HBM now dominates AI chip component costs

Epoch AI’s new component-cost analysis puts a clear number on a shift many hardware buyers already feel. High-bandwidth memory rose from 52% of AI chip component spending in Q1 2024 to 63% in Q4 2025. The estimate covers AI chips designed by Nvidia, AMD, Google, and Amazon, weighted by production volume. Logic dies stayed near 13%, while advanced packaging fell from 19% to 15% and auxiliary components declined from 15% to 9%.

The absolute figures are just as important. HBM spending across those four designers grew from roughly $12 billion in 2024 to $32 billion in 2025. Total AI chip component spending grew from about $22 billion to $52 billion. In other words, much of the new cost in AI accelerators is not only in the compute die. It is in the memory stack attached to it.

The Hacker News thread connected the data to everyday hardware markets. Commenters pointed to rising RAM prices, delayed PC upgrades, and the awkward position of buyers who are not building AI clusters but still pay into the same constrained supply chain. Others asked why hyperscalers do not integrate further upstream if component spending is now measured in tens or hundreds of billions of dollars. The answers are not simple: memory manufacturing requires capital, process expertise, patents, and long supplier relationships.

Epoch AI expects HBM’s share could rise further in 2026 if supply remains tight and prices continue upward. That has direct implications for cloud AI economics. Microsoft and Meta have already discussed higher capital expenditure expectations tied to component prices. Model competition therefore depends not only on algorithmic efficiency or accelerator availability, but also on HBM3 and HBM3e capacity, advanced packaging, and purchasing power across the supply chain.

Source: Epoch AI’s data insight. Community context: Hacker News discussion.