Advanced Micro Devices (NASDAQ:AMD) is repositioning itself for a new phase of artificial intelligence infrastructure, telling investors that the rise of agentic AI workloads is creating a surge in demand for server CPUs just as the company pushes beyond selling individual chips toward delivering complete rack-scale systems.
At the Goldman Sachs Communacopia + Technology Conference, senior executives laid out a vision in which AMD acts as a high-performance computing partner capable of supplying everything from endpoint devices to massive AI data centers. The shift reflects a broader change in how AI systems are being architected, with CPUs taking on a more central role than they did during the initial wave of GPU-centric AI deployments.
Dan McNamara, senior vice president and general manager of Compute and Enterprise AI, said agentic AI represents a structural departure from earlier prompt-response inference applications. Those earlier systems followed a largely linear path through web servers, application servers, databases, storage and GPU servers. Agentic systems, by contrast, involve continuous activity across multiple agents executing control-plane operations, API calls, database queries and tool execution.
“When you can drive a faster time to market, that’s where the real rubber hits the road,” McNamara said, describing how AMD deploys its own technologies internally before offering them to enterprise customers.
Those agentic tasks are predominantly CPU-based, he said, generating demand for a new class of processors designed specifically for agentic environments as well as traditional general-purpose servers that support increased system activity. AMD is targeting three deployment areas with its upcoming Venice portfolio: high-core-count processors for agentic AI, including a 256-core device; high-frequency processors for head-node workloads that keep GPUs supplied with work; and general-purpose server processors for broader enterprise and cloud use.
The company’s view of the CPU opportunity has expanded dramatically. Matt Ramsay, corporate vice president of financial strategy and investor relations, said AMD has raised its estimate of the server CPU total addressable market from $60 billion to $120 billion and then to $220 billion through 2030. While he did not provide a detailed market share forecast, he said the company expects its server business to address more than half of that market.
That aligns with comments from CEO Lisa Su, who has previously noted that the ratio of GPUs to CPUs in servers used to be 4:1 or even 8:1, but is now shifting toward roughly 1:1 as AI agents require more general-purpose processing power. McNamara echoed that view, saying the CPU-to-GPU ratio is growing from what he characterized as approximately one-to-one today, though he cautioned that CPU content in a gigawatt-scale AI deployment cannot be expressed as a fixed ratio because it varies by customer workload.
MetricPrevious EstimateCurrent EstimateServer CPU TAM$60 billion$220 billionData center CAGR target–>60%AI business CAGR target–>80%Total AI TAM by 2030–~$2 trillionData center revenue by 2027–More than double
Note: Figures reflect AMD management commentary at the Goldman Sachs Communacopia + Technology Conference.
Beyond CPUs, AMD is emphasizing its evolution from a component supplier to a systems and software company. McNamara said the company has shifted from managing individual product roadmaps to coordinating a data-center roadmap spanning CPUs, GPUs and networking, with the principal focus now on delivering rack-scale solutions. 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.
The company is scaling its AI business through rack deployments, including Helios systems and MI455 products. Ramsay reiterated growth targets previously outlined by management, including more than 60% compound annual growth for the data-center business and more than 80% for the AI business over the company’s strategic timeframe. AMD has described a total addressable market of roughly $2 trillion by 2030, representing a 40% growth rate, and expects to grow faster than that market.
AMD is also positioning itself as a significant internal user of AI. McNamara said the company began by optimizing its data layer for AI agents and open-sourced a solution called Optima. The company is now applying AI to domain-specific work including electronic design automation, coding, debugging, kernel development and broader software development. Ramsay said AMD has adopted tools from Anthropic and OpenAI across its engineering organizations to accelerate software development, debugging and chip-program production. He acknowledged higher token costs but said the productivity gains have been larger.
On the competitive landscape, McNamara downplayed the significance of the x86 versus Arm instruction-set architecture debate, describing the competition as primarily a matter of optimization. The x86 software ecosystem remains important for the large installed base of general-purpose server workloads, he argued. AMD sees strong demand for its current Turin products and expects Venice to have the company’s broadest launch to date across original equipment manufacturers, original design manufacturers, cloud providers and independent software vendors.
For the AI model landscape, Ramsay said the market will likely include frontier models, open-weight models and smaller language models tailored to specific tasks. AMD’s objective 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, with policy-based routing directing workloads based on performance, latency, security and cost considerations.
“Enterprises are infinitely hybrid, and we believe that will continue,” McNamara said.
The strategic messaging comes at a time when AMD’s shares have faced pressure despite strong financial results. The company reported a 50% jump in total sales to $11.5 billion in the second quarter, with data center revenue surging 107% to $6.7 billion. Non-GAAP earnings increased 245% to $1.66 per share, outpacing consensus estimates. Yet the stock has trended lower over several months, partly reflecting a trailing price-to-earnings ratio of 117 that some investors consider too rich relative to the tech sector average of 32.
AMD is not alone in facing skepticism about AI infrastructure spending. Broadcom (NASDAQ:AVGO) has also seen its shares pull back despite reporting an 86% sales increase to $29.6 billion in its latest quarter. Broadcom management has projected that its AI sales will double in 2027 to reach $100 billion, and Counterpoint Research estimates the company will hold roughly 60% of the ASIC market by the end of next year.
The broader spending environment suggests demand remains robust. Research from S&P Global indicates the world’s largest AI companies will spend more than $1 trillion next year on AI infrastructure, up from $750 billion this year. That backdrop supports the argument that current pullbacks in chip stocks may reflect valuation concerns rather than deteriorating fundamentals.
Looking ahead, McNamara said investors may increasingly view AMD as a software and systems company as it expands its rack-scale capabilities and develops a software stack intended to shorten customers’ time to value. Ramsay said management remains focused on execution as it seeks to scale data-center revenue and grow gross-margin dollars faster than expenses.
The message from AMD’s leadership is clear: the AI infrastructure opportunity is broadening beyond GPUs, and the company intends to capture a larger share of the expanding CPU and systems market as agentic workloads reshape data-center architecture.