Summary:
1. CPUs have become ‘chokepoints’ in AI infrastructure, but they are not being cannibalized by GPUs.
2. Foundry shortages of Intel 10/7 nodes have led to increased pricing for higher-end offerings.
3. Enterprises can prepare for price increases by diversifying suppliers.
In the realm of AI infrastructure, CPUs have taken on a new role as ‘chokepoints’, rather than being overshadowed by GPUs. CPUs like Granite Rapids play a crucial part in GPU clusters and managing agentic AI workloads. They are also essential for orchestrating distributed inference tasks.
The demand for higher-end offerings has caused foundry shortages of Intel 10/7 nodes, which make up a significant portion of the company’s production volume. This scarcity has led to an increase in pricing, with Intel facing pressure until at least Q2 2026 when chip production is expected to rise. Xeon manufacturing capacity for 2026 is already sold out, and custom silicon programs are experiencing lead times of 6 to 8 months, stretching into 2027.
In the data center space, memory is identified as the primary bottleneck, with projected price hikes of over 65% year over year in 2026. NAND Flash prices could also see an increase of up to 25%. Some products have already witnessed price inflation exceeding 1,000% since 2025, and new memory capacity is not anticipated until 2027 or 2028. Despite these challenges, there is a glimmer of hope as memory prices are predicted to stabilize this year on the client side, awaiting additional capacity in 2027.
To navigate the turbulent waters of rising prices, enterprises are encouraged to diversify their supplier base. While this may introduce complexity, it allows data center operators to mitigate price shocks by reallocating resources among suppliers with better planning or more resilient supply chains. Supplier diversification is seen as a strategic move to prepare for future price fluctuations and ensure a more stable procurement process.
