One Thor platform, every deployment scale.

YUAN High-Tech, an NVIDIA ecosystem partner, has begun development of its next-generation edge AI platforms on the new NVIDIA Jetson T3000 and T2000 modules. Both are built on the same NVIDIA Blackwell architecture and software stack as the flagship Jetson AGX Thor T5000 — which means applications written for one migrate to the others with minimal friction, and deployments can be sized to the job rather than the other way round.

The timing matters. On the flagship T5000, YUAN has already shipped production-ready edge AI platforms for factory inspection, autonomous robotics, and intelligent safeguarding — systems that keep inference low-latency and data on-site. As robotics enters large-scale commercialisation, developers increasingly need platforms that balance performance against power, size, and unit cost.

T3000: ninety per cent of the flagship, half the footprint

The Jetson T3000 cuts memory capacity, power consumption, and module size by roughly half compared with the T5000, while preserving the full NVIDIA software stack and delivering around 90 per cent of the flagship's AI inference performance. Developers shrink system footprint, power draw, and deployment cost without giving up capability. The IGX T3000 also retains the Functional Safety Island, the safety architecture required for industrial robots, autonomous systems, and other safety-critical edge AI.

T2000: Thor for the cost-sensitive edge

The Jetson T2000 carries the Thor architecture into a broader class of systems. With 400 FP4 TFLOPS of AI compute, an NVIDIA Blackwell GPU, 16 GB of LPDDR5X memory, and a power envelope of just 40 W, it is a cost-effective entry point for visual AI agents, autonomous mobile robots, robotic arms, and other price-sensitive intelligent edge applications.

Start building now, in emulation

The modules themselves are expected in Q1 2027, but development does not have to wait. T3000 emulation mode arrives later this month with JetPack 7.2.1: on YUAN's existing Jetson AGX Thor development platform, customers can validate applications against the complete NVIDIA software stack — including open models such as NVIDIA Nemotron, NVIDIA Cosmos 3, and NVIDIA Isaac GR00T, along with the latest Jetson AI agent capabilities. T2000 emulation support is planned for a future release.

From the flagship T5000 through the efficient T3000 to the cost-optimised T2000, YUAN now offers a complete Jetson Thor portfolio — one platform, smoothly migrated, from first prototype to high-volume production.