Good morning. It's Thursday, April 16th, 2026. Here's what you need to know in tech and science today.

**The White House just dropped a sweeping Space Nuclear Initiative.** OSTP Director Michael Kratsios rolled out the National Initiative for American Space Nuclear Power at the Space Symposium in Colorado Springs, alongside NASA Administrator Jared Isaacman. The plan lays out interagency coordination between NASA, DOD, and the Department of Energy to develop nuclear power and propulsion for deep space. The headline goal: a lunar surface fission reactor ready for launch by 2030. But Isaacman went even further — he announced a plan for SR-1 Freedom, a small interplanetary fission reactor that could launch as early as 2028. It would use nuclear electric propulsion to carry micro-helicopters to Mars. The repurposed Power and Propulsion Element from the canceled Gateway station would serve as the power backbone. This is arguably the most serious push for space nuclear power since the NERVA program. And yes, it's a bipartisan-level priority now.

**Stanford's AI Index 2026 landed, and the numbers are staggering.** Global AI compute capacity has grown three-point-three times every year since 2022 — that's a thirty-fold increase since 2021 alone. Nvidia still dominates with over 60 percent of total AI compute, though Amazon and Google are closing the gap with custom silicon. U.S. companies released 50 notable models in 2025, but China's output is accelerating. China also leads robot deployment by a massive margin — 295,000 industrial robots installed in 2024, compared to just 34,000 in the U.S. The report also raises red flags on emissions: training a single frontier model like xAI's Grok Four can generate over 72,000 tons of carbon equivalent. That's more than 13 times the estimated footprint of GPT-4's training run. The industry is racing toward IPOs — OpenAI and Anthropic both planning public offerings later this year — while local governments increasingly push back on data center development. Growth and friction, happening at the same time.

**South Korean AI chip startup DEEPX is partnering with Hyundai to build chips for generative AI robots.** The company's DX-M2 chips will be manufactured on Samsung's two-nanometer process and go into volume production late next year. DEEPX's current chips are already twenty times more power-efficient than Nvidia's Jetson Orin — at a fraction of the cost — and the next generation is optimized for on-device generative AI, meaning robots that learn from experience without calling to the cloud. Hyundai's planning a thirty-thousand-unit-per-year humanoid robot factory by 2028, and DEEPX wants to be the silicon underneath it. The company is also preparing for an IPO, raising over 400 million dollars in its current round. Founded by a former Apple engineer, this is one of several South Korean startups positioning itself as an edge AI champion. It's exactly the kind of hardware story that matters when you want robots that don't need a server farm attached.

**And finally, Blue Origin's New Glenn rocket is back on the pad, but the launch keeps slipping.** The massive 321-foot rocket has been raised at Cape Canaveral for its third flight, now targeting no earlier than April 19th. It still needs a hot fire test before any official date is locked. The payload: an AST SpaceMobile next-generation Block 2 BlueBird satellite. After that, Blue Origin has even bigger ambitions — the Blue Moon Mark 1 lunar lander is supposed to launch later this year as part of NASA's Artemis program, competing directly with SpaceX's Starship HLS for the Artemis 4 crewed landing. Three flights in and New Glenn is still proving itself, but the hardware is real and getting bigger every time it goes up.

That's all for today. Stay curious.