An eight-GPU Nvidia H100 server lists at $4.50 to $6.16 per GPU-hour on demand at three GPU clouds, by our read of the Nebius, Lambda and CoreWeave pricing pages early on 18 November. Hours earlier, on Nvidia’s third-quarter call, interim CEO Colette Kress said “Hopper GPUs have re-contracted at up to twice their original price under the multi-year agreements”.
Our number: the highest of those eight-GPU H100 prices is 1.37 times the lowest, and one such H100 rented all year at those rates would bring in about $39,400 to $53,900, by our arithmetic. Kress also said “the B300 rates have risen more than 60% since their introduction”; Nebius lists the B300 at $9.50 an hour.
Business Pill · Capex and Depreciation Explained
An evergreen Business Engineer explainer on capex and depreciation, how the cost of long-lived assets such as GPUs is spread over their useful life. It teaches the general idea only and does not discuss these companies.
The key insight: An eight-GPU Nvidia H100 server lists at $4.50 to $6.16 per GPU-hour on demand at Nebius, Lambda and CoreWeave, a 1.37-times spread by our arithmetic, on the day Nvidia’s interim CEO said Hopper GPUs re-contracted at up to twice their original price.
What the Price Pages Show
Nebius lists on-demand prices per GPU-hour of $4.50 for the H100, $5.40 for the H200, $8.50 for the B200 and $9.50 for the B300. Lambda lists $4.59 for an H100 and $7.69 for a B200 in its eight-GPU instances; its single-GPU H100 PCIe instance, a lower-spec card, lists at $3.79.
CoreWeave prices eight-GPU nodes: $49.24 an hour for the H100, $50.44 for the H200 and $68.80 for the B200. That is about $6.16, $6.31 and $8.60 per GPU, by our arithmetic. For the B300 node, CoreWeave lists no on-demand price and says to contact sales; its spot price is $35.84 a node.
Lambda prices committed clusters above its instances. A 16-GPU H100 cluster for two weeks to a year lists at $6.16 per GPU-hour, 1.34 times its $4.59 instance rate, by our arithmetic; the cluster price falls to $5.54 at 256 GPUs.

Spot Prices Move Daily
Nebius also publishes live spot prices. At our read, the B300 stood at $5.90 against a seven-day average of $5.16, the B200 at $2.85 against $3.16, and the H200 at $1.58 against $2.15, across all regions for 11 to 18 November.
The B300 spot price was 62% of its on-demand list and the H200’s 29%, by our arithmetic. Spot capacity can be reclaimed, so it is not the same product as an on-demand or contracted GPU.
What Kress Said on the Call
The rent lines came in Kress’s prepared remarks, in our transcription of the public relay of the call. She said “Token prices have fallen nearly 60% since May, accelerating token usage and demand for compute” and “As a result, GPU rental prices are rising across generations”.
She added “NVIDIA Compute remains fully utilized across every hyperscaler and Neocloud we serve”. After those remarks, CoreWeave and Nebius rose 9.0% and 11.0% after hours, about $5.9 billion of market value by our arithmetic.
The Business Lever
As we read it, the lever is what a GPU earns over its life. A GPU cloud borrows to buy chips and repays from rent, so the hourly price is the revenue line behind that debt. At list prices and full use, one H100 brings in about $39,400 to $53,900 a year before power, networking and staff, by our arithmetic.
The index-based H100 rent of $2.88 an hour we reported in October measures market rents, a different measure from list prices, so the two are not directly comparable.
The Structural Read
As we read the pages, list prices differ more by provider than by chip generation: the H100 spread across three clouds is wider than the gap between Nebius’s H100 and H200 prices.
As we read Kress’s lines, they describe contract renewals and rental trends; a single read of list prices cannot confirm or refute them, and Nvidia gave a ceiling (up to twice), not an average.
Colette Kress, NVIDIA interim CEO, Q3 FY2027 earnings call, 17 November 2026 (our transcription of the public relay)
“As a result, GPU rental prices are rising across generations.”
Three Implications
FOR GPU BUYERS List prices for the same eight-GPU H100 server differ by up to 1.37 times across the three clouds, by our arithmetic; Lambda prices its committed clusters above its instances.
FOR LENDERS Hopper re-contracting at up to twice the original price is Nvidia’s statement; it did not name customers or terms, and list prices are not contract prices.
FOR READERS Prices were read once, early on 18 November UTC, and can change at any time. Not investment advice; for information only.
The Business Engineer Lens
This story maps onto the Business Engineer framework The AI Capex Map.
The framework sizes the cycle: “The total cycle: $725B Big Four + $50B Oracle + $13B Apple + $60B Neoclouds + $80B China + $60B Sovereigns + $48B Other = ~$1.04 trillion in 2026 compute capex, the first trillion-dollar year of compute capex in human history.”
As we read it, neoclouds are $60 billion of that sum, about 6% by our arithmetic, and the hourly rent on each GPU is the revenue that has to repay it. The framework predates this report and does not discuss these list prices.
What Is Not Established
List prices are not contract prices; large buyers sign multi-year deals the pages do not show. This is one read, so it does not show whether list prices themselves rose. We did not read Amazon Web Services or Google Cloud prices. Nvidia did not say which customers re-contracted Hopper GPUs at up to twice the price, and “up to” is a ceiling, not an average. The yearly figures assume every hour is sold, and the range covers eight-GPU H100 servers only.
The Bottom Line
An eight-GPU H100 server lists at $4.50 to $6.16 per GPU-hour on demand at Nebius, Lambda and CoreWeave, a 1.37-times spread, on the day Nvidia said Hopper GPUs re-contracted at up to twice their original price. At those rates one H100 would bring in about $39,400 to $53,900 a year at full use, by our arithmetic. Not investment advice; for information only.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Prices were read from the Nebius, Lambda and CoreWeave public pricing pages early on 18 November 2026 UTC; they are list prices and can change at any time. CoreWeave per-GPU figures, ratios and yearly revenue figures are our arithmetic. Quotations from the Nvidia call are our transcription of a public relay of the 17 November webcast. Not investment advice.
Sources: Nebius, GPU pricing page (read 18 Nov 2026) · Lambda, AI cloud pricing page (read 18 Nov 2026) · CoreWeave, pricing page (read 18 Nov 2026) · NVIDIA Q3 FY2027 earnings call, public PreMarket Prep relay (our transcription)









