The AI infrastructure race is accelerating, but market focus has shifted beyond just GPU performance toward a re-evaluation of the entire computing ecosystem.
A July 23 report from Barclays argues that the most significant signal from Advanced Micro Devices' Advancing AI event was not its new GPU or AI server, but a dramatically increased long-term forecast for the server CPU market.
Advanced Micro Devices now expects the global server CPU market to reach $220 billion by 2030, nearly double its previous estimate of $120 billion. This revision is driven by the rise of Agentic AI, which is transforming the CPU from a supporting role in AI servers into a core computing resource.
Building on this, Advanced Micro Devices has also raised its outlook for the broader computing market. The company now forecasts the data center AI accelerator market will hit $1.4 trillion by 2030, while its total addressable compute market, spanning server CPUs and AI accelerators, is expected to grow from $365 billion in 2025 to approximately $2 trillion.
Barclays believes this indicates that AI infrastructure competition is evolving from a focus on single GPU performance to a full-stack battle encompassing CPUs, GPUs, networking, software, and system architecture.
Server CPU Market Forecast Significantly Raised
The most market-moving aspect of the event was not the Helios rack or new GPU, but Advanced Micro Devices' redefinition of the future server CPU market size. The company now projects a $220 billion market by 2030, with the compound annual growth rate jumping from over 35% to more than 50%. Barclays views this aggressive upgrade as a clear signal of a fundamental shift in Advanced Micro Devices' assessment of future AI server architecture.
The core logic supporting this forecast is the rapid development of Agentic AI. Advanced Micro Devices expects that "Agentic CPUs" designed for agent workloads will account for more than half of the total server CPU market. As more AI applications move from training to actual deployment, CPUs will take on more responsibilities, including model scheduling, agent execution, task coordination, and data management. They will no longer be mere supporting components in GPU servers, but will become an independent and significant growth market.
Inference Becomes the New Focus of AI Infrastructure Competition
Beyond the market size re-evaluation, Advanced Micro Devices' assessment of AI workload structure also caught Barclays' attention. The company predicts that inference will become the largest AI computing scenario for the first time in 2026, representing about 60% of all AI workloads, up from 50% in 2025 and 40% in 2024. This suggests the industry competition is shifting from "who can train the largest model" to "who can run models at the lowest cost and highest efficiency."
Barclays believes this explains why Advanced Micro Devices simultaneously launched the Helios platform, the MI455X GPU, Ethernet interconnect, and the ROCm.AI software platform. Future AI infrastructure competition will no longer be limited to GPU compute power, but will revolve around the entire system's cost, memory capacity, bandwidth, network interconnect, and software ecosystem.
The report notes that the entire industry remains in a state of tight compute supply. With major cloud providers continuing to expand their deployments and existing customers entering volume production phases, Advanced Micro Devices' data center business is expected to grow faster than the overall market. However, the event did not announce new major customers, as the Anthropic partnership was previously disclosed.
From Chip to Platform: AMD Refines Its Product Roadmap Through 2030
Supporting its new outlook, Advanced Micro Devices has further refined its product plans for the coming years. On the hardware side, the company officially launched the Helios AI rack, integrating the MI455X GPU, Venice CPU, Salina DPU, and Vulcano AI NIC into a complete system solution. The MI455X features 432GB of HBM4 memory with 23.3TB/s bandwidth and supports up to 40 PFLOPS of FP4 compute power. Compared to the previous generation, Advanced Micro Devices expects up to a 34x improvement in token throughput and an 18x reduction in token cost.
Advanced Micro Devices is also focusing its competitive efforts on system efficiency. The company predicts that, compared to Nvidia's Vera Rubin platform, Helios will offer not only larger HBM capacity and higher lateral interconnect bandwidth but also higher token output per dollar, indicating that the competitive dimension is shifting from single GPU performance to the total cost of ownership (TCO) of the entire system.
Simultaneously, the company has extended its CPU and GPU roadmap to 2030. The Venice processor has entered production, with deployment in cloud platforms and OEM servers expected in the fourth quarter. The Florence and Ravenna CPU generations are planned for 2028 and 2030, respectively. On the GPU side, the MI500 is expected in 2027, followed by the MI600 in 2028. The MI500 will be the first to simultaneously use both copper and optical interconnect technologies.
Beyond hardware, Advanced Micro Devices also launched the ROCm.AI development platform and continues to expand its product portfolio around inference, networking, and robotics ecosystems. This includes a decoupled inference solution in partnership with Cerebras, a dedicated inference card, the MI350P, and the KRIA AI System-on-Module for robotics, aiming to further build out its full-stack AI infrastructure capabilities.
Barclays believes that what truly changed market expectations at this Advancing AI event was not specific product performance, but Advanced Micro Devices' redefinition of future computing demand and its latest assessment of AI infrastructure evolution. If inference demand continues to accelerate as the company predicts, the importance of server CPUs and system-level solutions could significantly surpass previous market expectations.
Comments