The Times is now putting Microsoft at the center of its copyright case, not at the edge of it. If a court treats AI infrastructure as active participation in scraping, cloud companies have a much bigger problem than bad headlines.
Microsoft has spent most of the New York Times copyright fight looking like the quieter defendant: the cloud partner, the compute supplier, the company behind Azure rather than the company deciding which articles went into a training set. That position is harder to hold after the Times filed a third amended complaint on June 25 accusing Microsoft of helping build the machinery used to copy news at scale.
According to reporting from Bloomberg Law and MLex, the new complaint says Microsoft supplied OpenAI with a supercomputing platform designed to download, store, and reproduce copyrighted works for AI training. It also alleges that Microsoft provided internet scraping software and unauthorized content. That is not a small tweak. The Times is trying to move Microsoft from landlord to participant.
The legal distinction is blunt. The Times is not just saying Microsoft servers happened to hold material that later became part of OpenAI models. It is saying Microsoft knew what the system was built to do, helped build it, and then benefited when those models showed up inside Bing Chat and Copilot. Contributory infringement turns on knowledge and material contribution. The complaint is written to hit both.
Here is the thing: you should not read this as a narrow fight over one newspaper and one chatbot. The day before the Times filing, Bloomberg reported that a coalition representing nearly 400 local and regional newspapers sued OpenAI and Microsoft in the Southern District of New York. That lawsuit, filed by the Local News Copyright Alliance, accuses the companies of crawling news sites, including paywalled and restricted pages, then copying articles onto their own servers to train ChatGPT and Copilot.
Two newspaper cases landing within 24 hours against the same defendants is not background noise. It is the industry saying the old bargain has broken. Newsrooms published work for readers, subscribers, search engines, and licensing partners. They did not publish it so the largest software companies in the world could turn it into training data, sell the output back to the market, and call the whole thing inevitable.
Microsoft's own AI spending makes the issue larger. Business Insider reported in April that Microsoft raised its 2026 capital spending plan to about $190 billion, far above analyst expectations, as it races to add data center capacity for AI demand. The Wall Street Journal reported last October that Microsoft planned to double its data center footprint within two years and lift AI capacity by more than 80% in the year ahead. Azure is not a side business here. It is the road Microsoft is paving for Copilot, OpenAI, and the rest of its AI stack.
That is why the Times's theory bites. If a court accepts that purpose-built training infrastructure can be part of the infringement, the question no longer stops with the lab that trained the model. It moves upstream to the company that supplied the compute, the scraping tools, the storage, and the commercial route to market. Any AI startup buying cloud capacity should care about that, because your procurement team can no longer treat training data as someone else's legal footnote.
Buyers already ask harder questions about data provenance before they sign enterprise AI contracts. Where did the training data come from? Was it licensed? Can the vendor prove it? A ruling against Microsoft on the active-participation theory would push those questions into every cloud deal behind the model. Frankly, that is where they belong. If your product depends on scraped work, the problem does not disappear because the copying happened inside a very expensive data center.
The Times case is still far from over. In March 2025, U.S. District Judge Sidney Stein allowed the core copyright claims against OpenAI and Microsoft to proceed while dismissing some other claims, including unfair competition. In a later discovery fight, Magistrate Judge Ona Wang ordered OpenAI to produce 20 million de-identified ChatGPT logs for review under protective restrictions, a ruling OpenAI attacked publicly as a privacy threat. Those details matter because they show the case has already moved past press-release combat into the grinding work of discovery.
Microsoft will fight the new framing hard. It has every reason to argue that cloud infrastructure is general-purpose technology, not an infringement device. OpenAI will keep leaning on fair use. The courts have not resolved the central question yet, and no one should pretend the Times has already won.
But the direction is clear enough. The AI industry built its training pipelines during a period when scraping the open web looked like an engineering problem first and a legal problem later. Later has arrived. If you run an AI lab, fund one, or sell the cloud capacity behind one, the Times's new complaint is a warning written in plain legal language: the infrastructure is now part of the story.
Also read: OpenAI is raiding Apple's Vision Pro talent as the headset quietly dies; Robert Kuok's grandson is staking €5.3 billion on Italy becoming Europe's AI infrastructure capital; and Apple's touchscreen MacBook arrives on M5 chips as AI memory costs reshape the whole product line
