Good morning. It's Sunday, July fifth, twenty twenty-six.

**Story one.** SpaceX flew something new — not just another batch of Starlink satellites, although that too. Starlink Group Ten to Fifty. Twenty-nine satellites launched from Cape Canaveral just before seven AM Eastern this morning aboard a Falcon 9. But the real story was riding attached to the Falcon 9 first-stage booster: two semiconductor manufacturing test pods, called Fabships, built by Washington DC startup Besxar Space Industries. The entire first stage basically turned into a flying fab lab for an eight-minute suborbital ride. The vacuum of space enables ultra-pure substrate production that's impossible to replicate on Earth. Besxar has booked twelve dedicated Falcon 9 flights for this, and SpaceX is actually an investor. Nvidia's Inception program backs them too. The goal — autonomous orbital manufacturing plants for the semiconductor precursors that power AI chips. It's a vertical integration story wrapped in a launch story: the company making the rockets is funding the company using rockets to build the things the rockets need.

**Story two.** Meituan, China's food delivery and local services giant, has open-sourced a 1.6-trillion-parameter AI model called LongCat-2.0 under the MIT license. It's a Mixture-of-Experts architecture with 33 to 56 billion active parameters per token, a million-token context window, and built for agentic coding. But here's what makes it geopolitically significant: Meituan says it completed both training and inference entirely on domestic Chinese chip clusters. VentureBeat confirmed the training was done on Chinese silicon, making it a test case for how far domestic hardware can go when Nvidia GPUs are off the table. With US export controls tightening on Nvidia GPUs to China, this is proof of concept that China can still build frontier-scale models using homegrown silicon. The weights are already on HuggingFace. It raises the question — at what point do domestic Chinese chips become good enough that export controls are no longer a bottleneck, just a tax?

**Story three.** Agility Robotics is going public via a two-point-five-billion-dollar SPAC merger with Churchill Capital Corp Eleven. The ticker will be AGLT on Nasdaq. The company's two-legged humanoid robot, Digit, has logged sixty-five thousand hours of real commercial warehouse operation with customers including GXO, Toyota, and Schaeffler. They've got a three-hundred-million-dollar order book — backlog, not revenue. This is the first pure-play humanoid robotics company to hit public markets, and the valuation will be a temperature check on whether investors believe the humanoid thesis or think we're past the hype peak.

**Story four.** Singapore's dConstruct Robotics closed a hundred-and-twenty-five-million-dollar Series A for spatial tech that lets autonomous robots operate in GPS-denied environments — warehouses, underwater, underground. The kind of positioning infrastructure that makes or breaks real-world deployment. When you can't rely on satellite signals, robot autonomy gets a lot harder. This is the unsexy foundational layer that the flashy humanoid demos can't function without.

That's all for today. Have a great Fourth of July weekend.