Prices for Nvidia's rack-scale AI servers are now changing in weeks rather than quarters, with some components swinging as much as 40% week to week, leaving GPU cloud providers struggling to forecast costs.
AI Compute · Market Volatility
AI Server Racks Now Trade Like a Commodity
Nvidia AI rack prices swing week to week — not by the quarter. An HBM memory squeeze has pushed some components up 40% in a single week, turning compute cost forecasting into spot-market guesswork through all of 2026.
$7.8M
Per-rack cost, VR200 NVL72 — roughly double a year earlier
+485%
Memory cost jump in a single generation (435–485%)
8×
Cost-efficiency gap for the same workload across zones
40%
Swing in some components within a single week
Same Chip, Wildly Different Price
Hourly cloud GPU rates span an enormous range by provider, region and timing.
Regional H100 differences have ballooned to the equivalent of $7,400 a month for the same usage.
Why Racks Got Expensive — Memory's Rising Share
Memory now makes up about a quarter of a rack's total cost.
Contract DRAM forecast:+50–55% QoQ (Q3 2026)+30–40% further (Q4 2026)Server memory prices could double.
The Root Cause — A Supply Squeeze Loop
AI demand surges (Blackwell / Rubin)
→
Makers shift to HBM memory
→
Standard DRAM crowded out
→
Rack & cloud prices spike weekly
The HBM shortage is expected to persist through all of 2026 — with inventory described as effectively sold out. Providers are turning to multi-cloud low-price zones, dedicated instances and FinOps-style forecasting to cope.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…