whole internet, curated to disproportionately feature Times works, to train the most capable large language model in history.

If you publish anything online, the claim here is the one to watch: that an infrastructure provider can be held liable not for hosting data, but for designing a system whose purpose was to ingest copyrighted work without permission.

How the alleged infringement machine works

The mechanical argument matters because it is what separates a cloud landlord from an active participant.

According to the Times, the supercomputer did two jobs that go beyond renting out compute. It helped select which works were scraped, and it provided the means to seize those works without permission. The paper alleges both firms wanted to train on the highest-quality journalism available so that level of writing could be confidently mimicked in outputs, and that Times articles were therefore weighted more heavily inside the training data.

The result, the Times says, showed up in product. It alleges Microsoft deployed Times-trained models across its product line, and that doing so helped boost the company market capitalization by a trillion dollars in a single year.

The evidence the Times leans on is the model output itself. During discovery it obtained a large slice of user ChatGPT sessions. In some, users openly said they were trying to skirt paywalls and got significant chunks of articles simply by asking to see the next paragraph. In others, the Times says, models spit out several paragraphs with no prompting trick at all. The complaint includes side-by-side comparisons of ChatGPT text against original articles to argue substitution, plus screenshots of allegedly infringing outputs.

How each side is framing it

The two camps are telling opposite stories about the same filing.

• The Times calls it a strengthening of an already strong case, built on new law and new evidence uncovered in discovery. Spokesperson Graham James said the paper has long alleged Microsoft actively encouraged OpenAI to steal its copyrighted works.
• Microsoft calls it a last-ditch effort to save a claim from unfavorable precedent set in other recent rulings.

There is a tell in the procedural details. The Times voluntarily dismissed two claims, contributory copyright infringement and trademark dilution, against all defendants. That is the streamlining of a case down to its most potent arguments, and it cuts both ways: it removes weaker claims but also signals the plaintiff is consolidating around the fights it thinks it can win.

Why the amendment may be allowed

The Times argues neither company would be prejudiced by the change. Its reasoning is procedural: courts generally permit plaintiffs to revise arguments when the underlying legal standard shifts, and the Times says it is not seeking any additional discovery to support the amended claims, so the case schedule would not slip.

What happens over the next 24 to 72 hours

The immediate question is narrow and procedural: will the judge grant leave to amend? Expect Microsoft and OpenAI to oppose the motion, leaning on the argument that the new framing is a reaction to losing ground on precedent rather than a legitimate clarification.

Watch for a few things in the near term. A ruling on whether the amended complaint is accepted could come quickly given the Times insistence that no new discovery is needed. If it is accepted, the supercomputer-as-infringement-tool theory becomes a live part of the case rather than background color.

Beyond the docket, the redactions are their own story. The filing is heavily redacted, and any unsealing of the technical allegations about how the system was built and how Times content was weighted would be the most consequential development to track.

The bigger stake for AI training

Strip away the parties and this is a test of where liability sits in the AI supply chain. The Cox ruling made it harder to pin infringement on a passive intermediary. The Times response is to argue Microsoft was not passive at all, but a designer and operator of a purpose-built ingestion system.

If that theory survives, it reshapes the risk calculus for every company that builds or rents specialized AI training infrastructure. The defense will keep insisting this is a plaintiff scrambling after a bad precedent. The next ruling will tell us which read the court finds more convincing.

Source: Ars Technica