AMD Expands Into Rack-Scale AI as Agentic Workloads Boost CPU Demand

Advanced Micro Devices NASDAQ: AMD executives said the company is expanding its role in artificial intelligence from a semiconductor supplier toward a provider of rack-scale systems and software, while positioning its server CPU portfolio for increased demand from agentic AI workloads.

Speaking at the Goldman Sachs Communacopia + Technology Conference, Dan McNamara, AMD’s senior vice president and general manager of Compute and Enterprise AI, said AMD has also become a significant internal user of AI. The company began by optimizing its data layer for AI agents and open-sourced a solution called Optima, he said.

McNamara said AMD is now applying AI to domain-specific work, including electronic design automation, coding, debugging, kernel development and broader software development. He also cited automation and efficiency improvements across the company’s operations.

“When you can drive a faster time to market, that’s where the real rubber hits the road,” McNamara said, adding that AMD deploys its own technologies in its data centers before bringing them to external enterprise customers.

Matt Ramsay, AMD’s corporate vice president of financial strategy and investor relations, said the company has adopted tools from Anthropic and OpenAI across its engineering organizations to accelerate software development, debugging and chip-program production. He said AMD has incurred higher token costs but has seen larger productivity gains.

On the outlook for AI models, Ramsay said the market will likely include frontier models, open-weight models and smaller language models tailored to specific tasks. AMD’s objective, he said, is to provide CPU and GPU products that deliver favorable token-per-dollar economics across those use cases.

McNamara said enterprises are expected to use hybrid architectures, combining cloud-based frontier models with on-premises systems running open-weight models. He described a policy-based routing approach in which workloads could be directed based on performance, latency, security and cost considerations.

“Enterprise are infinitely hybrid, and we believe that will continue,” McNamara said.

McNamara said AMD has shifted from managing individual product roadmaps to coordinating a data-center roadmap spanning CPUs, GPUs and networking. The company’s principal focus is now delivering rack-scale solutions, he said.

Ramsay said AMD can approach major computing customers as a broader high-performance-computing partner rather than solely as a supplier of a CPU, GPU, FPGA or semi-custom component. That approach can extend across endpoint devices, edge inference, on-premises servers, cloud deployments, large-scale AI data ce...