Nvidia Tells Top Customers AI Server Prices to Rise Over 15%
According to Bloomberg (Aug 23) and The Information, Nvidia has notified its largest customers—including Microsoft, Google and Oracle—that prices for data center servers with its AI chips will rise by more than 15% (about 17% per The Information), applying to systems shipping early next year and covering flagship Vera Rubin and Grace Blackwell products. The increase stems not from chip manufacturing costs but from surging DRAM and HBM prices at Samsung, SK Hynix and Micron, showing memory supply tightness propagating downstream through the AI hardware supply chain. Server makers and contract manufacturers have also issued price-increase notices.
Confirmed
- Magnitude: server system prices up more than 15%, about 17% per The Information; @pstAsiatech reports server manufacturers have begun notifying customers of roughly 17% increases
- Scope: systems shipping early next year, covering Vera Rubin and Grace Blackwell; @量子位 notes some GB300 and Vera Rubin 200 systems up about 17%; @新智元 and @机器之心 say the increase takes effect in early 2027 and that contract manufacturers have also notified customers
- GB300 servers cost about $600,000 (per @zephyrz9); a Rubin NVL72 rack with 72 GPUs is expected to cost about $8 million, up from roughly $6.6 million (per @ChrisGPT)
- Cause: surging DRAM (including HBM) costs from Samsung, SK Hynix and Micron, not GPU manufacturing costs; @新智元 cites TrendForce data showing DRAM prices rising and memory supply tightening in Q1 2026, with contract manufacturer gross margins falling below 5%
- @rohanpaulai relays an estimate that memory inflation could add roughly $5 billion to a 1GW-scale deployment
Unconfirmed
- @AccBalanced's relayed cloud-vendor impact is only partially verifiable: NeoClouds such as CoreWeave face rising costs but are supply-constrained rather than demand-constrained, and $NBIS says it can sell out all 2027 compute capacity; this is a single-source account and the pass-through path remains unclear
Why it matters
- This is a rare large-scale price hike on flagship AI servers by Nvidia, showing memory chip tightness pushing costs downstream through the AI hardware supply chain
- @dinabass notes the hike underscores strong demand for AI compute and Nvidia's dominance in the hardware supply chain, giving it leverage to pass costs to major customers
- @emmanuelvivier argues the increases may affect downstream model training and inference operating expenses, tightening compute budgets
- @新智元 adds that contract manufacturer margins have been squeezed below 5%, pressuring profitability across the AI server industry chain
- @AccBalanced points out that for cloud providers the bottleneck is supply, not demand, so cost pressures will likely flow through cloud pricing to downstream AI applications; memory cost inflation plus server price hikes will significantly raise build budgets for next-generation AI data centers
2026-08-23 ~ 2026-08-24 · 15 related posts
Primary sources
- [source] Nvidia AI Server Prices to Rise 15%+, GB300s Around $600k — zephyr_z9 · 2026-08-23
- Nvidia reportedly hiking AI server prices by 15%+ for major clients — Polymarket · 2026-08-23
- [source] Nvidia tells big customers server prices to rise over 15% next year — firstadopter · 2026-08-23
- Nvidia Tells Customers AI Servers Getting ~17% Pricier as Memory Costs Surge — rohanpaul_ai · 2026-08-23
- Nvidia AI Server Prices to Rise Over 15% Next Year, Adding $5B to 1GW Data Center Costs — 量子位 · 2026-08-23
- [source] NVIDIA reportedly raising GB300 and Rubin system prices by 17% — ChrisGPT · 2026-08-23
- Nvidia hikes some AI product prices by over 15% on surging memory chip costs — emmanuelvivier · 2026-08-23
- Nvidia Notifies Customers of AI Price Hikes Over 15% — fallingdowndizzyvr · 2026-08-24
- Nvidia raises AI server prices 15% as memory becomes new bottleneck — heyshrutimishra · 2026-08-24
- DRAM Shortage Drives AI Server Price Hikes Over 15%, ODM Margins Under Pressure — 新智元 · 2026-08-24
- Nvidia raises server prices 15%; how neoclouds pass costs to end users — AccBalanced · 2026-08-24
4 near-duplicate retellings: dinabass · pstAsiatech · emmanuelvivier · 机器之心