AMD has been hammering the physical AI message and did again so during its 2025 financial analyst day on Tuesday. The impact of humanoids and robotaxis on the world of sensors will be profound, as Intel, Nvidia, Tesla and many others have noted in the past year.
The embedded division at AMD (which is responsible for physical AI and edge technologies) is not its largest revenue segment (especially when compared to data center) and actually declined 8% year-over-year in the latest third quarter report to $857 million. Even so, it represents an area of long-term growth, especially as big competitors gear up for advanced edge technologies in the next decade.
CEO Lisa Su noted at earnings on Nov. 4 that embedded revenue did increase quarter-to-quarter as demand strengthened across multiple markets including emulation testing, aerospace and defense and industrial vision and healthcare. “The design momentum remains very strong across our embedded portfolio,” she said on the earnings call.
In a summary report on analyst day, AMD called its embedded business a “powerhouse in embedded and semi-custom silicon solutions—positioning AMD to lead the next wave of intelligent, connected systems.” This embedded business includes the FPGA businesses of Xilinx, acquired in 2022. In its third quarter earnings report, AMD also highlighted an expanded x86 embedded processor portfolio including EPYC Embedded 4005 series for network security appliances and entry level industrial edge servers. Also, AMD Ryzen Embedded 9000 processors were recently announced for low latency, high performance per watt for machine vision, industrial PCs and automation systems, which Su said includes robotics, edge computing and smart factory applications. “Design momentum remains very strong across our embedded portfolio,” Su said at earnings time.
AMD projects a $200 billion total addressable market by 2035 globally and the company declared it is “ready to lead in physical AI,” in a slide presentation given to financial analysts by Salil Raje, general manager of adaptive and embedded computing.
“The scope of AMD embedded has increased tremendously over the last three years,” he told analysts. AMD has “thousands of loyal embedded customers.”
“As AI expands from the cloud to the physical world, billions of intelligent devices will shape the next wave of embedded computing,” noted one of his slides, then another listed a series of AMD silicon devices for work in digital twin, vision AI/smart camera, motor drive and predictive maintenance, lidar and real-time sensor fusion, perception and multiview camera, real-time industrial networking and robot controller and generative AI. AMD lists partners such as ABB, Intuitive, Robotec AI, Sick and Subaru. These silicon chips work across perception, decision-making at inference and action, with real-time control.
“With the FPGA business, AMD is in a position to use the FPGAs as AI accelerators for embedded systems and autonomous devices like robotics and machines tools,” said Jack Gold, president of J. Gold Associates, in an interview with Fierce. “FPGAs can be custom configured to run specific models efficiently as opposed to general purpose GPUs. And given their GPU capabilities, AMD is also in a position to supply for vision and imaging systems like medical and also for robots that need to be able to see where they are going.”
AMD faces “stiff competition” from the ARM camp as AMD still is x86 centric, Gold added. “That’s not a huge negative but it’s a competitive arrier where small, power efficient systems are required.”
Nvidia has a clear lead in GPUs for AI training and has a strong focus on work with associated systems for inference and connections to physical AI. In September, Intel laid out a number of applications for its edge processor, code-named Panther Lake.
“Since acquiring Xilinx, AMD has executed well in embedded,” said Austin Lyons, analyst at Creative Strategies. “With a full suite of compute, connectivity and security IP, AMD is well-positioned for robotics and physical AI, especially as it exploures semi-custom system leveraging its chiplet and advanced packaging expertise. And don’t forget AI software for these emerging use cases, which they continue to invest in. AMD will make a nice contender.”
Summit Research noted that AMD expects its embedded segment to expand at 10% annually over the next three to five years, representing a “nascent recovery” following cyclical declines over the past two years. But Summit added this silver lining: “The embedded segment is highly margin accretive and sustained long-term growth on this front will be especially favorable to AMD’s long-term earnings target.”
For developers creating applications that rely on sensors and processors, the emergence of edge processing from a strong of group of vendors like AMD should offer plenty of promising alternatives.