According to data analysis released on May 21 by Venkat Somala, a researcher at Epoch AI, the share of high-bandwidth memory (HBM) in the total component costs of four major AI chip designers—Nvidia, AMD, Google, and Amazon—has risen from 52% in Q1 2024 to 63% in Q4 2025, when weighted by production volume. Meanwhile, advanced packaging (CoWoS) saw its share drop from 19% to 15%, while auxiliary components fell from 15% to 9%; logic dies remained roughly unchanged at about 13%. In absolute terms, HBM-related spending by these four firms climbed from approximately $12 billion throughout 2024 to around $32 billion in 2025, outpacing growth rates of all other components; overall spending on AI chip components rose from roughly $22 billion to about $52 billion over the same period.
Epoch AI notes that given ongoing supply constraints and rising prices for HBM, its share in component costs is expected to climb further in 2026. Major hyperscale cloud providers have already factored this trend into their capital expenditure forecasts: Microsoft’s planned $190 billion in capex for fiscal year 2026 includes roughly $25 billion attributed to higher component prices; Meta likewise raised its 2026 capex guidance by $10 billion, citing escalating component costs as the primary driver. These figures stem from Epoch AI’s AI Chip Component Explorer tool, which compiles bill-of-materials (BOM) data for various chip models using financial disclosures, supplier filings, and analyst reports; cost estimates for each component come with a 90% confidence interval.