Europe and Southeast Asia are getting more AI compute. The fight over who controls it is about to get louder. New to this story? Here’s the setup. Alibaba has been tying its chip push to cloud expansion: it unveiled the Zhenwu V900 AI chip and says Alibaba Cloud aims to operate more than 20 gigawatts of global data-center capacity by 2032. The question is whether it can match domestic accelerators with enough cloud capacity to compete as Chinese and Western AI infrastructure bets scale up. This is AI Daily Briefing. We’ve got the buildout, the China-policy fight, and an agent that may have found a theorem—so, who gets to claim the result? We're staying with this story: Alibaba AI infrastructure buildout. Follow the show and you won't miss what comes next. From Roselyne Min at Euronews:
Alibaba is expanding its overseas data centres to Europe, the Chinese tech giant announced Wednesday at its annual conference. The company said it would "establish its first cloud regions" in Finland, the Netherlands and Turkey, while expanding its data centre footprint in Germany, France, the United Arab Emirates, Malaysia and Hong Kong over the next 12 months.
Alibaba’s infrastructure story has gone from chip-and-capacity ambition to specific European cloud regions and near-term expansion. Finland, the Netherlands, and Turkey are new regions, with Germany and France expanding within 12 months. Brussels can now see where this is headed. And the Zhenwu V900 changes the picture. If Alibaba can pair its own accelerator with its own cloud regions, European customers would be choosing an inference stack, with all the supply-chain and export-control baggage that comes with it. Eddie Wu says machine thinking is under 3% of human thinking and could reach a thousand times human capacity. Fine. The measurable claim is 20 gigawatts by 2032—and whether these European regions are actually running competitive workloads on Alibaba silicon. Exactly. Finland and Amsterdam look good on a rollout slide. The test is latency, chip availability, and enterprise demand. Announcing a cloud region is easy; keeping a proprietary accelerator fleet busy is where the spreadsheet starts arguing back. From TechNode Global:
Singapore-founded AI cloud provider Aolani is working with NVIDIA to deploy 22,000 Blackwell Ultra GPUs across new AI infrastructure in Malaysia and the Philippines, expanding accelerated computing capacity in Southeast Asia. The companies expect the new deployments to begin in early 2027.
Twenty-two thousand Blackwell Ultras across Malaysia and the Philippines, starting early 2027. That puts more than 100 megawatts on Aolani’s books—and the utilization test starts when those racks light up. Aolani says it has revenue-sharing and credit support tied to demand, which is better than ordering hardware and praying. But the GPU split, facility locations, and deal value are still under wraps. The Philippines is the interesting bet. Aolani calls it one of the country’s first large-scale AI platforms. Now it needs enough serious regional workloads before 22,000 very expensive chips start aging in public. Alibaba’s European push is the same infrastructure wave taking a different route. NVIDIA’s cloud-partner network is making regional capacity strategic, one deployment agreement at a time. From Yang Cai and colleagues at arXiv:
The main result was entirely obtained by Cogentic, an agentic framework for mathematical discovery, using an interval version of Gemini as the base model. The authors contextualized the findings and verified the proofs. The full exposition here is due to the authors aided by different AI models.
An arXiv paper carrying Google Research and DeepMind names says Cogentic, running an interval version of Gemini, entirely obtained its main theorem. The humans contextualized it and verified the proofs. That attribution may be the bigger precedent. And the result matters on its own: it closes the anytime-regret gap to essentially the fixed-horizon constant without knowing the horizon. But “verified the proofs” needs a much clearer public account. What was machine-checked? What was independently reconstructed? Where did humans step in? We’re getting disclosure norms in theoretical computer science before we’ve even agreed on them for agent demos. According to the authors, Cogentic gets credit for the theorem; the humans wrote the paper. Good. Keep that in the record—don’t let it disappear behind a long author list. From Ars Technica:
To Graylin, the US pretending that China achieving artificial superintelligence first would pose the “greatest AI security threat to the United States” seems flawed, and it’s obstructing cooperation that would help both countries prepare for actual imminent AI risks. Those include threats to workers, national economies, and national security.
Alibaba’s Europe build cuts right into this. Export controls decide whether those new regions can run genuinely competitive workloads; data-center announcements alone don’t settle it. Graylin’s point is what Washington keeps trying to blur: a smaller, specialized model in rogue hands can create a near-term bio, chemical, or cyber threat. “Who reaches superintelligence first” makes for a clean slogan and a lousy policy filter. And his proposed trade is uncomfortable: narrow controls around actual military and CBRNE threats, with open channels for safety testing. I’m not sold on every lever, but pretending zero contact makes American systems safer is a pretty heroic assumption. Trump and Xi can compete over chips and still talk about misuse thresholds, incident reporting, and dangerous-model access. Brussels and Ankara will face versions of that choice as this infrastructure lands on their soil. If you want a deeper dive on the risks and guardrails behind today’s AI news, try AI Safety Daily. It covers AI alignment, model evaluations, emerging risks, and governance—what changed, what the evidence says, and why it matters. Find it wherever you listen to podcasts.
We’ll be watching Alibaba’s planned cloud-region and data-center expansion over the next 12 months, along with Aolani and NVIDIA’s expected Blackwell Ultra deployments in Malaysia and the Philippines, starting in early 2027.
Links to every story are in the show notes. Check out the ones you want to explore further. That’s AI Daily Briefing for today. This is a Lantern Podcast.