NVIDIA Jetson Thor T3000 brings Blackwell AI to compact edge robots

NVIDIA has introduced the Jetson Thor T3000 and T2000 modules to shrink the footprint of edge AI systems. These new computers target robots and industrial machines that require high performance in a compact form factor. The shift matters because it allows engineers to deploy powerful AI models without the bulk of previous generations. Buyers can now fit advanced inference capabilities into tighter spaces while managing power consumption more effectively.

New modules shrink footprint while maintaining high inference performance

The Jetson Thor T3000 serves as the flagship option in this new lineup. It pairs a Blackwell GPU with an eight-core Arm Neoverse processor to handle complex workloads. This combination supports demanding tasks in language, vision, and robotics models. The T2000 acts as a lower-end alternative for systems that need less raw compute power.

Spec comparison

The T3000 delivers 865 FP4 TFLOPS of AI performance using 32GB of LPDDR5X memory. This memory configuration provides 273GB/s of bandwidth to keep data flowing efficiently. It also includes 25 GbE connectivity for high-speed network communication. The T2000 offers 400 FP4 TFLOPS and 16GB of memory for scaled-down applications.

NVIDIA claims the T3000 is roughly half the size and uses about half the power of the existing T5000 module. Despite these reductions, the company states it delivers similar inference performance for key workloads. This efficiency gain is critical for mobile robots and battery-powered industrial devices. The T2000 provides a cost-effective entry point for smaller edge deployments.

Both modules are scheduled to become available in the first quarter of 2027. This timeline gives developers time to prepare their integration strategies for the new hardware. The T3000 emulation is also available now via JetPack 7.2.1 for early software development. Engineers can start building and testing their applications before the physical hardware ships.