AMD Takes On NVIDIA in Robotics With New Kria Module and X100 Chip

Key points

• AMD launched the Ryzen AI Embedded X100 series, a new system-on-a-chip designed for robotics, in early 2025.
• AMD claims the Kria AI system-on-module delivers 3.4 times better real-time performance than NVIDIA's Thor T5000 chip.
• Foundation Robotics, a humanoid robot developer, is migrating from Intel and NVIDIA hardware to AMD's X100 platform.
• The Kria module uses an open industry standard connector called COM-HPC, meaning other manufacturers can build compatible hardware without AMD's permission.
• AMD's robotics partner network already includes Open Robotics, OpenCV, and sensor suppliers.

AMD has a well-known name in PC processors, but its latest announcement targets a very different machine: robots. The company this week unveiled the Kria AI Robotics platform and the Ryzen AI Embedded X100, a system-on-a-chip (an SoC, meaning a single chip that combines processor, graphics, and AI compute into one piece of silicon), aimed squarely at factory robots, autonomous mobile robots, and humanoid machines.

NVIDIA currently dominates this corner of the market. AMD is positioning the X100 as a direct alternative.

What does the X100 chip actually do differently?

The key change is memory. Traditional robot computers use a main processor and a separate graphics chip, each with their own memory pool, constantly passing data back and forth. The X100 puts the central processor (CPU), graphics processor (GPU), and AI accelerator (NPU) on one chip, sharing the same memory. That cuts the delays caused by shuffling data between chips.

For a robot, those delays matter. Perceiving an obstacle, fusing data from multiple sensors, and deciding how to move all happen in milliseconds. Shaving microseconds off each step adds up.

AMD says the chip hits an interrupt latency (the delay before a chip reacts to a signal) below seven microseconds at very high reliability levels. For the hardest real-time tasks, it recommends pairing the X100 with a Zen hypervisor and FreeRTOS, two software tools that isolate critical tasks from everything else running on the chip.

How does AMD compare to NVIDIA on paper?

AMD's numbers, presented at the launch, are pointed. According to the company, the Kria AI module delivers:

Metric	AMD claim vs. NVIDIA Thor T5000
Real-time performance	3.4x better
Spare compute headroom	1.6x more
Agentic AI capacity	2.3x more
FP32 signal-processing compute	3x better than NVIDIA Thor

Those figures come from AMD, not independent testers, so treat them as a starting point for evaluation rather than settled fact.

What is the Kria module, and why does the open standard matter?

The Kria AI SoM (system-on-module, a small self-contained board that slots into a larger device) is built on the X100 chip. It uses a connector standard called COM-HPC rather than a proprietary one AMD controls. That means a robot maker is not locked into buying hardware only from AMD.

"We don't want to have a proprietary form factor," said KV Thanjuvar Bhaaskar, robotics lead at AMD. Any company on the planet can build a compatible carrier board using publicly available schematics.

A development kit ships with the module, including high-speed camera connectors, an inertial measurement unit (a sensor that tracks movement and orientation), Wi-Fi, Bluetooth, and a small on-board FPGA (a reprogrammable chip used for real-time safety and sensor tasks).

What should readers watch for?

AMD's performance claims need independent confirmation before robot makers stake production lines on them. Watch for third-party benchmarks from integrators in the Robotics Partner Network. Foundation Robotics' migration from NVIDIA to AMD X100 will be an early real-world test. If that goes smoothly, expect more humanoid developers to take a serious look.

As first reported by The Robot Report, Castec International is also using the platform for a semiconductor factory robot, which gives AMD a concrete industrial reference point beyond the press release.