Intel and AMD warn China customers of server CPU delays as AI demand squeezes compute

The AI build-out is starting to pinch parts of the datacenter stack that normally never get a headline. Intel and AMD have both warned customers in China about stretched lead times for server CPUs, with Intel reportedly flagging waits as long as six months for some parts. When the market is this tight, the consequences are predictable: higher prices, longer planning cycles, and more projects that slip because one critical component shows up late.

The details are uncomfortable if you are trying to stand up capacity quickly. Intel’s server CPUs in China are said to be “more than 10% higher” in price in general, with the usual caveat that it depends on contracts and who has leverage. AMD is not immune either, with lead times on some products pushed out to roughly eight to ten weeks. The broad takeaway is that CPU supply is no longer a given, and that is a problem because even “AI servers” still need a lot of conventional compute to keep everything fed and coordinated.

China is also not a small side market for Intel. It accounts for more than 20% of Intel’s overall revenue, and the reported shortages are centered on fourth- and fifth-generation Xeon models, which are apparently tight enough that Intel is rationing deliveries and carrying a backlog of unfulfilled orders. That is the sort of situation where customers do not just wait. They start hoarding allocations, pulling orders forward, and paying up to avoid being the one left holding an empty purchase order.

What is driving this is the same story we have seen elsewhere, just spreading sideways. The rush to build AI infrastructure is not only about accelerators. It drags CPUs, memory, storage, networking, and everything else along with it. Memory in particular has been the canary in the coal mine, with prices rising sharply, and that has knock-on effects: when buyers think the overall bill of materials is about to get worse, they accelerate purchases to lock builds in. That kind of pull-forward demand is great if you are a supplier in the short term, but it is brutal on availability.

On the supply side, the two vendors are constrained for different reasons. Intel has been battling manufacturing yield challenges while trying to ramp production, which makes any sudden demand spike harder to absorb. AMD’s situation is more about foundry capacity: it relies on TSMC, and when foundry lines are being prioritized for high-margin AI silicon, there is simply less room for mainstream server CPUs. In other words, even if demand is healthy, supply is only as flexible as the manufacturing pipeline, and that pipeline is already crowded.

One detail worth paying attention to is the mention of “agentic AI” demand. This is not just chatbot inference. These systems run multi-step workflows and tend to lean more heavily on CPU resources than people assume. If that trend accelerates, it makes the squeeze worse, because it increases demand for the very server CPUs that are now showing the longest queues.

There is also an awkward competitive angle here. Intel still dominates server CPUs, but its share has fallen significantly over the last several years while AMD’s has climbed. In a more balanced market you would expect shortages to create openings for a rival, yet both are still seeing constraints. That tells you this is not simply a “vendor problem.” It is a capacity and supply chain problem, made worse by AI-driven urgency and memory pricing chaos.

For buyers, the practical impact is simple: procurement gets slower and more expensive. If your CPU lead time slips by months while memory and accelerator delivery dates shift around, it becomes harder to align an entire server build. Even when the CPUs eventually arrive, the schedule damage has already been done.

Sources: