Intel Data Center CPUs Sold Out
Intel's data center CPUs are reportedly sold out for the coming quarters as hyperscalers lock down capacity at its fabs. The demand surge is attributed to agentic AI workloads shifting to CPUs, turning fab allocation into a new strategic battleground for AI compute supply.
The surge in CPU demand is driven by a fundamental shift in AI architecture. Agentic AI workflows, which involve complex orchestration, API calls, and logic, are inherently CPU-bound, moving the processor from a supporting role to the command layer for AI systems. This evolution marks a significant change from the purely GPU-centric focus of the past. A joint research paper by Georgia Tech and Intel from November 2025 quantified this bottleneck, finding that tool processing on CPUs can account for 50% to 90% of the total latency in agentic AI tasks. In these new workloads, expensive GPUs often sit idle, waiting for the CPU to manage and dispatch the next steps, completely inverting the previous infrastructure economics. This isn't an isolated event for Intel. Rival AMD has also reported an "unexpected" surge in demand for its EPYC server processors due to the same AI workload shift. Amid Intel's supply constraints, AMD captured a record 28.8% of the server CPU market in the fourth quarter of 2025. In response to the demand, Intel CFO David Zinsner confirmed that many of the company's factories are operating at over 100% capacity, with shortages expected to persist through the end of the year. To manage the crunch, Intel has been reallocating wafer production away from its lower-margin client PC chips to prioritize the high-end data center products. The renewed importance of the CPU is forcing a strategic rethink of data center design, pushing the CPU-to-GPU ratio in AI clusters back toward 1:1. Even GPU-leader Nvidia is responding to this trend, developing its "Vera" CPU platform specifically designed for agentic reasoning workloads, which can be deployed independently of a GPU. This capacity crunch coincides with Intel's multi-billion dollar bet on its foundry services, aiming to manufacture chips for other companies, including potential hyperscaler customers like Apple, Microsoft, and Amazon. The successful ramp-up of its advanced 18A process node is critical to competing with TSMC and attracting these anchor clients.
