AMD Paves Open Path for Robotics: Chips, Modules, Platforms, Ecosystem Target NVIDIA Jetson

At the Advancing AI 2026 conference in late July, AMD put 'Physical AI' on center stage. While data center chips dominated the spotlight, the real news was in the embedded products unveiled: the Ryzen AI Embedded X100, Kria AI SOM, and the Kria robotics development platform. This marks AMD's first complete robotics foundation—spanning chips, modules, software, and an ecosystem network.

The X100 series is built on Strix Halo, the highly integrated silicon that has shone in laptops and mini PCs over the past year. AMD has adapted it for embedded use with two key changes: industrial-grade certification for temperatures from -40°C to 105°C, and real-time firmware that cuts interrupt latency to under 7 microseconds in Linux.

The top-tier X199 features 16 Zen 5 cores, 40 RDNA 3.5 compute units, and a 256-bit LPDDR5X memory bus. For robotics running large models, the bottleneck isn't raw compute but memory bandwidth—this wide bus is the chip's key advantage.

The X100 is also part of AMD's 10-year life cycle promise, ensuring stable supply for years. Industrial equipment makers fear sudden chip shortages, so this long-term commitment provides crucial confidence.

Beyond the chip, AMD turned the X100 into the Kria AI SOM, a system-on-module that marks Kria's leap from Arm-based programmable SoCs to x86 high-performance processors. The module adheres to the COM-HPC open standard, minimizing friction for developers integrating it into their devices—a classic AMD strategy of using open standards to differentiate.

It's clear this board targets NVIDIA's Jetson directly. AMD claims superior performance, though RDNA 3.5 lacks tensor cores, imposing a hard ceiling on compute throughput. Still, the direction is unmistakable.

AMD also introduced the Kria AI Robotics development platform—a carrier board that combines the X199 module, FPGA, and various interfaces into a ready-to-use dev kit. This directly competes with NVIDIA's Jetson AGX Thor dev kit. AMD's approach is pragmatic: selling modules alone isn't enough; developers need a complete reference design to start working immediately.

On the ecosystem front, the Kria AI SOM won't be solely manufactured and sold by AMD. Instead, AMD sets the design and standards, with certified partners allowed to sell under the Kria brand. This keeps AMD in the standard-setting role while enabling partners to scale volume. It's a strategy of 'I won't compete with you for complete machines—I lay the foundation, standards, and toolchains; you build the house.'

This contrasts with two approaches to robotics. The international 'foundation' approach, exemplified by AMD and NVIDIA, layers computing, memory, software, standards, and partner networks, planning on a decade scale. The domestic Chinese 'body' approach focuses first on building robots that can move, using real-world scenarios to validate demand and drive supply chains. The starting points are opposite.

China's advantage lies in large-scale real-world data—feedback from robots in factories and outdoor environments, which labs cannot replicate. Ultimately, foundation and body must interlock: chip makers lay the computing groundwork, and robot builders feed real-world data back into the loop. Only then does the path open up.

AMD's move focuses Physical AI on robotics, building layer upon layer from chip to module to platform to partners, and locking in trust with a decade-long commitment. It doesn't build complete robots; it lays the foundation—and this foundation points to a more open Physical AI future than NVIDIA's.