HBM memory already accounts for 63% of AI chip costs; SK Hynix, Samsung, and Micron capture compute pricing power

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1279 words

According to the latest research from Epoch AI, the share of High Bandwidth Memory (HBM) in the component costs of AI chips has risen from 52% in Q1 2024 to 63% in Q4 2025; absolute spending on HBM surged from $12 billion to $32 billion over the same period, an increase of 167%. (Context: South Korea's KOSPI index soared 8%, triggering a circuit breaker! Samsung shares rose 7.4% intraday as strikes were averted, and SK Hynix surged 11.2%.) (Background: Gavin Baker's three contrarian bets: TSMC is saving the market, Trainium is undervalued, and space computing will be unveiled within two years.) The latest analysis tracks the component cost structures of four major chip designers—Nvidia, AMD, Google, and Amazon—spanning from Q1 2024 to Q4 2025. The research conclusion is clear: the cost bottleneck for AI computing is shifting from chip design to memory supply. High Bandwidth Memory (HBM) is based on the core concept of vertically stacking memory chips and packaging them directly next to the GPU. AI model training and inference require GPUs to read and write massive amounts of data in an extremely short time. Traditional memory bandwidth is insufficient, and HBM is designed to solve this bottleneck: by stacking multiple layers of DRAM chips, transmission speeds can reach several times that of traditional solutions, with significantly reduced latency. Simply put: the task of HBM is to ensure the GPU does not have to wait for memory, rather than making the GPU itself run faster. The problem is that the manufacturing of this memory is extremely complex, with only three companies worldwide—SK Hynix, Samsung, and Micron—possessing mass production capabilities. Since 2022, the demand for AI computing power has grown exponentially, and the bargaining power of these three manufacturers has increased significantly. Epoch AI's data clearly shows the shift in the cost center: in the chip procurement structures of Nvidia, AMD, Google, and Amazon, the cost share of HBM rose from 52% in Q1 2024 to 63% in Q4 2025, an increase of 11 percentage points. During the same period, the share of logic chips (the GPU itself) fell slightly from 14% to 13%, advanced packaging dropped from 19% to 15%, and auxiliary components decreased from 15% to 9%. While the shares of all other segments are shrinking, only memory continues to expand. The price hike in memory is not just a supplier issue; it is propagating down the supply chain. With HBM accounting for 63% of AI chip component costs, and AI chips making up the bulk of capital expenditures for tech giants, any increase in memory prices is directly reflected in financial reports. In its fiscal year 2026 capital expenditure plan, Microsoft separately listed $2.5 billion as a factor for component price increases, a clearly attributable portion of its $25 billion increase. Meta similarly raised its 2026 capital expenditure range by $10 billion, citing rising component costs. In terms of absolute figures, the scale of this propagation is equally staggering: HBM spending surged from $12 billion in 2024 to $32 billion in 2025, a 167% increase in one year. Total spending on AI chip components jumped from $22 billion to $52 billion, an increase of 136%. The entire supply chain has doubled in size within two years, making the structural problem of memory shortages even harder to resolve in the short term. Epoch AI expects that the share of HBM in AI chip component costs will continue to rise in 2026. In other words, an increase in capital expenditure does not necessarily mean a proportional increase in computing power; a significant portion is being used to absorb the costs of rising memory prices.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching2 items

2026-05-15

Inside SK Hynix's "A World Seen for the First Time": KBS Decrypts the HBM Memory Confidential Production Line, Employees Beaming with Smiles

Similarity 170%關鍵字海力士/hbm同分類 zh

2026-05-15

Samsung union confirms 18-day strike starting 5/21: Negotiations to resume after 6/7, HBM supply chain faces further strain

Similarity 120%關鍵字 hbm同分類 zh

💡 Currently matching via keywords + symbols (MVP) · Will be upgraded to embedding semantic search later

Raw Information

ID:2c101df056

Source:動區 BlockTempo

Published:2026-05-26 01:08:00

Category:zh_news · Export Category zh

Symbols:Unspecified

Community Votes:+0 / −0 · ⭐ 0 Important · 💬 0 Comments