NVIDIA Jetson Thor T3000 and T2000 modules bring Blackwell power to mainstream robotics

NVIDIA Expands Jetson Thor Lineup for Physical AI Applications

NVIDIA has unveiled two new computing modules that aim to accelerate the deployment of humanoid robots and autonomous machines across industries. The Jetson Thor T3000 and Jetson Thor T2000, both powered by NVIDIA Blackwell architecture, address the growing demand for compact, power-efficient AI supercomputers capable of running foundation models directly at the edge 1

1

. The announcement signals a shift as general-purpose robots move from research labs toward real-world mass-market deployment, creating opportunities for developers to build smarter systems without relying heavily on cloud infrastructure 3

3

.

Leading technology companies—including 1X, Agility, Agile Robots, Amazon Robotics, Boston Dynamics, FANUC, Hitachi, Medtronic, and Techman Robot—are already building on the NVIDIA Jetson Thor platform 1

1

. Both modules are scheduled to launch in the first quarter of 2027, with the T3000 available for testing in emulation mode this month through JetPack 7.2.1 2

2

.

Jetson Thor T3000 Delivers 865 Teraflops for Humanoid Robots

The Jetson Thor T3000 serves as a streamlined version of the flagship T5000 solution, delivering 865 FP4 teraflops of AI compute in a form factor roughly half the size and power consumption of its larger counterpart 1

1

. The module combines an NVIDIA Blackwell GPU with an eight-core Arm Neoverse CPU, 32GB of LPDDR5X memory, and 273GB/s of memory bandwidth, along with 25 GbE connectivity 1

1

.

Despite its smaller footprint, the T3000 achieves similar inference performance to the T5000 for multimodal workloads, including large language models, vision language models, vision language action models, and world foundation models 1

1

. The platform is rated at 70 Watts, making it suitable for humanoid robots and advanced robotics systems that require substantial AI compute while operating within power constraints 2

2

. The IGX T3000 variant delivers identical performance with integrated functional safety features, seamlessly running the NVIDIA Halos for Robotics full-stack safety system for robots operating alongside humans 1

1

.

Jetson Thor T2000 Opens Entry Point for Edge AI Systems

The Jetson Thor T2000 brings the Thor architecture to a broader range of edge AI systems and autonomous machines. With 400 FP4 teraflops of compute and 16GB of memory, it provides an accessible entry point for developers building visual AI agents, autonomous mobile robots, industrial manipulators, and other intelligent machines 1

1

. The T2000 platform operates at 40 Watts, positioning it as a cost-effective solution for mainstream robotics and edge AI deployments that don't require the full power of the T3000 2

2

.

With these new additions, NVIDIA now offers a scalable edge AI platform spanning performance from 70 TOPS to 2,000 teraflops, enabling developers to address virtually any edge AI workload across different price points and power envelopes 1

1

. This range allows companies to select the appropriate computing power for their specific physical AI applications without overprovisioning hardware.

Jetson Agent Skills Reduce Memory Usage by Up to 15GB

NVIDIA introduced new Jetson agent skills that automate memory optimization across the entire software stack, delivering significant cost savings for developers. These AI-powered tools support the entire Jetson portfolio, including both Jetson Thor and Jetson Orin modules, enabling developers to run more capable workloads on lower-memory configurations 1

1

.

Humanoid robotics leaders including UBTech and Agile Robots, along with industrial solutions provider Connect Tech, have reduced memory usage by up to 15GB, enabling them to move from NVIDIA Jetson AGX Orin 64GB to the 32GB module 1

1

. In smart retail, SandStar reduced memory usage by up to 4GB, enabling deployment on the NVIDIA Jetson Orin NX 8GB module instead of the 16GB configuration 2

2

. The result is lower system cost, faster deployment, and the flexibility to move down one memory SKU within the same product tier without compromising performance 1

1

.

In intelligent transportation, NoTraffic reduced memory usage by 30% on Jetson TX2 NX, creating headroom to add more AI capabilities into its smart traffic platform without increasing hardware requirements 1

1

. These optimizations matter as memory prices remain elevated, and the ability to achieve similar performance on lower-cost hardware configurations directly impacts the economics of mass-market robot deployment.

Cosmos 3 Edge Brings World Foundation Models to Thor Platforms

NVIDIA expanded its Cosmos 3 frontier open world foundation model family with Cosmos 3 Edge, a lightweight 4-billion-parameter model compatible with NVIDIA Thor platforms 1

1

. This model helps embodied systems see the world, reason over it in real time, and predict and generate actions through on-device inference. Using the open Cosmos framework, developers can post-train Cosmos 3 Edge for specific embodiments and sensors in about a day, closing the sim-to-real gap before deploying on physical AI applications 1

1

.

The integration of Cosmos 3 Edge with the Thor platform demonstrates how industrial automation and robotics companies can leverage pre-trained foundation models and customize them quickly for specific use cases. Partners providing Thor-based solutions include ADLINK, Advantech, AAEON, Aetina, Auvidea, AVerMedia, Connect Tech, ForeCR, JWIPC, NEXCOM Robotic Solutions, Realtimes, Seeed Studio, Twowin, TZTEK, and YUAN 2

2

. This ecosystem approach accelerates time-to-market for companies developing autonomous machines across healthcare, manufacturing, logistics, and retail sectors.