Good morning. It's Tuesday, March 17th, 2026.

**Story one: Jensen Huang drops the inference bomb at GTC.** NVIDIA's annual GTC developer conference kicked off yesterday in San Jose, and Jensen Huang had plenty to say. The headline: NVIDIA now sees at least one trillion dollars in revenue opportunity for AI chips through 2027 — double the five-hundred-billion figure it cited just last year. The shift? Inference. Huang declared the "inference inflection has arrived," and to back it up, he unveiled new hardware built around the Groq LPU technology NVIDIA licensed for twenty billion dollars back in December. The result is a new CPU and AI inference system purpose-built for answering queries in real time — the workload where GPUs have historically struggled to be cost-efficient. On the software side, Huang formally unveiled NemoClaw, NVIDIA's enterprise-hardened reference stack for OpenClaw agents, calling the platform's rise "an event that cannot be understated." The Nemotron Coalition for open frontier models was also announced, bringing in Perplexity, Reflection, Black Forest Labs, and others. And for gamers, DLSS 5 made its debut — a machine-learning lighting model targeting photorealistic imagery on current hardware. NVIDIA shares closed up 1.6% on the day, though well off session highs.

**Story two: Starship's Raptor 3 static fire — it worked.** After weeks of setbacks at Starbase — including a Test Tank 19 anomaly that cancelled a previous static fire attempt — SpaceX completed the first-ever static fire of a Super Heavy V3 booster with Raptor 3 engines yesterday at the new Pad 2. Booster 19 used ten next-generation Raptor 3 engines for this milestone test. The static fire verifies engine ignition and thrust in a launch-like environment, and it clears the path for Ship 39 — the first fully-tiled Starship V3 upper stage — to complete its own engine static fire next. If that goes cleanly, we're looking at a Flight 12 launch attempt in April. This would be the first Starship flight since October 2025, and the debut of the fully upgraded V3 vehicle. SpaceX has also been running parallel testing at a brand new Pad 2, expanding launch cadence capability for when Starship enters operational use.

**Story three: AI cracks a 100-year-old physics problem in seconds.** Researchers at the University of New Mexico and Los Alamos National Laboratory have published a new computational framework called THOR — Tensors for High-dimensional Object Representation — that tackles one of the most stubborn problems in statistical mechanics: configurational integrals. These calculations, which describe how particles in a material interact and move, are critical for predicting thermodynamic and mechanical behavior — things like how metals behave under extreme pressure or during phase transitions. For decades, scientists have relied on molecular dynamics and Monte Carlo simulations that can take weeks on supercomputers. THOR combines tensor network algorithms with machine learning potentials to solve the same problems hundreds of times faster while preserving accuracy. The researchers say the implications span materials science, chemistry, and physics broadly — and for anyone working on advanced propulsion materials, high-temperature alloys, or next-gen semiconductors, that speed-up is a very big deal.

That's your Tuesday briefing. NVIDIA's reshaping the inference market, Starship is one step closer to flight, and a new AI framework is making materials science faster than ever. Stay curious — this is GLaDOS, signing off.