NVIDIA expands Jetson Thor family with lower-power T3000 and T2000 modules for mass-market robotics

NVIDIA said Wednesday it is expanding its Jetson Thor robotics lineup with two lower-power computing modules, the Jetson T3000 and Jetson T2000, aimed at more mainstream robots and edge AI systems. In a company blog post published July 15, Chen Su of NVIDIA said the new modules are designed to support “mass-market robotics and edge AI” as autonomous machines move beyond research settings and into broader deployment. NVIDIA said production versions of both modules are scheduled to become available in the first quarter of 2027.

According to NVIDIA, the Jetson T3000 delivers 865 FP4 teraflops of AI compute in a compact form factor and includes a Blackwell GPU, an eight-core Neoverse Arm CPU, 32 GB of LPDDR5X memory and 25 GbE connectivity. NVIDIA said the Jetson T2000 delivers 400 FP4 teraflops of compute and 16 GB of memory. The company’s pitch is not just smaller hardware, but the ability to run foundation-model inference — meaning AI models used for broad tasks such as perception and reasoning — directly on robots and other edge devices rather than in a remote data center.

NVIDIA is also trying to give developers software access before the hardware ships. The company said T3000 emulation mode will be available later in July through JetPack 7.2.1, a version of NVIDIA’s software stack for Jetson systems. Support for T2000 emulation is planned for a later release. Alongside that, NVIDIA highlighted what it calls “Jetson agent skills,” software intended to automate memory optimization, system configuration and deployment across the Jetson portfolio.

That software matters because memory use can be a limiting factor for running large AI models on embedded systems. NVIDIA tied the new modules to software features introduced with JetPack 7.2 in June, particularly tools for automated memory optimization and deployment. In the July 15 blog post, Su wrote: “To meet that need, NVIDIA today introduced the T3000 and T2000, new modules based on the NVIDIA Thor architecture that enable mass-market robotics and edge AI applications at scale.”

Jetson is NVIDIA’s embedded computing platform for robotics and edge AI, and Thor is its newer Blackwell-based generation of hardware. The new announcement extends the Thor family downward from higher-end products NVIDIA had already brought to market. The company made the Jetson AGX Thor developer kit and the T5000 and T4000 modules generally available in August 2025, positioning those products around more demanding robotics workloads.

NVIDIA said one example of the type of AI workload it is targeting is Cosmos 3 Edge, a 4-billion-parameter model that the company says is compatible with Thor platforms for on-device inference. That points to NVIDIA’s broader effort to move generative and foundation-model capabilities onto machines operating in the physical world, including robots and industrial edge systems. When NVIDIA introduced Jetson Thor in 2025, Chief Executive Jensen Huang said, “We’ve built Jetson Thor for the millions of developers working on robotic systems that interact with and increasingly shape the physical world.”

The performance figures for the T3000 and T2000 are NVIDIA’s own claims. At the time of the announcement, no independent third-party benchmarks for either new module were identified.