Victor Peng on interoperability
From AI Chip Competition with Victor Peng, President, AMD · · Stanford Graduate School of Business
“You need to interoperate with CPUs and GPUs and any one of those could be your bottleneck to the end delivered performance, and also the software that enables that end-to-end performance with AI. What you have to be careful of is blockages in interoperability and the lack of customer choice at different levels of the stack, both the software and the hardware.”
On , Victor Peng, President of AMD at Advanced Micro Devices, spoke about interoperability during AI Chip Competition with Victor Peng, President, AMD on Stanford Graduate School of Business.
About Victor Peng
Victor Peng, President of AMD, has been a prominent spokesperson for the company's AI strategy, emphasizing a three-word approach: "Open, Proven and Ready." In various appearances in 2023 and 2024, Peng discussed AMD's focus on providing customer choice through open-source software and hardware interoperability, describing the company's approach as building a "bridge over the moat" that competitors had. He highlighted AMD's participation across the full AI stack, from data center GPUs (MI Instinct) and server CPUs (EPYC) to embedded and client devices, and noted the company's work with partners like Microsoft and other cloud providers. Peng stated that AMD's culture of collaboration and listening to customers positions the company to evolve with changing algorithms and use cases. Peng has also discussed the broader AI chip market, explaining that GPUs are purpose-built for training large language models and differ from general-purpose CPUs. He addressed AMD's fabless model, noting the company outsources manufacturing to foundries and works with partners to mitigate geographic concentration risks. In earlier appearances as CEO of Xilinx, Peng described the company's transformation from a device company to a platform company, introducing the adaptive compute acceleration platform (ACAP) architecture. He has spoken about the importance of modular design, such as AMD's chiplet approach, and the need for interoperability across CPUs, GPUs, and software stacks to avoid bottlenecks in AI performance.