NVIDIA Introduces Jetson Thor T3000 and T2000 Modules for Mainstream Robotics and Edge AI

1. Physical AI Drives Demand for Mainstream Edge Compute

General-purpose robots and autonomous machines are moving from research labs into real-world commercial deployment. New robots rely on natural language interaction, physical perception, trajectory prediction, and autonomous reasoning to complete complex tasks. Humanoid robots, industrial arms, autonomous mobile robots, drones, and intelligent transportation devices are all moving quickly in this wave. Edge deployment, however, faces hard constraints: payload, battery capacity, thermal space, and memory resources are limited, making memory shortage a key bottleneck for large models and multi-model concurrency.

Against this backdrop, NVIDIA introduced the Jetson T3000 and T2000 modules. Both use Blackwell GPU technology and a compact credit-card-size form factor similar to Orin Nano. They fill the mainstream middle tier in the Jetson lineup, creating a scalable platform from entry-level 70 TOPS products to flagship 2,000+ FP4 TFLOPS systems. Leading robotics companies are already building on the Jetson AGX Thor platform.

2. T3000: Half the Size and Power, Near-Flagship Inference Performance

Jetson T3000 is positioned as a main compute module for humanoid robots and high-end robotic systems. In a compact footprint, it provides 865 FP4 TFLOPS of AI performance, while its size and power are roughly half of the flagship T5000. It integrates an NVIDIA Blackwell GPU with 1,536 CUDA cores, an eight-core Neoverse Arm CPU, 32 GB LPDDR5X memory, 273 GB/s memory bandwidth, and 25 GbE networking for high-speed sensor data transmission.

Although smaller, T3000 delivers inference performance close to T5000 on LLM, VLM, VLA, and world foundation model workloads. As memory costs remain high, moving to T3000 can reduce system BOM cost. The IGX T3000 industrial safety version adds functional safety support and can run NVIDIA Halos for Robotics, helping robots operate safely in human-machine collaboration scenarios and meet industrial safety requirements.

3. T2000: Bringing Blackwell to Scaled Edge AI

Jetson T2000 brings the Thor architecture to a broader range of edge AI systems. With 400 FP4 TFLOPS and 16 GB memory, it gives developers an entry option for visual AI agents, autonomous mobile robots, industrial arms, and other intelligent devices. It includes 1,024 Blackwell GPU cores, a six-core Arm CPU, and a standard 40W power envelope. Its form factor is compatible with Orin Nano carrier boards and provides several times the compute performance of Orin Nano.

T2000 is especially suitable for compact AMRs, warehouse AGVs, single-camera inspection devices, small inspection drones, retail recognition terminals, and lightweight educational humanoid robots. Developers can begin with Orin Nano during algorithm validation and upgrade to T2000 when multi-model concurrency or larger VLM inference creates memory pressure, without redesigning the carrier board.

4. T3000 and T2000 Core Specifications

5. Jetson Agent Skills: Memory Optimization in Days

NVIDIA also introduced Jetson Agent Skills, turning code assistants such as Claude, Cursor, and Codex into specialized Jetson development tools. These skills can automate memory audits, inference tuning, system trimming, and application deployment. Developers can optimize the software stack in days rather than weeks and reduce memory usage effectively.

The capability supports Jetson Thor and Jetson Orin products, allowing developers to run more complex workloads on lower-memory configurations or switch to lower-memory SKUs without sacrificing performance. NVIDIA showed customer examples in humanoid robots, smart retail, companion robots, and intelligent transportation where memory usage was significantly reduced.

6. Cosmos 3 Edge Comes to Thor Platforms

NVIDIA also expanded the Cosmos 3 open world foundation model family with Cosmos 3 Edge, a lightweight model compatible with Thor platforms. With 4 billion parameters, it helps embodied systems perceive the world, reason in real time, and predict or generate actions through edge inference. Developers can fine-tune Cosmos 3 Edge for specific robot bodies and sensors, then deploy it on Jetson Thor for real-time visual analysis and robot policy execution.

7. ChipMark Helps Bring the Jetson Thor Family to Market

As an NVIDIA core partner, ChipMark follows the Jetson Thor roadmap closely and provides product selection, developer kit supply, and localized technical support for the full line from T5000 to T3000 and T2000. ChipMark can provide Jetson AGX Thor developer kits and help developers begin software work using simulation modes before final module availability.

Beyond hardware delivery, ChipMark provides carrier board customization, system adaptation, and AI model deployment optimization services, helping enterprise customers move from prototype validation to mass production. Together with NVIDIA Isaac, Isaac GR00T, Nemotron, and Cosmos 3 Edge, ChipMark aims to be a preferred technology partner for robotics and edge AI developers in China.

Development and availability information: T3000 simulation mode is available through JetPack 7.2.1, while T2000 simulation mode will follow in a later release. Jetson T3000 and T2000 modules are expected to be available in the first quarter of 2027. Developers can obtain Jetson AGX Thor developer kits through ChipMark or NVIDIA channel partners.

NVIDIA Jetson Thor, T3000, T2000, Blackwell architecture, edge AI, humanoid robots, physical AI, Jetson Agent Skills, Cosmos 3 Edge, JetPack 7.2, IGX T3000, Halos for Robotics, autonomous mobile robots, visual AI agents, edge computing modules, ChipMark, NVIDIA partner, robot compute, on-device inference, embodied intelligence