H100 Rental Price: Neocloud Indexes Vs Hyperscaler List Prices in November 2026

Summary

The H100 rental price in November 2026 is about $3.06 to $3.10 per GPU-hour on neocloud benchmarks. Hyperscaler list prices run roughly $5 to $12 per GPU-hour, depending on the pricing tier.

• Two neocloud readings. Silicon Data's SDH100RT read $3.06 per GPU-hour on Nov 10, 2026, and Ornn's OCPI-H100 settled at $3.10 on Nov 10, 2026.
• Hyperscaler on-demand lists. On-demand lists at AWS, Google Cloud and Azure run about $6.88 to $12.29 per GPU-hour (calculated); 3-year commitments narrow the gap.
• Rising rates, contracted value. H100 rates rose through 2026, and lenders value the contracted share of that revenue more than the on-demand share.

Why the H100 Rate Is the Market's Benchmark

H100 prices look opaque because list prices, index readings and contract rates each measure something different. A hyperscaler list price is the published rate anyone pays to start using an instance. An index reading averages or settles rates across the neocloud market, and a contract rate is whatever an operator and its customer signed.

For an operator that owns H100s, the H100 rental price is the revenue line. The contract behind that rate shapes how lenders value it. A fixed rate on a multi-year offtake contract carries less uncertainty than hourly sales at the going rate. That difference follows you into GPU financing for neoclouds on your next deployment, whatever hardware it uses.

The H100 is the benchmark because the market's pricing tools are built on it. Silicon Data and Ornn both publish H100 indexes, and the compute futures CME Group has proposed reference an H100 rental index. The sections below cover current index readings, hyperscaler tier prices and why rates rose in 2026. They close on how lenders read contracted and on-demand revenue, leaving cluster build costs aside.

H100 Neocloud Rental Indexes in November 2026

Silicon Data's neocloud H100 index (SDH100RT) read $3.06 per GPU-hour on Nov 10, 2026. Ornn's OCPI-H100 settled at $3.10 per GPU-hour on Nov 10, 2026. The two readings come from different methods, so treat each as its own data point.

Silicon Data standardizes each rental record to a per-GPU-hour rate and adjusts it to a benchmark contract. It then averages the rates by provider and weights those averages across providers to produce one reading. The result describes what neocloud providers charge for comparable H100 capacity.

Ornn builds OCPI-H100 from executed compute trades, so the reading reflects prices at which compute changed hands. Ornn also sells compute, which makes it a market participant as well as a publisher. Keep that in mind when you weigh the two readings against each other.

The next step for this benchmark is a futures market, and that step hasn't happened yet. The H100 contract in CME compute futures references the Silicon Data Daily H100 Rental Index. Each contract covers 730 GPU-hours, with monthly contracts out 36 months. CME Group shows the contracts as pending regulatory review.

A 730 GPU-hour contract equals one GPU running for an average month, which matches how operators bill and how lenders model revenue. A benchmark rate is what lets lenders and buyers price compute the way they price other assets. With a reference price in place, you can compare an offtake contract rate against the market and measure the revenue risk behind it.

Hyperscaler H100 List Prices by Pricing Tier

Hyperscaler on-demand H100 list prices sit at roughly two to four times the neocloud index readings, and multi-year commitments close part of the gap. On demand, AWS, Google Cloud and Azure work out to about $6.88 to $12.29 per GPU-hour, calculated from 8-GPU instance prices. Each provider sells the same hardware on several tiers:

• On-demand: hourly billing with no commitment, at the full list price.
• Spot: spare capacity at a discount, which the provider can reclaim at short notice.
• Committed or reserved: a 1- or 3-year commitment in exchange for a lower hourly rate.
• Capacity blocks: prepaid GPU capacity reserved for a fixed window at a set price.

Google Cloud's a3-highgpu-8g listed $88.49 an hour on demand for 8 GPUs when read on Nov 10, 2026. The page doesn't label the region, and its spot, 1-year and 3-year prices appear in the table below.

Microsoft's Azure Retail Prices API lists the ND96isr H100 v5 at $98.32 an hour pay-as-you-go in East US on Linux. Its reservations are priced per VM for the whole term, at $551,221 for one year and $1,134,310 for three years. These are standing list prices, in effect since December 2023.

AWS published the p5.48xlarge, its 8-GPU H100 instance, at $55.04 an hour on demand in May 2026. EC2 Capacity Blocks priced P5 capacity at $5.970 per H100-hour on a page updated Nov 5, 2026. AWS says these prices "are updated regularly based on trends in supply and demand."

Per-GPU figures below are calculated as the 8-GPU instance price ÷ 8. For Azure reservations, the calculation divides the VM price by the hours in the term (8,760 or 26,280) and then by 8. The Capacity Blocks price is published per GPU-hour, so it needs no calculation.

Silicon Data tracks the same gap in its own numbers. Its hyperscaler H100 reading was $7.10 as displayed on Nov 10, 2026, against the $3.06 neocloud reading that day.

For a neocloud, these lists are the ceiling a customer compares against when it negotiates an offtake contract. Even Google's 3-year rate of about $4.86 per GPU-hour sits above both neocloud readings. That leaves room for a neocloud to price a multi-year contract below hyperscaler commitments and still earn a rate near or above the index.

Why H100 Rental Rates Rose in 2026

Providers raised H100 prices through 2026 as demand outran available capacity. The clearest evidence comes from the providers themselves, in earnings calls, releases and changelogs.

CoreWeave said on its Aug 11, 2026 earnings call that its July pricing changes raised rates about 25% across SKUs. The company also described its near-term capacity as "effectively sold out."

