The most expensive component in a modern AI accelerator is the memory bolted to it, and the world cannot make enough of it.
Oracle is booking committed future business roughly forty-seven times faster than it converts it into revenue.
Microsoft global data center fleet draws roughly 12 gigawatts of power today; by 2032, the company intends for that number to exceed 38 gigawatts.
The Pentagon spent the past year learning how to buy artificial intelligence. It is now learning how to finance it.
Australia entire data center fleet draws about 1.6 gigawatts today. Nvidia has now lined up eight local operators to add as much as 2 gigawatts by the end of 2027.
Huawei most advanced AI accelerator is not shipping yet, and it has already become up to 50 percent more expensive.
Adobe Q3 beat came with a decelerating ARR base, a Q4 guide below consensus, and a CEO handoff to Anil Chakravarthy on December 1.
The model was never the hard part. The real work lives in the unglamorous plumbing underneath, and on September 10 OpenAI started renting that plumbing out.
The most capable coding model an American application company has ever shipped did not begin in San Francisco. It began as a file a Chinese lab posted for free.
For two generations, the way you learned investment banking was by doing the worst job in it. On Thursday, OpenAI shipped a product built to do that in minutes.
Universal Music has spent more than two years telling federal judges that generative AI companies built their products on stolen recordings. On Thursday it signed a deal with one.
For 17 years NASA Lunar Reconnaissance Orbiter has photographed the Moon at roughly one meter per pixel. On September 10, the agency and IBM released the model that read it.
Tom Loosemore asked an AI agent whether the council tax band on his house was fair compared with his neighbours. It came back at once, with an offer to do something about it.
Thom cannot prove his case, and OpenAI cannot disprove it by any means a skeptic would accept. Both facts are features of the architecture.
California began dismantling self-certified AI safety claims on September 9, without writing a single rule about how a model must behave.
The question at the center of the first congressional investigation into an autonomous AI security incident is not how the agents broke out. It is why the company switched the tests back on.
Maria Raine stood a few feet from the governor of California on Thursday and watched him sign a law named for her son.
For two decades, threat intelligence analysts treated craft as a fingerprint. In a report published Thursday, Anthropic says that heuristic is finished.