If you buy servers, plan for tight Xeon supply through at least the first half of 2027. Intel is shifting manufacturing toward data center processors and says it still cannot meet demand, so lead times and allocation are the risks to budget around, not list price. Two details remain open: Intel has not said how it will split Arizona capacity between client and server products, and the Fab 24 expansion in Ireland might arrive sooner than planned if Intel can accelerate it.
What happened
According to Tom’s Hardware, which covered Intel’s third-quarter earnings call, the company is moving some capacity from client to data center CPUs. Its Data Center and AI unit reported $7.3 billion in revenue, up 17% from the prior quarter and nearly 80% from a year earlier. CEO Lip-Bu Tan called the situation “one of the most significant supply shortages in history” and said he expects ongoing difficulty keeping pace with server demand.
The main lever is Intel 3, the process used for Xeon 6 Granite Rapids. Intel plans to convert some Oregon resources from Intel 18A to Intel 3. Those resources were originally meant for Panther Lake client processors, and the company says weaker PC demand let it prioritize data center products. It expects the extra Intel 3 capacity to enter production in the second half of 2027. The Fab 24 expansion in Ireland is due to come online by the end of 2027, unless Intel can speed it up. Intel also reports that Intel 3 manufacturing costs have fallen about 15% since the first quarter of 2026.
Wafers are not the only constraint. CFO Dave Zinsner said Intel is also limited on the substrate side, and the company expects about 40% more substrate capacity in the second half of 2027. Intel is also expanding advanced packaging in New Mexico and Malaysia. It expects capital spending of about $20 billion in 2026 and a significant increase in 2027, and it says it is improving yields and cutting cycle times on existing capacity.
Why it matters
AI demand is pulling on the whole compute stack, not only accelerators. Intel expects CPU-heavy agentic and inference workloads to keep data center CPU demand strong in 2027 and 2028. Our analysis: if general-purpose CPUs are becoming a constrained input to AI infrastructure, enterprise buyers may find themselves competing with the largest operators for the same parts. Intel did not say this directly.
The fix also takes time. New tools must be installed and qualified, wafers pass through long manufacturing cycles, and substrate suppliers need their own equipment. Intel says it has secured more tools, with firm purchase orders to equipment suppliers up nearly 40% from the prior quarter. Even so, it expects constraints to be particularly severe in the first half of 2027. Buyers should treat the second half of 2027 as the earliest point where added capacity could show up, not as a guarantee.
The client side is less clear. Intel says the PC market will keep shrinking, yet it expects to gain some client share with its Nova Lake processors, which will need high-volume production. Intel has not said how it will divide 18A capacity in Arizona between client and data center products.
What it means for you
Start with procurement timing. If you need Xeon-based servers in the first half of 2027, order early and ask your vendor for committed delivery dates rather than estimates. Ask whether quotes hold if allocation changes. You could also ask whether any configurations ship later than others, since packaging is one of the constraints Intel describes.
Build slack into capacity plans. If a refresh depends on new hardware arriving in a given quarter, keep a fallback such as extending existing server lifecycles, adding cloud capacity for peaks or qualifying a second platform. Check that your workloads run on more than one CPU generation so you can accept whatever ships.
For PC fleets, Intel has not described a client shortage, but it has said it is moving some capacity to servers. Confirm availability of specific models before committing to large rollouts. Teams that compare cloud costs with owned hardware should rerun the numbers, since hardware delays and cloud providers’ own supply pressure can both shift the balance. Our hardware coverage tracks related supply moves.
What to watch
- Whether Intel accelerates Fab 24 in Ireland ahead of the end of 2027.
- The Oregon conversion to Intel 3 entering production in the second half of 2027.
- Progress on the roughly 40% substrate capacity increase, since substrates can limit shipments regardless of wafer supply.
- Whether Intel discloses how it will divide Arizona 18A output between Nova Lake client chips and server products.
