AI This Week

Meta announced a far-reaching plan to secure nuclear power for its data centers, aiming to meet the surging electricity demand from AI training and inference. The company intends to pursue long-term power deals and partnerships with nuclear operators, targeting firm, carbon-free supply that can scale. The plan seeks round-the-clock power for new and existing sites. It also spotlights the escalating energy race among tech giants and the strain AI could put on grids. The plan faces hurdles, including regulatory approvals, project timelines, and high upfront costs. If it proceeds, it could reshape how digital infrastructure draws power.

MiniMax Group, the second of China’s “AI tigers” to go public, has seen its shares skyrocket. The stock doubled in value during its first day of trading in Hong Kong. Investors enthusiastically pursued the startup, which develops popular consumer AI applications like the video tool Hailuo AI and the character interaction app Talkie. The company raised $620 million in its initial public offering to fund further research and development. MiniMax’s successful debut outshined that of Zhipu AI, the first AI tiger to list. The high investor demand highlights a strong appetite for consumer-facing AI technology. Analysts suggest MiniMax’s consumer focus presents higher growth opportunities compared to Zhipu’s enterprise-oriented model.

OpenAI has launched ChatGPT Health, a new feature in the US that analyzes personal medical records. The system can also review data from health apps like Apple Health and MyFitnessPal to provide users with personalized advice. OpenAI states the new feature is not for diagnosis or treatment. The company also claims health-related conversations will be stored separately and will not be used to train its AI models. Currently available to a small group of US users via a waitlist, the feature has not been introduced in the UK or Europe, which have stricter data protection rules

A new artificial intelligence model can forecast a person’s risk for serious diseases, including dementia and heart failure, using data from a single night of sleep. Researchers developed the system, called SleepFM, by training it on a vast library of over half a million hours of sleep studies from 65,000 individuals. The model examines complex biological signals recorded overnight, including brain waves, heart rhythms, and breathing patterns. From this single snapshot, it identifies subtle indicators of future health problems. The results are startling. SleepFM predicted all-cause mortality with a C-index of 0.84 and dementia at 0.85. It also demonstrated strong predictive performance for heart attack, stroke, and chronic kidney disease. A novel design allows it to function effectively even when a patient’s sleep data is incomplete.

The world’s largest technology conference is presenting the next wave of gadgets in Las Vegas. CES 2026 is filled with advancements in artificial intelligence, from a new AI pet that physically grows as you care for it to a smart ring that records and transcribes your work meetings. Robotics also had a key presence. Boston Dynamics debuted its new Atlas robot, and Qualcomm-powered humanoids demonstrated backward-bending flexibility. Other notable reveals include LG’s ultra-thin wallpaper TV, a pocket-sized lab that detects food allergens, and exoskeletons built to make hiking easier. Samsung also exhibited a creaseless foldable phone concept and a massive 130-inch television.