High-bandwidth memory now dominates AI chip costs, Epoch AI data shows

High-bandwidth memory (HBM) has become the primary cost driver in artificial intelligence chip manufacturing, now accounting for 63% of total component costs. This figure represents a significant escalation from the first quarter of 2024, when memory made up 52% of the same cost base, according to a recent breakdown by Epoch AI.

The data, published by the research organisation Epoch AI, tracks cost shifts across major chip designers. The increase in the memory share suggests a tightening supply dynamic or a structural shift in how value is allocated within the bill of materials for advanced AI processors.

Epoch AI’s analysis underscores the growing financial importance of HBM in the production of artificial intelligence hardware. As chip architectures evolve to support more complex computational tasks, the proportion of expenditure dedicated to memory continues to outpace other component categories.

The specific timeframe for the current 63% figure is not explicitly detailed in the source material beyond the comparison to the first quarter of 2024. The breakdown covers major chip designers, though the specific entities included in the analysis were not listed in the provided data.

Analysts note that the term 'AI chip component costs' is broad, and caution is advised regarding whether this metric reflects total bills of materials, specific chip generations, or average costs across different architectures. The claim is attributed solely to Epoch AI, with no independent verification from other industry analysts provided in the source text.

The rise in memory costs highlights the critical role HBM plays in the current artificial intelligence infrastructure build-out. For investors and institutions monitoring the semiconductor supply chain, this shift indicates that memory procurement and pricing will remain a central factor in the economics of AI chip production.