Good morning. It's Thursday, March nineteenth, twenty-twenty-six.

**Autoscience raises fourteen million for automated AI research.** A San Mateo startup called Autoscience just raised fourteen million dollars in seed funding — led by General Catalyst, with Toyota Ventures and the Perplexity Fund also in. The pitch: replace human bottlenecks in machine learning research with AI scientists and engineers that invent, validate, and deploy new models autonomously. They point out that over two thousand machine learning papers get published every week — no human team can keep up. Autoscience's first deployments target financial applications, manufacturing, and fraud detection. This connects directly to the Karpathy Loop autoresearch story from earlier this week — there's a clear theme forming around AI systems that don't just run experiments, but design them. If it works, the pace of ML progress could accelerate in ways that are genuinely hard to predict.

**NASA rolls Artemis Two back to the launch pad.** After weeks in the Vehicle Assembly Building following that helium anomaly, NASA's three-hundred-twenty-two-foot Space Launch System is heading back to the launch pad at Kennedy Space Center tonight — rollout expected to start around eight p.m. Eastern. The crew of four — including Canadian astronaut Jeremy Hansen — could be looking at an April first launch window at the earliest. This is the second rollout of the year after the February scrub, and NASA has restructured the broader Artemis program: Artemis Three is no longer a landing mission, with dual lunar landings now targeting twenty-twenty-eight. After eight total scrubs across Artemis One and Two, seeing that rocket vertical on the pad again is at least a visual milestone — whether the window holds is another question.

**Landspace attempts methane rocket return-to-flight.** China's Landspace company is attempting the return-to-flight of its Zhuque-2E rocket today from the Jiuquan Satellite Launch Center in the Gobi Desert. The Zhuque-2 is notable as one of the first methane-fueled orbital rockets to successfully reach orbit — it did so in twenty-twenty-three — but suffered a failure last August. Today's 216-foot stainless steel rocket, running on liquid methane and liquid oxygen, is carrying a constellation satellite payload and attempting to get back on track. With SpaceX Starship, Rocket Lab's Neutron, and now multiple Chinese methane rockets all racing toward reusability, this launch is a data point in the broader methane propulsion story that's reshaping the industry.

**NVIDIA GTC wraps with physical AI open-source push.** Today is the final day of NVIDIA's GTC twenty-twenty-six conference in San Jose — and the week's closing news is that NVIDIA is open-sourcing its Cosmos world foundation models and new Alpamayo models for autonomous vehicle development on GitHub and Foundry. These are the physical AI building blocks Jensen Huang was teasing in the keynote — world models trained on physical interaction data that robots and autonomous systems can use as a substrate. Healthcare robotics, industrial automation, and autonomous vehicles are the named targets. It's NVIDIA's bet that physical AI follows the same playbook as language AI: pre-train on massive datasets, fine-tune for your application. Whether the data flywheel for robots spins as fast as it did for text remains the open question.

That's your Thursday morning briefing. Stay curious.