來源钛媒体•較早收集於 2h
讓黃仁勳「低頭」的存儲,到底有多金貴?

💡HBM shortages choke AI compute—plan hardware scaling now before delays hit
⚡ 30 秒速覽
有什麼變化
HBM讓黃仁勳低頭
為什麼重要
HBM限制可能延遲依賴GPU叢集的AI模型訓練與推論擴展。這強調供應緊俏下需多元化記憶體採購策略。
下一步行動
Audit HBM availability in Nvidia GPU quotes for your next AI training cluster procurement.
誰應關注:Developers & AI Engineers
關鍵要點
- •HBM讓黃仁勳低頭
- •因稀缺性價值極高
- •卡住AI算力時代脖子
🧠 深度解析
背景與延伸:來自公開資料,非原文內容。引用 4 個來源。
🔑 增強重點摘要
- •SK Hynix holds 62% HBM market share as of Q2 2025, supplying ~90% of Nvidia's HBM needs, followed by Micron at 21% and Samsung at 17%.[2]
- •HBM capacity is sold out through 2026 across SK Hynix, Micron, and Samsung, with suppliers reallocating fabs from DDR to HBM4 production.[1][2]
- •Nvidia is reducing gaming GPU production by 30-40% in H1 2026 due to GDDR7 constraints, prioritizing AI chips amid HBM demand.[2]
- •Micron expanded its HBM customer base to six by Q4 2025, supplying both Nvidia and AMD, and claims HBM4 leadership in performance and power efficiency.[3]
🛠️ 技術深入
- •HBM4 enters production in 2026 with 16-Hi stacks targeting Q4; SK Hynix and Samsung delivered samples to Nvidia.[2]
- •Nvidia Vera Rubin NVL72 offers 1.5x HBM capacity of Blackwell and 2.8x HBM4 bandwidth; Rubin Ultra (2027) targets HBM4e with 12-16-Hi stacks up to 1TB per GPU.[1]
- •Nvidia GB200 supports 13.4TB HBM, GB300 NVL72 up to 21.7TB, 34x higher than H100's 640GB.[3]
🔮 前景展望基於引用來源的 AI 分析
HBM market TAM reaches $100B by 2028
Projections show ~40% CAGR from $35B in 2025 due to AI demand and capacity constraints.[2]
Memory suppliers gain unprecedented pricing power through 2026
ASICs like TPUs and Tranium exacerbate HBM shortage
Growing ASIC adoption requires similar HBM scaling as Nvidia GPUs, tightening supply without production increases.[1]
⏳ 時間線
2025-06
SK Hynix reaches 62% HBM market share per Q2 data.
2025-12
Micron expands HBM customers to six including Nvidia and AMD.
2026-01
SK Hynix and Samsung deliver HBM4 samples to Nvidia.
2026-03
HBM4 mass production preparations complete by SK Hynix.
2026-04
Nvidia announces gaming GPU cuts for H1 due to memory shift.
📎 來源 (4)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源: 钛媒体 ↗
每週電子報
每週一封,可隨時退訂。



