NVIDIA Expands Jetson Thor Lineup with Cost-Effective T3000 and T2000 Boards
Key Takeaways
- ▸T3000 delivers ~72% of T4000 GPU performance but maintains full memory bandwidth (273GB/s), making it ideal for bandwidth-bound AI inference workloads at significantly lower cost
- ▸NVIDIA's strategic focus on memory bandwidth over capacity optimization allows the T3000 to match T5000-level inference performance on multimodal tasks while dramatically reducing price
- ▸T2000 offers the most aggressive cost reduction with half the memory bus width and 10GbE networking, serving price-sensitive applications
Summary
NVIDIA announced a major expansion of its Jetson Thor robotics and edge AI board lineup with two new mid-range models set to launch in early 2027. The Jetson Thor T3000 and T2000 are designed as lower-cost alternatives to the existing high-end T4000 and T5000 boards, directly addressing rising component costs that have been pricing out robotics and industrial AI developers. The T3000, the more powerful of the two, features 8 Arm Neoverse V3AE CPU cores, a 1536-core Blackwell iGPU, and 32GB of memory (half that of the T4000), while maintaining the same full memory bandwidth of 273GB/s. This memory bandwidth optimization means the T3000 is expected to deliver similar inference performance to the T5000 for multimodal workloads that are memory-bandwidth bound rather than memory-capacity limited.
The T2000 represents the most affordable option in the Thor lineup, with a 6-core CPU configuration, 1024 CUDA cores delivering 400 TFLOPS of sparse FP4, and just 16GB of memory on a half-width memory bus for 137GB/s bandwidth. It also steps down to 10GbE networking compared to the T3000's 25GbE. NVIDIA is specifically targeting the robotics sector with both boards, while also offering functional safety-certified IGX versions. The entire T3000 board consumes approximately 65 watts, roughly half the power of the T5000, making it an attractive option for power-constrained deployments.
- Launch scheduled for early 2027 directly addresses market demand for affordable edge AI and robotics solutions amid elevated component costs
Editorial Opinion
NVIDIA's expansion of the Jetson Thor lineup is a strategically sound response to real market pain points. By maintaining memory bandwidth while cutting capacity and GPU performance proportionally, the T3000 cleverly positions itself as a true value option for the significant portion of AI inference workloads that are memory-bandwidth bound—which covers many real-world robotics and multimodal AI scenarios. This move could substantially broaden NVIDIA's addressable market in robotics and industrial edge AI, though final pricing will ultimately determine adoption. The early 2027 timing also suggests NVIDIA is betting that component costs will remain elevated, making mid-range options increasingly attractive to developers.
