If you're trying to source Intel Xeon or AMD EPYC right now, you already know it's not easy. What you may not know is how fast things are actually moving, and how quickly the window to act is closing. Here's what the market looks like from where we sit.
At Fusion, we're fielding calls daily from data center operators who can't get allocation. Here's what we're actually seeing and what we're telling our customers to do about it.
The server CPU market in 2026 is facing one of its most severe supply disruptions in recent memory. A collision of structural demand drivers (agentic AI infrastructure buildouts, continued data center expansion, and accelerated enterprise refresh cycles) has overwhelmed the manufacturing and distribution capacity of both Intel and AMD. What was already a tight market entering the year has become critically constrained, with the shortfall affecting customers of every size.
This isn't a temporary blip. Both Intel and AMD are capacity-constrained. Foundry limits, swelling backlogs, and longer-range purchase orders are all compounding the problem at once.
A fundamental shift in how AI infrastructure is being built is at the heart of this demand surge. Hyperscalers and enterprise data centers are rapidly scaling agentic AI deployments — systems that operate autonomously across complex, multi-step tasks — and these workloads require a very different compute ratio than traditional AI inference.
In a standard GPU-centric AI cluster, the historical CPU-to-GPU ratio has been roughly 1 CPU for every 4 to 8 GPUs. Agentic AI breaks that model entirely. Because agentic workloads require continuous orchestration, memory management, and parallel decision-making across many concurrent tasks, the CPU carries a far heavier load. Hyperscalers are now architecting infrastructure at closer to a 1:1 CPU-to-GPU ratio. AMD CEO Lisa Su has noted the ratio could eventually flip, with some deployments requiring more CPUs than GPUs as agentic AI scales.
This is why AMD doubled its server CPU TAM forecast to $120 billion in its most recent earnings call. It's not just more AI; it's a structurally different kind of AI that needs far more general-purpose compute. The same dynamic is driving Intel Xeon demand, as data center operators scramble to build out the CPU-dense infrastructure that agentic workloads require. Supply simply hasn't caught up, and at current trajectory, it won't by end of 2026.
Below is a detailed breakdown of the current state for Intel and AMD server CPUs, including real market pricing data, the most-requested SKUs, and what buyers should be planning for.
Intel's server CPU supply situation has deteriorated sharply in 2026. Distributors are currently fulfilling only approximately 40% of yearly backlog allocations, leaving the majority of outstanding orders unfilled. Based on Fusion's order flow data, Intel is under-shipping real demand by roughly 20% at best, suggesting the true gap between supply and actual need is larger than even allocation figures indicate.
Lead times now stretch 8 to 22 weeks depending on SKU. Customers in Asia are reporting wait times of 6 to 8 months on certain models, while North American customers are receiving some degree of delivery prioritization, though that advantage is narrowing as global demand intensifies.
The shortfall is especially acute at the low end of the 5th Generation Emerald Rapids lineup, where Intel has little to no available capacity for entry-level models. This is pushing some customers toward higher-tier SKUs or accelerating their evaluation of AMD alternatives.
Intel has raised pricing 10% for most customers in 2026, with further increases expected later in the year. On the open market, prices are tracking well above Intel's official list, particularly for high-demand SKUs. The Xeon Silver 4510 (SRN60), for example, is trading at $2,300–$2,400 on the open market, a meaningful premium over historical norms.
The top six most-requested Intel SKUs over the past month span Sapphire Rapids, Emerald Rapids, and Granite Rapids: the Xeon Silver 4510, Xeon 6517P, Xeon Gold 6530,Xeon 6747P, Xeon Gold 6526Y, and Xeon 6767P.
After RDIMM shortages, Intel CPU constraints have become the #2 reason server builders are delaying shipments to end customers. Customers who cannot meet their Intel allocation needs are increasingly evaluating AMD EPYC as an alternative, a trend that could reshape server platform preference heading into 2027.
AMD EPYC is effectively sold out through the end of 2026. That's not hyperbole; it's what analysts are broadly reiterating, and it matches what we're seeing in our own order flow. Official lead times are listed at 12–18 weeks, but actual delivery windows are stretching beyond 30 weeks depending on SKU and customer tier. Top VIP customers are being given delivery priority, which means mid-market customers with upside volume needs or spot requirements face particularly difficult sourcing conditions.
The shortage is cascading across generations. Top customers are receiving partial shipments on Turin (5th Gen) CPUs, driving renewed interest in securing older-generation Genoa (4th Gen) and Milan (3rd Gen) EPYC processors from the open market. Demand for Genoa and Milan has been accelerating in recent weeks, with models like the EPYC 7763 and EPYC 9454 seeing notably increased activity.
AMD EPYC open-market prices are trending upward across the board, with Turin commanding the steepest premiums. Customers who secured pricing earlier in the year are sitting on increasingly valuable inventory. Those coming to market now are paying significantly more, and extended lead times mean the cost of waiting only compounds.
The seven most-requested AMD SKUs over the past three weeks have been the EPYC 9124 (Genoa), EPYC 9455P (Turin), EPYC 7763 (Milan), EPYC 9454 (Genoa), EPYC 9475F (Turin), EPYC 9375F (Turin), and EPYC 9555 (Turin). The mix of generations is telling — Turin demand is high but so is the fallback to Genoa and Milan as customers work around allocation gaps.
AMD's most recent earnings call delivered a signal that every server buyer should understand. AMD has doubled its server CPU total addressable market forecast from $60 billion to $120 billion, not because the market was undersized before, but because agentic AI is fundamentally expanding what CPU demand looks like. CEO Lisa Su cited shifting CPU-to-GPU ratios as the core driver: from 1:4 or 1:8 toward 1:1, and potentially inverting entirely. AMD's next-generation platform, EPYC Verano, is being built specifically for this AI-first workload profile. The demand wave this represents is not priced into most procurement plans, and supply is nowhere near ready for it.
This is the guidance we're giving our customers right now.
Fusion carries current inventory across Intel Xeon (Sapphire Rapids, Emerald Rapids, Granite Rapids) and AMD EPYC (Milan, Genoa, Turin), including many of the most-requested SKUs listed above.
Browse Server CPU Inventory at fusionww.com →
Need a firm quote on a specific MPN or volume? Contact Fusion's server CPU specialists directly for reconfirmed availability and pricing based on your volume, location, and lead-time requirements.