Foundation picks AMD X100 to power Phantom humanoid robots

AMD just got its first flashy humanoid reference customer, and it is a pointed one. Foundation Future Industries says its Phantom humanoid robots will run on AMD's Ryzen AI Embedded X100 Series instead of Nvidia silicon, per Forbes. Foundation's own claim is that the AMD parts deliver 2.5x faster inference and 3x faster training than Nvidia. Take that as a vendor-supplied number, not a settled benchmark.

The more interesting technical detail is what AMD is bringing beyond a CPU-GPU-NPU sandwich. The Xilinx acquisition gave AMD FPGA fabric, and Foundation is using it to run a deterministic control loop for the robot's 23-degree-of-freedom hands, fusing high-frequency tactile feedback in a single loop. That is the part of a humanoid stack where Nvidia's Jetson lineage is thinnest: real-time motor control at the millisecond level is not what a GPU was born to do. If AMD can credibly say "we do the reasoning and the reflexes on one platform," that is a genuinely differentiated pitch to robotics teams.

Why it matters if you are not building humanoids: the physical-AI compute market has, until now, been treated as basically Nvidia's to lose. A design win in a defense-adjacent humanoid, with a Phantom MK-2 variant rated to IP67 and 100G vibration tolerance for industrial and defense customers, is exactly the kind of lighthouse deal AMD needs to be quoted in every other robotics RFP for the next twelve months.

The honest caveat is that Foundation is the one making the speed claims, and the reporting does not give you the benchmark methodology, the Nvidia SKU that lost, the model sizes, or the shipped robot count behind the "24/5 at customer sites" line. Foundation also carries its own political baggage that could color how enterprise and government buyers read this win. Treat the specifics as reported, not proven.

What is worth watching is whether any other named humanoid program follows Foundation over to AMD in the next few quarters. One design win is a story. Two is a trend Nvidia has to answer.