NVIDIA Introduces New Jetson Thor Computers to Advance Mainstream Robotics and Edge AI

As general-purpose robots and autonomous machines transition from the controlled environments of research laboratories to the complexities of mass-market deployment, the industry faces a critical bottleneck: the need for compact, energy-efficient AI supercomputers capable of executing foundation models directly at the edge. Addressing this challenge, NVIDIA has unveiled the T3000 and T2000, two new modules built upon the powerful NVIDIA Thor architecture, designed to scale robotics and edge AI applications for the real world.

The Jetson AGX Thor platform is rapidly becoming the backbone for the next generation of humanoid and robotic systems. A growing roster of industry leaders—including 1X, Agile Robots, Amazon Robotics, Boston Dynamics, FANUC, Hitachi, and Techman Robot—are already leveraging this architecture to power their latest innovations.

Unlocking Humanoid and Robotics Deployment with T3000

At the heart of this new hardware rollout are the Jetson and IGX T3000 modules. These units deliver a staggering 865 FP4 teraflops of AI compute power, all within a form factor that is roughly half the size and power consumption of the existing T5000.

The T3000 is engineered for high-performance demands, featuring:

• NVIDIA Blackwell GPU architecture.
• An eight-core Neoverse Arm CPU.
• 32GB of LPDDR5X memory with 273GB/s of bandwidth.
• 25 GbE connectivity for high-speed data transfer.

For environments requiring stringent safety protocols, the IGX T3000 offers identical performance while integrating functional safety features. It is designed to run the NVIDIA Halos for Robotics full-stack safety system, ensuring that robots can operate securely alongside human workers. Despite its reduced footprint, the T3000 maintains inference performance parity with the T5000, effectively handling multimodal workloads such as large language models (LLMs), vision-language models, and world foundation models. By migrating to the T3000, developers can significantly optimize costs in an era of fluctuating memory prices.

Expanding Edge AI Capabilities with T2000

While the T3000 targets high-end robotics, the Jetson T2000 is designed to bring the power of the Thor architecture to a broader spectrum of edge AI systems. Delivering 400 FP4 teraflops and equipped with 16GB of memory, the T2000 serves as an accessible entry point for developers focused on:

• Visual AI agents.
• Autonomous mobile robots (AMRs).
• Industrial manipulators.
• Intelligent machines requiring localized processing.

With the addition of these new modules, NVIDIA’s edge AI platform now offers a comprehensive, scalable range of performance—stretching from 70 TOPS to 2,000 teraflops—allowing developers to tailor their hardware choices to virtually any edge workload.

Automating Efficiency with New Agent Skills

To further boost developer productivity, NVIDIA has introduced new Jetson agent skills. These tools automate complex tasks such as memory optimization, system configuration, and deployment—processes that previously demanded extensive manual intervention and specialized domain knowledge.

"With the newly released Jetson agent skills, developers can optimize the entire software stack and achieve significant memory savings in days instead of weeks."

These skills are compatible across the entire Jetson portfolio, including the Thor and Orin series. By enabling developers to run sophisticated workloads on lower-memory configurations, these tools reduce system costs and accelerate time-to-market. Real-world results are already promising:

• UBTech and Agile Robots have reduced memory usage by up to 15GB, allowing them to transition from the 64GB Jetson AGX Orin to the 32GB module.
• SandStar achieved a 4GB memory reduction, enabling deployment on the Jetson Orin NX 8GB module.
• NoTraffic successfully reduced memory usage by 30% on the Jetson TX2 NX, creating the necessary headroom to integrate additional AI capabilities into their traffic management platforms.

Introducing Cosmos 3 Edge

NVIDIA has also expanded its Cosmos 3 frontier open-world foundation model family with a lightweight version tailored for the Thor platform. Cosmos 3 Edge is a 4-billion-parameter model that allows embodied systems to perceive their surroundings, reason in real time, and generate actionable insights through on-device inference. Using the open Cosmos framework, developers can post-train this model for specific sensors and embodiments in approximately one day, effectively bridging the gap between simulation and real-world performance.

Seamless Development and Future Availability

NVIDIA is ensuring a smooth transition for developers by offering a unified software stack across the Thor family. Developers can begin prototyping today using the Jetson AGX Thor developer kit, which supports emulation for both the T3000 and T2000 modules.

• T3000 Emulation Mode: Available later this month via JetPack 7.2.1.
• T2000 Emulation Mode: Scheduled for a future software release.
• Hardware Availability: The Jetson T3000 and T2000 modules are slated for release in Q1 2027.

A robust ecosystem of partners—including ADLINK, Advantech, AAEON, and Connect Tech—is already preparing Thor-based solutions. As physical and embodied AI move toward mainstream adoption, these new NVIDIA computers provide the scalable, high-performance foundation required to bring the next generation of intelligent machines into the real world. Developers are encouraged to visit the NVIDIA marketplace to secure their developer kits and begin building the future of robotics today.