What's New
A surprising CPU demand surge driven by agentic AI workloads has caught Intel and AMD off guard, with both companies reporting supply constraints as datacenters rediscover the critical role of general-purpose processors in AI infrastructure. AMD CEO Lisa Su revealed that CPU demand "far exceeded expectations," while Intel CFO David Zinsner declared "the CPU has become cool again this year" [Tom's Hardware].
Unlike traditional AI training dominated by GPUs, agentic AI systems that observe, reason, plan, and act independently require substantial CPU resources for orchestration, scheduling, and coordination tasks. This architectural reality has pushed CPU-to-GPU ratios back toward 1:1 in AI clusters [SemiAnalysis].
Technical Deep Dive
Agentic AI represents a fundamental shift in AI architecture. While GPUs excel at parallel matrix operations for training and inference, CPU-intensive orchestration tasks are essential for autonomous agents that must coordinate multiple models, manage state, schedule tasks, and handle decision logic in real-time.
Why CPUs Matter for Agentic AI
Modern agentic systems require CPUs for:
- Task orchestration – Coordinating multi-step reasoning processes across distributed systems
- State management – Maintaining context and memory across agent interactions
- Scheduling – Dynamic resource allocation and workload balancing
- Data preprocessing – Preparing inputs before GPU processing [TRG Datacenters]
Next-Generation Server CPUs
Both Intel and AMD are pivoting to address this demand with high-core-count processors optimized for AI infrastructure:
Intel Diamond Rapids (H2 2026):
- Up to 192 cores per socket
- 16-channel DDR5 memory support
- Designed specifically for AI cluster deployments [ServeTheHome]
AMD Zen 6 "Venice":
- Up to 256 cores via 32 Zen6c cores per CCD
- 16-channel memory architecture
- New SP8 platform with up to 128 dense cores for enterprise [SemiAnalysis]
Intel has canceled mainstream next-gen Xeon plans to focus exclusively on these high-core-count designs, signaling a strategic pivot toward AI infrastructure [ServeTheHome].
Market Impact
AMD's data center segment posted record Q4 2025 revenue of $5.4 billion, representing a 39% year-over-year increase [Igor's Lab]. The company's 5th Gen EPYC processors captured over 50% of server revenue share for the first time, according to Mercury Research [CRN].
AMD's server market trajectory has been remarkable:
- Server share rose from under 2% in 2018 to nearly 30% by 2025 [TechRadar]
- EPYC revenue grew from under $100 million to over $3.5 billion – a 3400% increase in eight years [TechRadar]
Intel's server share fell from 97% in 2019 to 72% by 2025, with unit shipments declining from 12 million in 2019 to under 8.5 million in 2025 [TechRadar]. The company admitted it could not meet demand for Xeon processors at any price point [The Next Platform].
Industry analysts project CPU demand to grow 34.9% through 2029, potentially outpacing GPU growth by 2028 [Futurum Group]. Meta has already signed a massive deal with AMD for up to 6 gigawatts of CPUs and GPUs, highlighting the scale of hyperscaler demand [Tom's Hardware].
China is experiencing notable supply shortages for server CPUs from both vendors, indicating the global nature of this demand spike [Igor's Lab].
What It Means
For Infrastructure Teams: The GPU-centric AI infrastructure narrative needs revision. Planning for agentic AI deployments requires rebalancing CPU-GPU ratios toward parity, with sufficient general-purpose compute for orchestration layers.
For Procurement: Expect extended lead times for server CPUs through 2026. Both AMD and Intel are prioritizing hyperscale customers, potentially squeezing enterprise availability. Consider long-term supply agreements now.
For Architects: The shift validates hybrid CPU-GPU architectures. EPYC's multi-threaded efficiency (e.g., EPYC 9654 with 96 cores/192 threads) makes it well-suited for AI orchestration workloads, while Xeon maintains advantages in single-threaded stability [ServerMania].
For the Industry: The CPU renaissance signals AI's maturation from training-focused to deployment-focused workloads. As agentic systems proliferate, expect sustained demand for high-core-count x86 processors alongside specialized accelerators.