The NYT Says Microsoft Built a Copyright-Infringement Machine

The Gray Lady isn't just suing OpenAI for scraping articles. The new claim is that Microsoft built a bespoke supercomputer specifically to steal them. The trillion-dollar question is whether the courts will agree.

The idea that language models are trained on copyrighted text is old news. The legal fights have been brewing for years. But the New York Times just changed the game. In a new amended complaint, the newspaper isn't just going after OpenAI for scraping its articles. It's aiming directly at Microsoft, arguing the tech giant built a purpose-built supercomputer for the explicit goal of committing mass copyright infringement. This isn't about providing a generic cloud server anymore. The accusation is that Microsoft knowingly built the weapon, not just rented out the armory.

The legal shift hinges on proving intent. Following a recent Supreme Court precedent where the court sided with Cox Communications, plaintiffs now have to show a party actively induced illegal conduct. To meet that standard, the NYT's updated filing argues Microsoft’s supercomputer wasn’t a neutral platform. It was an "unusually complex" machine, ranked among the world's most powerful and custom-built for one job: training OpenAI's models. The complaint alleges this system was designed to vacuum up the internet and that The Times's content was disproportionately featured and weighted, all in an effort to create models that could mimic high-quality journalism. This reframes Microsoft's role from a passive cloud provider to an active co-conspirator who engineered the very means of infringement.

This fight is about who owns the economic upside of AI. The original suit centered on ChatGPT substituting for a NYT subscription. The amended complaint escalates the stakes to a planetary scale. The filing directly claims Microsoft’s strategy helped boost its market capitalization by a trillion dollars in the past year alone. Microsoft gets a historic valuation bump; the newspaper gets its core asset devalued into free training data. While a Microsoft spokesperson told Ars this is a plaintiff's "last-ditch effort," for content industries this lawsuit is a line in the sand. If the architecture of the internet's most valuable new products is built on uncompensated source material, the business model for creating that material effectively dies.

Forget a quick settlement. In a heavily redacted court filing, the NYT is airing discovery evidence, including user chat logs where people admit to using ChatGPT to bypass its paywall. The lawsuit will now grind forward on the specific question of whether building custom hardware for training constitutes intent to infringe. A loss for the Times could establish a powerful precedent, blessing the practice of training models on anything digital. A win could force a catastrophic and expensive re-engineering of the entire AI stack for Microsoft and its peers. The question isn't whether AI can replicate journalism. It's who gets paid when it does: the creator of the original work, or the creator of the machine that copied it?