AMD Bets on Open Standards and Real-Time Control with Kria AI Robotics Platform

Advanced Micro Devices Inc. is making a direct push into robotics compute. The company announced the Ryzen AI Embedded X100 series and the Kria AI Robotics platform, positioning the combined hardware and software stack as an alternative for developers building physical AI systems — one it claims can outrun NVIDIA on real-time performance.

Unified silicon for robot workloads

The Ryzen AI Embedded X100 series is built on a unified CPU–GPU–NPU architecture and is designed for high-end robotic and edge AI applications. AMD optimized the X100 series specifically for physical AI applications, and the line targets both "firm" and "hard" real-time control through BIOS and Linux optimizations. For firm real-time behavior, AMD cites an interrupt latency target below 7 μs at six-nines — a figure that will matter to engineers closing tight control loops on the same silicon that runs perception and planning.

For hard real-time requirements, AMD points to virtualization using the Zen hypervisor and a FreeRTOS virtual machine, consolidating workloads that would otherwise require separate devices.

The Kria module: COM-HPC, not proprietary lock-in

On top of the X100 sits the Kria AI system-on-module, which AMD is positioning as the foundation for robotics development. The SoM offers a 1.5 μs control-loop closure and — notably — adopts COM-HPC instead of a proprietary module. That choice is deliberate.

“What makes this [Kria] SoM differentiated is tying it back to the feedback [from customers],” said KV Thanjuvar Bhaaskar, Robotics Lead and Senior Manager at AMD.

The accompanying Kria AI robotics development kit includes a robotics carrier card with GMSL, high-speed Ethernet, IMU, and Wi-Fi/Bluetooth connectivity. An integrated FPGA on the baseboard handles real-time I/O, sensor fusion, and safety features. Relative to consumer-grade parts, the Kria platform offers extended reliability, burn-in, test coverage, and fit rates appropriate for industrial applications.

Performance claims against NVIDIA

AMD is not being subtle about the competitive target. The company claims the Kria AI SoM delivers 3.4x better real-time performance than NVIDIA Thor T5000, along with 1.6x spare compute capacity and 2.3x more agentic AI capacity.

Rob Bauer, Senior Manager of Product Management and Marketing for the x86 Embedded APU portfolio at AMD, extended the comparison into signal-processing territory:

“In that market [aerospace and defense market, with a focus on signal-processing workloads], we’re actually delivering three times the FP32 compute performance relative to NVIDIA Thor,” Bauer said.

Software and ecosystem

The Kria system supports ROS 2 / Nav2 acceleration and is compatible with current robotics workflows, while AMD ROCm serves as a common AI stack from cloud to robot. AMD Robotics Sophie Suite rounds out the open-source software offering.

To build out the ecosystem, AMD is assembling a Robotics Partner Network of sensor and software companies, including Open Robotics, Open Navigation, and OpenCV. Early customers already named include Castec International and Foundation Robotics.

The open-standard strategy is the real story here. Whether the performance claims hold up in independent benchmarking remains to be seen, but AMD is clearly wagering that robot builders value avoiding vendor lock-in as much as they value raw compute.