AMD Makes Strategic Power Play in AI Data Center Market

AMD Makes Strategic Power Play in AI Data Center Market Recent design wins with Meta, Open AI signal chipmaker is dead serious on carving a stronger share of market dominated by Nvidia AMD has long operated in Intel's shadow within the processor space. As artificial intelligence exploded in prominence, rival Nvidia ascended to chipmaking giant status while Intel faced setbacks. However, at its Advancing AI event in San Francisco, streamed live Thursday, AMD demonstrated it won't surrender the lucrative AI data center market without an aggressive fight. While AMD's event lacked the buildup and spectacle that now characterizes Nvidia's annual GTC conference, it was rich in technical content. AMD CEO Lisa Su presented a comprehensive roadmap, highlighted by announcements from key customers deepening their commitments to AMD in the competitive data center sector. Customer partnerships signal market momentum Su welcomed Meta's Head of Infrastructure, Santosh Janardhan, who detailed his company's expanding AMD relationship. The collaboration spans multiple generations of AMD's EPYC processors and Instinct GPUs. Meta plans to deploy up to 6 gigawatts of AI infrastructure powered by AMD GPUs—a massive vote of confidence in AMD's data center capabilities. OpenAI's Vice President of Compute Strategy, Sachin Katti, also joined Su on the keynote stage. According to Katti, OpenAI and AMD are collaborating on data center infrastructure, leveraging AMD's Instinct GPUs, Helios rack-scale solution, and EPYC CPUs in a full-stack co-design approach. Expanded product portfolio targets AI workloads AMD has been aggressively expanding its lineup for product for AI applications in recent years, and during the keynote Su discussed the company’s launch of additional new products that expand AMD’s initiatives into agentic AI and data center infrastructure. The company unveiled its Instinct MI400 series GPUs for AI inference and frontier-model training. Applications range from hyperscale AI factories to sovereign AI and scientific computing centers requiring accuracy, security, and long-term operational control. Emphasizing an open ecosystem approach, the MI400 Series provides an open software foundation with programming models, compilers, libraries, runtimes, and deployment tools for AI and HPC workloads. AMD also announced its Helios AI rack-scale solution, a full-stack AI platform addressing large-scale inference and agentic workloads. The solution integrates Instinct MI455X GPUs, AMD EPYC Venice CPUs, and ROCm open software for maximum flexibility and performance. More EPYC Processors AMD has expanded its EPYC CPU family for agentic workloads, cloud computing, and general-purpose applications. The variants include: EPYC 9006 SP7: Targets agentic computing with up to 256 cores and 512 threads for maximum parallel processing power. EPYC 9006 SP8: Built for efficient, right-sized performance across enterprise workloads and agent sandbox execution. With flexible 8- to 128-core configurations and low-power designs, it supports edge deployments, power-constrained racks, and general-purpose servers engineered for leadership performance per dollar. EPYC 9006X SP7: Optimized for HPC, technical computing, and data-intensive workloads benefiting from high cache and memory bandwidth. It accelerates simulation, modeling, large-scale analytics, retrieval, and in-memory processing. EPYC 9006 LP (formerly codenamed "Verano"): Purpose-built as an AI host node CPU for next-generation rack-scale AI systems. Leveraging LPDDR memory and frequencies up to 5GHz, it keeps accelerators fed across dense, accelerator-rich racks, enabling system utilization to scale from single nodes to AI factories. AMD added new variants to its EPYC series CPUs. (AMD) Analysts laud AMD's efforts Industry analysts believe AMD's aggressive product launches demonstrate serious intent to capture the burgeoning AI data center market. A Seeking Alpha report titled AMD: Major Data Center Deals with Meta & OpenAI noted: "In terms of major AI and hyperscaler customers, many companies had primarily relied on Nvidia GPUs for AI infrastructure. However, OpenAI, Meta, and Oracle have now adopted AMD GPUs as part of their AI infrastructure plans, reducing their reliance on Nvidia. The main benefit is that customers can avoid full dependency on Nvidia GPUs while improving supply access and bargaining power." The report acknowledged: "While AMD continues executing on its product roadmap with the launch of its latest CPUs and GPUs, AMD still trails market leader Nvidia in data center GPUs, as Nvidia maintains a performance advantage in many areas. That said, we credit AMD for making strong improvements compared to previous generation chips." Memory capacity favorable The report said AMD's products compete favorably in critical metrics like memory capacity. "Interestingly, AMD surpasses Nvidia in some specifications. Its Helios system features 50% higher HBM memory than Nvidia's comparable solution. It also delivers 11% higher FP8 performance on paper, while scale-up bandwidth remains similar. Scale-up refers to GPU connections within a data center rack and data transfer speeds—higher bandwidth benefits large AI workloads significantly." For a look at AMD’s AMD Advancing AI Keynote, go here.