Technical Staff

Pentagon AI contracts skip Anthropic and Nvidia loses China. Robert Hattala May 5, 2026 p>Big day in AI news, and not the kind where someone ships a new model with a clever logo. This was the kind of day where the bills come due and the politics show up. Three stories worth your time today. Pentagon money, Nvidia's China problem, and a real medical win from Mayo Clinic. Let's get into it. Pentagon hands out AI contracts and Anthropic walks. On May 1, the Pentagon awarded classified-network AI contracts to AWS, Google, Microsoft, Nvidia, OpenAI, SpaceX, and Reflection AI. Anthropic was not on the list. The why is the part that matters. Anthropic refused to let Claude be used for "all lawful" purposes. They flagged the language as too broad, saying it could open the door to domestic mass surveillance or autonomous weapons. Why it matters: there's a real split now between AI labs willing to take any defense dollar and labs that draw a line in the sand. The Pentagon wrote a blank check, and one company said no thanks. My take: I respect Anthropic for this. You can't build a "safe AI" brand and then hand the keys to whatever a contracting officer feels like doing on a Tuesday. The dollars are big. The trust they keep is bigger. Other labs ought to pay attention because customers in regulated industries are watching too. Nvidia's China AI share just hit zero. Jensen Huang said this week that Nvidia's market share for AI accelerators in China is now zero. Huawei, Cambricon, and other domestic chip outfits are filling the gap. Huang argued the U.S. export controls have "already largely backfired." Hard to disagree when your share goes from dominant to nothing in a couple of years. Why it matters: export controls were supposed to slow China down. Instead they pushed China to build its own stack. Now there are two separate AI hardware worlds, and the American one no longer ships into the bigger market. My take: this was always the risk with broad export controls. You don't strangle a country with this much capital and this many engineers. You just give them a five-year head start on import substitution. The U.S. lost a customer and gained a competitor in the same trade. That's a bad day on the chess board. Mayo Clinic says AI can spot pancreatic cancer years early. Mayo Clinic researchers showed an AI model that flags pancreatic cancer years before clinical diagnosis. Pancreatic is one of the worst because by the time you feel something, it's usually late. Why it matters: this is the kind of work that justifies all the noise around AI in medicine. Not chatbots that summarize charts. Not slick demos. Pattern recognition on imaging and labs that actually saves lives. My take: more of this please. Less PR theater about agents that can book your dentist appointment, more boring research that catches a cancer two years before a human radiologist would. Mayo is doing it right by publishing real results instead of dropping a press release and hiding the data. Putting it together. One day, three signals. AI vendors are getting sorted into camps based on who they will sell to. Hardware policy is shifting under everybody's feet. And the actual humans-helping-humans use cases are quietly winning while the rest of the field argues on Twitter. Worth paying attention to all of it.

Shinsegae Group and Reflection AI to build 250MW sovereign AI data center in South Korea. Will offer compute to government and enterprise clients March 19, 2026 Shinsegae Group and Reflection AI are teaming up to build a 250MW data center in South Korea, which they claim will be the country's largest sovereign AI factory. Korean conglomerate Shinsegae and AI lab Reflection have signed a Memorandum of Understanding (MoU) laying out their intention to set up a joint venture (JV) to build the data center. The MoU was signed in San Francisco earlier this week at a ceremony attended by US Commerce Secretary Howard Lutnick. A Shinsegae department store in Centum City, Busan, Korea - Modamoda/Wikimedia Commons Under the JV agreement, Shinsegae will secure a plot of land and build the data center, for which Reflection AI will be responsible for the design and operation of the equipment inside. The project is expected to cost at least 10 trillion won ($6.8bn). Further details haven't been shared. "The Reflection and Shinsegae Group AI factory will equip the Korean government and enterprises with fully sovereign frontier capabilities built and operated on home soil, setting a new standard in technological self-sufficiency," a statement from the companies said. Shinsegae Group operates department stores, as well as e-commerce, hospitality, and real estate businesses, and is apparently keen to become South Korea's answer to Amazon and the other hyperscalers. "The data center that we are building in partnership with Reflection AI will not only be a growth opportunity for Shinsegae, but also a pivot point for the Korean AI ecosystem," said Yongjin Chun, chairman of Shinsegae Group. Reflection AI is developing open AI models akin to China's DeepSeek, rather than the proprietary systems owned by the likes of OpenAI and Anthropic. It was set up by Misha Laskin and Ioannis Antonoglou, two former Google DeepMind engineers. It is backed by Nvidia and raised $2 billion in October 2025. Earlier this month, it was reported to be seeking further funding in a round that would value the business at $20bn. In November, it announced a deal to use GMI Cloud's US-based infrastructure to train and run its models. It is also one of a number of companies that are part of an Nvidia-founded open AI coalition, Nemotron, which was launched at the GPU giant's GTC event in California earlier this week. Get a roundup of the latest regional news across asia fortnightly. More in construction & site selection.