120 billion → 220 billion USD! AMD reassesses CPUs for the AI era
```
The competition for AI infrastructure continues to intensify, but the market focus is no longer limited to GPU performance. Instead, it is shifting toward a reassessment of the entire computing system.
A research report released by Barclays on July 23 believes that the biggest signal from AMD's Advancing AI conference is not the new generation of GPUs or AI servers, but a significant upward revision of its long-term outlook for the server CPU market.
The company expects that by 2030, the global server CPU market size will reach $220 billion, nearly doubling from the previous estimate of $120 billion. This reflects how, under the drive of Agentic AI, CPUs are shifting from a supporting role in AI servers to core computing resources.
On this basis, AMD has further raised its forecasts for the overall computing market. The company expects that by 2030, the data center AI accelerator market size will reach $1.4 trillion, and the total addressable compute market (Compute TAM), which includes server CPUs, AI accelerators and related businesses, will expand from $365 billion in 2025 to around $2 trillion.
Barclays believes this means that AI infrastructure competition is gradually evolving from single GPU performance to a full-stack competition covering CPUs, GPUs, networking, software, and system architecture.
Significant upward revision of the server CPU market size
The most noteworthy aspect of this event is not the Helios rack or the new generation of GPUs, but AMD's redefinition of the future server CPU market size.
The company expects that by 2030, the server CPU market size will reach $220 billion, with the corresponding compound annual growth rate rising significantly from over 35% to more than 50%. Barclays believes this aggressive revision shows that AMD has undergone a significant change in its assessment of future AI server architecture.
The core logic supporting this forecast is the rapid development of Agentic AI. AMD predicts that in the future, "Agentic CPUs" designed for agent workloads will make up more than half of the server CPU market.
As more and more AI applications shift from training to actual deployment, CPUs will take on more roles such as model scheduling, agent execution, task collaboration, and data management. They will no longer just be auxiliary components in GPU servers but will become an important independent growth market.
Inference becomes the new focus of AI infrastructure competition
Aside from the market re-evaluation, AMD's analysis of AI workload structures has also become a key focus for Barclays.
The company expects that inference will become the largest AI computing scenario for the first time in 2026, accounting for about 60% of overall AI workloads, compared to 50% and 40% in 2025 and 2024, respectively. This means that industry competition is shifting from "who can train the largest models" to "who can run models more efficiently and at lower cost."
Barclays believes this is also why AMD launched the Helios system platform, MI455X GPU, Ethernet interconnect, and ROCm.AI software platform simultaneously. The future competition in AI infrastructure will not be limited to GPU computing power but will revolve around the overall system cost, memory capacity, bandwidth, network interconnectivity, and software ecosystem.
The report points out that the industry is still experiencing a supply shortage in computing power. With major cloud providers continuing to expand deployments and current customers ramping up, the growth rate of AMD's data center business in the future is expected to exceed overall market growth. However, the conference did not announce any new major customers—partnership with Anthropic was previously disclosed.
From chips to platforms: AMD improves its product roadmap through 2030
Based on the above assessment, AMD 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, 23.3TB/s memory bandwidth, and supports up to 40 PFLOPS of FP4 computing power. AMD expects that compared to the previous generation, its token throughput can be up to 34 times higher and token cost up to 18 times lower.
AMD is also focusing competition on system efficiency. The company expects that compared to Nvidia's Vera Rubin platform, Helios not only offers larger HBM capacity and higher lateral interconnect bandwidth, but also delivers higher token output per dollar, showing that competition is shifting from single GPU performance to overall system total cost of ownership (TCO).
At the same time, the company extended its CPU and GPU roadmap to 2030. The Venice processor is in production and expected to be deployed on cloud platforms and in OEM servers starting in Q4; two more generations of CPUs, Florence and Ravenna, are scheduled for 2028 and 2030, respectively. For GPUs, the MI500 is expected in 2027, and the MI600 in 2028, with MI500 being AMD's first to use both copper and optical interconnect technology.
Beyond hardware, AMD has also launched the ROCm.AI AI development platform and continues to expand its offerings around inference, networking, and robotics ecosystems, including a decoupled inference solution with Cerebras, as well as introduction of the MI350P inference card and the KRIA AI System-on-Module for robotics, seeking to further enhance the full-stack strength of its AI infrastructure.
Barclays believes that compared to the performance of specific products, what truly changed market expectations at the Advancing AI conference was AMD's redefinition of future compute demand scale and its latest insights on the evolution of AI infrastructure. If inference demand continues to accelerate as expected, the importance of server CPUs and system-level solutions is likely to exceed prior market expectations.
Risk Disclosure and DisclaimerThe market involves risks, and investment requires caution. This article does not constitute personal investment advice, nor does it take into account any particular user's special investment objectives, financial status, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific situation. If investing based on this article, you do so at your own risk. ```