Photo: Grujoungarot / Wikimedia Commons / CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0) AMD's new Kria AI platform combines CPU, GPU, NPU, and FPGA in one package, taking direct aim at Nvidia's Jetson Thor dominance in edge robotics
On July 23, 2026, at its Advancing AI 2026 event in San Francisco, AMD launched the Kria AI system-on-modules and a full Robotics Developer Platform built around the new Ryzen AI Embedded X100 Series processors.
The X100 platform folds a CPU, GPU, NPU, and FPGA into a single unified memory architecture, meaning each compute element shares the same memory pool instead of passing data back and forth across separate chips.
AMD is comparing its platform directly against Nvidia’s Jetson Thor T5000. The company claims a 3.4× advantage in real-time processing capabilities, 2.3× more concurrent agents running simultaneously, and 1.6× more spare CPU capacity left over after those workloads run. It also claims 3× improved FP32 performance in specific signal-processing tasks.
The platform runs on the COM-HPC standard, an open hardware spec that allows developers to swap out modules without redesigning entire systems. AMD is also building its software stack around ROS 2, plus ROCm, AMD’s GPU compute framework. Schematics are publicly accessible.
The company announced a Robotics Partner Network alongside the hardware launch. The most concrete early adopter is Foundation Future Industries, which is transitioning its Phantom humanoid robots to the X100 platform. Foundation Future Industries was previously working with Nvidia and Intel setups.
AMD has supplied FPGAs and adaptive SoCs through its Xilinx acquisition for years, often quietly powering control systems inside industrial automation equipment. The Kria brand itself already existed as an FPGA-based embedded platform. What changed is the integration level, combining all the compute types developers need under one roof and giving it a proper software stack for the AI era.
For developers, the ROCm software ecosystem, while improving, still trails CUDA in terms of tooling maturity. That gap is the main risk AMD has to close. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our