AI News21 Nov 2026
An Anthropic mathematician says he and Claude have proved the Riemann hypothesis
The most famous unsolved problem in mathematics may have fallen this week, with an AI sharing the credit. Levent Alpöge, a mathematician at Anthropic, says he has proved the Riemann hypothesis, a 167-year-old conjecture about how prime numbers are spread. He found the proof together with Claude, Anthropic's AI model, and published it as computer code that software can check. Outside mathematicians, one at MIT among them, report that it passes. Nobody has yet explained in ordinary mathematics why it works. The same day, Stanford's Ravi Vakil published a rival proof by Alessandro Masullo, "all human, with no AI". Experts need time to judge both. I did not expect to write this in 2026.
Read more: TechCrunch →AI News21 Nov 2026
Trump says American companies may no longer build data centres abroad
Much of Europe's business data sits in data centres that American companies built here. On Friday, Donald Trump said American companies would no longer be allowed to build data centres outside the United States, or would pay "a tremendous tax" if they did. He named Finland, where Google plans to invest about 13 billion euros. He also said he had not yet told his own staff, and I have seen no explanation of which law would allow it. So it may come to nothing. If you are planning around new cloud capacity in Europe, though, ask your supplier where it will come from.
Read more: washingtontimes.com →AI News21 Nov 2026
AI shares have their worst week since 2020
The week of the largest stock market listing on record ended with investors running from AI. The Nasdaq 100, the index of America's biggest technology companies, fell 10 percent this week, its worst week since 2020, and has lost all of this year's gains. Nvidia, which makes the chips AI runs on, dropped 14 percent on Friday alone, after Reuters reported that the SEC, America's financial regulator, is examining its accounts. Anthropic, whose listing was the record, gave back most of its first-day jump. A week is short, and these companies are still growing fast. When investors turn cautious, AI suppliers find cheap prices harder to fund, and that is where we would notice it.
Read more: stockanalysis.com →AI News21 Nov 2026
Britain's AI minister says the country does not need its own top AI model
Should a European country build its own top AI model, or buy access to America's? Britain's AI minister, Kanishka Narayan, gave his answer in Silicon Valley this week: a British model is not "a matter of necessity as long as we have trusted interdependence and access". He would rather put public money into chip design and science, where others could come to depend on Britain. British founders are furious, the startup site Sifted reports, and the trade body UKAI has warned of a "vassal state". I think most European governments would reach the same conclusion. Set it beside Trump's threat above: access is what he just put in question.
Read more: cityam.com →AI News21 Nov 2026
Alibaba says it is training its next AI model on its own chips
Alibaba now says that it, too, can train AI without American chips. On Thursday I wrote about Huawei making that claim. Now Eddie Wu, Alibaba's chief executive, has told investors that the next version of Qwen, its AI model, is being trained on processors from Alibaba's own chip unit. He gave no figure for how much of the work they do, and I have seen no outside check. Washington tightened its export rules again on Monday. So this week America fenced its AI in, Britain chose to depend on it, and China said it is building its own hardware. European companies sit in the middle of that, which makes it our business too.
Read more: technode.com →