AI
May 24, 20261
Analysis finds HBM memory taking growing share of AI chip component spending through 2025
Epoch estimates that HBM memory is taking a larger share of AI chip component spending from Q1 2024 to Q4 2025, rising from 52% to 63%, while packaging and auxiliary shares decline and logic remains stable. The group estimates total AI chip component spending grew from about $22 billion in 2024 to about $52 billion in 2025, with roughly $20 billion of the increase tied to HBM.
Quick Facts
- Estimated per-chip costs for AI chip components across four categories: memory (HBM), logic dies, advanced packaging (CoWoS), and auxiliary components
- Multiplied per-chip costs by estimated quarterly production volumes to estimate total component spending by category
- Computed each category’s share of total component spending per quarter from Q1 2024 to Q4 2025
- Reported that memory’s share of AI chip component spending increased while packaging and auxiliary components decreased; logic die share remained roughly constant
- Reported growth in total AI chip component spending from 2024 to 2025 and attributed most of the increase to HBM spending
A new analysis by research group Epoch estimates that high-bandwidth memory (HBM) is consuming an increasing portion of AI chip supply-chain spending, even as the shares for advanced packaging and other supporting parts decline. The work models per-chip component costs and production volumes to estimate quarterly spending across major AI chip component categories from Q1 2024 through Q4 2025.
Epoch’s breakdown covers four categories: memory (HBM), logic dies, advanced packaging (including CoWoS), and auxiliary components. Using estimated per-chip costs for each category and multiplying them by estimated quarterly production volumes, the analysis derives both total component spending by quarter and each category’s share of that total.
Across the period, Epoch reports that memory’s share of AI chip component spending rose from 52% to 63%. Over the same timeframe, advanced packaging’s share fell from 19% to 15% and auxiliary components declined from 15% to 9%, while the logic die share remained roughly steady at about 13% to 14%.
In absolute terms, the analysis estimates that total component spending on AI chips increased from about $22 billion in 2024 to about $52 billion in 2025. Epoch attributes most of that growth to HBM, estimating that memory spending accounted for roughly $20 billion of the year-over-year increase.
The estimates are presented as a way to understand how AI chip supply-chain costs are distributed over time for systems used by companies such as Nvidia, AMD, Google and Amazon. Epoch said the work is available under a Creative Commons BY license, allowing reuse with attribution to the source and authors.
Topics
Why This Matters
For investors, suppliers, and chip buyers, the report signals that AI hardware economics are becoming increasingly memory-heavy. That means HBM capacity, pricing, and supply constraints may have an outsized effect on the cost and availability of AI accelerators, influencing procurement plans, vendor negotiations, and where capital gets allocated across the AI supply chain.
Timeline & Sources
Jan 1, 2024
WireTotal component spend on AI chips estimated at approximately $22 billion.
Jan 1, 2025
WireTotal component spend on AI chips estimated at approximately $52 billion; HBM spending estimated to account for roughly $20 billion of the increase.
May 24, 2026
WireArticle published on hacker_news.