AMD announced a comprehensive approach to advancing robotics and other physical AI on Thursday that combines System-on-Chip silicon with open software designed to rival Nvidia’s CUDA along with an ecosystem of partners including nine robot makers.
“We’ve been in the body of the robot and now we are going after the brain of the robot with a complete offering,” KV Thanjavur Bhaaskar told Fierce. He serves as robotics lead at AMD’s embedded unit. “Robotics is more than AI. You have to service that entire system.”
AMD’s updated approach comes amidst a flurry of announcements in early 2026 from competitors like Nvidia and Intel that put sensor inputs like visual perception and tactile pressure center stage in robot systems that rely on AI logic to make real-time decisions.
Robots are one of the most challenging examples of physical AI because they are emerging in spaces alongside humans, some called cobots, where outcomes could prove disastrous without adequate safety designs. A false robot move during a surgery or on an assembly line comes to mind, not that different from an autonomous vehicle careening down a busy highway or plowing through a neighborhood filled with children playing.
The company has been battling its foes for more than two years by introducing its silicon in data center servers as AI agents and LLM have dominated the news. However, at its annual AAI (AMD Advancing AI) developer event Thursday, AMD CEO Lisa Su also moved to set the company’s sites clearly on physical AI with an embedded processor family that combines CPU, GPU and NPU acceleration into a single System-on-Chip to power robots, industrial automation and what it terms broadly “embedded intelligent systems” that work in aerospace and healthcare settings, among others.
IDC analyst Phil Solis remarked on the comprehensive nature of AMD’s announcement in comments to Fierce: “Overall, AMD is positioning its products for robotics to be capable, targeted, and friendly to the vast open-source robotics community.”
Called the AMD Ryzen AI Embedded AI Embedded X100 Series, the CPU, GPU and NPU processors will work together to accelerate perception and reasoning.
The seeds of the SoC were planted at CES in January. The processors integrate up to 16 AMD Zen 5 CPU cores, an AMD RDNA 3.5 iGPU and a power-efficient NPU. Unified memory provides low latency, the company said, while the entire system is designed to be rugged and meet thermal demands to prevent unsafe overheating.
The Embedded X100 will simplify development tasks and help companies move applications to market faster, AMD engineers told Fierce.
More than an SoC with Kria modules, partner network
These SoC advances will be helped with a related announcement AMD made through a platform called AMD Kria AI Solutions, a portfolio made of modules and a robotics carrier card powered by the Embedded X100 Series processors and a Kria AI robotics developer platform. The entire approach will be available to early access customers in fourth quarter 2026.
AMD also formally introduced an open robotics partner network for physical AI development that encourages partners to use open tools like the AMD ROS 2 perception node and the AMD ROCm open software ecosystem and Ryzen AI software. AMD named Analog Devices, SICK and Voyant Photonics as sensors members in that network, among a group of 30 companies in related areas of the network.
No single company will build the future of physical AI, AMD officials said. With the open concept, AMD “wants to remove vendor lock-in,” Baaskar said. “Open source helps innovation.”
The module and carrier card will support connections to various sensors. With an FPGA that expands the capabilities of the X100, “you just plug it in and it works,” he said. “There’s a ton of connectivity.” One AMD official even called ROCm “AMD’s answer to CUDA,” noting that ROCm software has been on the market for 10 years and has 100,000 users. With ROCm software, developers will be able to drop in code from CUDA to be able to stay within same development stack, AMD said.
AMD named Castec as a user of Ryzen AI x86 processors for human-machine collaboration needed in semiconductor fabs to identify thousands of different objects and respond to anomalies. Other robotics collaborators include Xaba, Rexroth, Foundation Future Industries, Universal Robots, ABB, Generative Bionics, Muks Robotics and Weston Robot.
Foundation CEO Sankeet Pathak issued a statement calling the X100 with its FPGA used in the Phantom MK-2 robot a “balanced architecture by AMD [that] gives us the optimal compute performance needed to put robots to work at scale.” He noted 2.5x faster CPU inference, equal model quality and 3x faster training when compared to Nvidia’s products.
The X100 processor family is hardened for use in harsh environments with extreme temperature ranges, allowing it to keep running at higher temperatures, IDC's Solis noted. It also uses an XDNA 2 NPU with more robust error correction with Error Correction Code memory. “It is very good for industrial applications and robots, and because they support industrial applications they will have 10 years of support,” he noted.
