On 15 July NVIDIA launched two new Jetson Thor modules - T3000 and T2000 - on the Blackwell architecture, for robotics and edge AI. The cheaper one lands exactly where earlier robot boards were too expensive for a device made in volume.
- NVIDIA announced Jetson Thor T3000 and T2000 (Blackwell) for robotics and edge AI.
- T2000 is entry-level (per NVIDIA, around 400 FP4 TFLOPS, 16 GB memory); T3000 comes close to the pricier T5000 while using less memory.
- Emulation mode via JetPack 7.2.1 this month; production modules ship in Q1 2027.
Robots don't think in the cloud - or at least not entirely. Whatever has to happen instantly (not crashing into something, catching an object in time) gets computed on the device itself. Until now, that edge compute was either weak or expensive. NVIDIA is launching two new ones, and one of them is deliberately built for the mass-market end.
This is where the craft shows - not in the most powerful chip, in the cheapest one. T2000 is the move that matters. The powerful module has existed for a while; what was missing was a board cheap enough to go into a device made at mass scale. Price, not peak performance, decides what stays a prototype and what reaches the market.
We don't make robots. But the cheaper edge compute gets, the more devices will want to think for themselves - in the car, in the warehouse, at home. This chip is the quiet start of a lot of those devices. And every new thing that reasons on the spot is also a new thing that someday you have to let into your network, or keep out.