In a Sep 17, 2026 release, CoreWeave said new 3- to 6-month contracts were signed at about $40 million per megawatt of annualized revenue. That figure is stated per megawatt of capacity. Converting it to an H100 GPU-hour rate would need hardware and power assumptions the release doesn't give, so we leave it as published.

Nebius noted in its changelog that compute prices changed on Oct 1, 2026 for VMs with NVIDIA H100, H200, B200 and B300 GPUs. In our analysis of why companies will own compute, Nebius's H100 on-demand price moved from $3.85 to $4.50 per GPU-hour. That is a rise of about 17%.

Committed rates moved as well. Our market analysis cited SemiAnalysis's 1-year H100 contract index, which rose from $1.70 per GPU-hour in October 2025 to $2.35 in March 2026. That matters for operators signing new offtake contracts, because the term rate a customer will accept has risen alongside the hourly rate.

AWS Capacity Blocks reprice on the same pressure, since AWS sets them on trends in supply and demand. For an operator that owns H100s, higher rates mean more revenue per GPU-hour sold. On-demand revenue reprices in both directions, though, so the exposure that lifted revenue in 2026 would cut it if rates fall. Revenue locked into an offtake contract at a fixed rate holds through either move.

How Lenders Read Contracted and On-demand H100 Revenue

Lenders and rating agencies give more credit to revenue fixed by contract than to revenue earned at the going on-demand rate. Two 2026 credit reports set out the reasoning.

Morningstar DBRS rated Lambda Compute I on Oct 2, 2026. It described the deal's leases as "take-or-pay capacity contracts with no exposure to volume risk." It also found no exposure to re-leasing risk. The collateral is NVIDIA GB300 and Vera Rubin (VR200) hardware, so the deal matters here only for how contracted revenue is treated.

A take-or-pay contract obliges the customer to pay for reserved capacity whether it uses that capacity or not. Volume risk falls on the customer, and the operator's revenue stays fixed for the term.

KBRA listed its key credit considerations for AI compute in July 2026 research. They include cash flow durability, contract structure, debt-service coverage, re-leasing and residual value. Each one turns on the same question: how much of the revenue is fixed, for how long and by whom.

Illustrative example. Assume one 8-GPU H100 server, Silicon Data's $3.06 neocloud reading from Nov 10, 2026, and 730 hours in a month. At full use, the server earns about $17,870 a month ($3.06 × 8 × 730). At 70% utilization on demand, it earns about $12,510.

A take-or-pay offtake contract at a fixed rate pays the same amount whatever the utilization. On-demand revenue moves with both the rate and utilization, so a lender modeling it has two variables to stress where the contract has none.

On-demand revenue still has a place in a financing. Documented on-demand revenue counts toward GPU-as-a-Service financing through USD.AI, priced to reflect its shorter commitment.

Three things drive how lenders value your revenue: the share under offtake contracts, their term and the customer's credit quality. A high H100 rental price on hourly sales lifts this month's revenue, while a three-year contracted rate lifts what a lender will count.

What the H100 Rate Means for Your Next Deployment

The H100 is the market's pricing benchmark, with neocloud readings of $3.06 from Silicon Data (Nov 10, 2026) and $3.10 from Ornn (Nov 10, 2026). Hyperscaler list prices sit far higher, rates rose through 2026, and lenders value contracted revenue above on-demand revenue.

Compute IQ scores every deal on a 100-point scale, using three inputs: the operator, the customer and the offtake contract. The score sets financeability and one of three pricing tiers, and deals scoring below 60 aren't financeable. On-demand revenue doesn't disqualify a deal, and it affects the score.

Approved collateral for GPU financing through USD.AI as of October 2026 is NVIDIA B200, B300, GB200, GB300 and Vera Rubin (VR200). AMD Instinct MI455X and MI430X are approved too. H100s aren't on that list, and their rate matters here as the benchmark for pricing offtake contracts on newer hardware.

Loans cover up to 80% of the deployment cost, with the operator contributing at least 20% as equity, over 3–5 years with straight-line monthly amortization. If you're pricing offtake contracts for a new deployment, Talk to USD.AI.

FAQ

These answers cover the questions operators and buyers ask most about H100 rental pricing. Each figure carries the date it describes.

How much does it cost to rent an H100 per hour?

An H100 costs about $3.06 to $3.10 per GPU-hour on neocloud benchmarks. Silicon Data read $3.06 on Nov 10, 2026, and Ornn settled at $3.10 the same day. Hyperscaler on-demand list prices work out to about $6.88 to $12.29 per GPU-hour, calculated from 8-GPU instance prices.

What is the difference between on-demand, spot and committed H100 pricing?

On-demand is the full hourly rate, spot is reclaimable capacity at a discount, and committed pricing trades a 1- or 3-year term for a lower rate. On Google Cloud's a3-highgpu-8g (read Nov 10, 2026), that's about $11.06 on demand, $6.59 spot and $4.86 on a 3-year commitment per GPU-hour, calculated.

Why did H100 rental prices rise in 2026?

H100 rental prices rose in 2026 because demand outran available capacity, and providers repriced. CoreWeave raised rates about 25% across SKUs in July 2026, and Nebius changed H100 prices on Oct 1, 2026. Our market analysis cited SemiAnalysis's 1-year H100 contract index, which rose from $1.70 per GPU-hour in October 2025 to $2.35 in March 2026.

Can you trade H100 rental futures?

Not yet. CME Group lists its H100 compute futures as pending regulatory review, so the contracts aren't trading.

Do lenders treat on-demand H100 revenue the same as contracted revenue?

No. Lenders and rating agencies give more credit to take-or-pay contracted revenue, which removes exposure to volume and re-leasing risk. On-demand revenue still counts, and Compute IQ prices it to reflect its shorter commitment.