Nvidia’s Jetson T3000 and T2000 land in 2027

Nvidia has introduced the Jetson T3000 and Jetson T2000, new Blackwell-based compute modules aimed at robots, autonomous transport, and industrial equipment. The pitch is straightforward: process camera and sensor data directly on the device instead of relying constantly on the cloud. Nvidia says sales will begin in the first quarter of 2027.

The launch fits a broader shift in edge AI, as compute moves closer to sensors and into deployed hardware. The market is already crowded: Qualcomm is pushing robotics and camera platforms, Intel is building out industrial AI, and Nvidia itself introduced Jetson AGX Thor this spring as a stepping stone for the ecosystem.

The top-end Jetson T3000 includes a Blackwell GPU, 8 Arm Neoverse cores, 32GB of LPDDR5X memory, and memory bandwidth of up to 273GB/s. Nvidia lists peak performance at 865 FP4 teraflops. The Jetson T2000 comes with 16GB of memory and up to 400 FP4 teraflops.

Recommended reading

Some AI problems stay unsolvable, even with infinite data

Nvidia is also planning an IGX-T3000 variant with enhanced security for the higher-end platform. The company says Jetson Agent Skills helps reduce memory use and simplify model deployment.

According to Nvidia, the Jetson T3000 takes up roughly half the space and uses less power than the Jetson T5000, while delivering comparable AI performance for computer vision, large language models, and robot control.

The Jetson T2000 is positioned for:

• autonomous mobile robots
• manipulators
• smart cameras
• logistics systems
• production lines

Nvidia also showed Cosmos 3 Edge, a 4 billion-parameter model designed for environmental analysis and real-time prediction of robot actions.

Before the new modules ship, developers will be able to work with Jetson AGX Thor. In July 2026, it will receive a Jetson T3000 emulation mode, with Jetson T2000 emulation coming later. That gives developers a way to start building ahead of the 2027 launch.