Nvidia Jetson AGX Thor T5000 Developer Kit with Compute Carrier for Next-Gen Physical AI

A New Benchmark for Physical AI

The NVIDIA® Jetson T5000™ module combined with its compute carrier board delivers a new standard for physical AI and robotics, offering up to 2070 FP4 TFLOPS of AI compute with 128 GB LPDDR5X memory. Power is configurable between 40 W and 130 W, giving developers flexibility for efficiency or maximum performance. Designed for advanced autonomous machines, humanoid robotics, and industrial AI applications, this setup combines high-density compute with extensive I/O to handle highly complex AI workloads at the edge.

Unmatched Hardware Architecture

The Jetson T5000 module features an NVIDIA Blackwell GPU with 2560 CUDA cores, 96 Tensor Cores, a transformer engine, and Multi-Instance GPU (MIG) support, paired with a 14-core Arm® Neoverse®-V3AE CPU for general-purpose and multi-threaded tasks. Integrated accelerators—including a third-generation PVA, dual video encoders/decoders, and an optical flow accelerator—enable real-time sensor processing, making the module ideal for demanding robotics and physical AI workloads.

Paired With Next-Gen Heatsink & Compute Carrier

When paired with the compute carrier board, the system provides extensive connectivity for prototyping and deployment. This includes camera inputs via HSB and USB, expansion slots such as PCIe M.2 Key M (x4 Gen5, 1 TB NVMe) and M.2 Key E (x1 Gen5, Wi-Fi/Bluetooth), front/rear I/O including 2× USB-A 3.2, 2× USB-C 3.1, 5 GbE RJ45, QSFP28 (4× 25 GbE), HDMI 2.0b, and DisplayPort 1.4a, plus internal I/O and control interfaces like CAN headers, automation headers, JTAG connectors, fan connector, audio headers, RTC backup, Microfit power jack, and power/force recovery/reset buttons—giving developers full control over sensors, cameras, and peripherals for complex AI systems.

Generative AI and Multi-Sensor Performance

Jetson T5000 is architected for next-generation physical AI workloads, supporting Vision Language Action (VLA) models like NVIDIA Isaac™ GR00T, as well as LLMs and VLMs. With real-time low-latency multi-sensor processing, it enables autonomous machines to perceive, plan, and act intelligently. The NVIDIA AI software stack runs seamlessly on this setup, including Isaac for robotics, Metropolis for visual agentic AI, and Holoscan for sensor fusion, with support for agentic AI workflows such as Video Search and Summarization (VSS) at the edge.

Production-Ready Platform for Physical AI

Together, the T5000 module and compute carrier board provide a complete, production-ready platform for humanoid robotics, industrial automation, smart spaces, retail, and medical AI applications. The compact carrier board houses all essential I/O, expansion, and connectivity for sensors, cameras, and storage while keeping the system compact and deployment-ready. Combined with the NVIDIA AI software ecosystem and partner solutions, Jetson T5000 accelerates solution development from prototype to production.

• 2070 FP4 TFLOPS AI compute performance in a compact module.
• NVIDIA Blackwell GPU with 2560 CUDA cores and 96 Tensor Cores.
• 14-core Arm Neoverse-V3AE 64-bit CPU for high-throughput processing.
• 128GB LPDDR5X memory, 256-bit, 273 GB/s bandwidth.
• Power range: 40W–130W configurable for efficiency or maximum performance.
• Supports MIG (Multi-Instance GPU) and transformer engine for generative AI and large models.