Weekly Pricing Pulse

Editor's note

Five AWS B300 price listings are newly visible in our October 11 catalog for EU (London), at $27.19 to $27.67 per GPU-hour for an eight-GPU instance. They are operating-system variants, detailed below. In the Ornn Compute Price Index (OCPI), A100 SXM4 made the largest weekly move of the four GPU groups: $0.88 on October 4 to $0.96 on October 11 (+9.1%). The October 4 value was the lowest weekly reading in our chart history, which begins May 18.

The lowest observed H100 SXM catalog price stayed at $3.90 per GPU-hour, while OCPI rose from $2.59 to $2.65 between the October 4 and October 11 settlements. That narrowed the catalog-to-index spread from $1.31 to $1.25 without a change in that catalog price. Across the nine providers we track, we found no verified public on-demand GPU price changes of more than 1%.

Eighteen apparent AWS changes were checked and are not price changes: AWS added new “without licenses” Windows with SQL Server listings for G7e instances in US East (Ohio), and the existing license-included prices did not change.

This week’s catalog contains 14,032 GPU price listings. We compared 14,021 directly with last week’s prices for the same provider, region and instance. The remaining eleven comprise five newly visible AWS listings and six returning Lambda listings. Three Lambda listings captured last week were absent this week. These Lambda counts describe region-specific records in its availability feed, including changes between named and unpublished locations. They do not establish product launches, retirements or current capacity.

We also applied the same updated Azure GPU-specification rules to both saved catalogs. This restates last week’s comparable-listing total from 14,001 to 14,024; the methodology below explains the adjustment. It is a change in how listings qualify for comparison, not a provider price change.

OCPI tracks the traded price of compute rather than quoted or listed prices. Its October 11 settlements and the full weekly comparison appear below. The public history values used here are rounded to cents, so percentage changes are approximate.

The AIForge Works Weekly Pricing Pulse is informational and is not a live quote, capacity signal, availability guarantee, workload-specific recommendation or personalized account analysis. For workload-specific options, use Cloud Advisor. To inspect neighboring SKUs, regions or terms, open the Multi-Cloud Pricing Analyzer. To compare your own GPU spend with market evidence that clearly identifies when prices were collected, use GPU Audit. Verify current price, capacity, availability, orderability, terms, fees and taxes directly with the provider before acting.

Ornn OCPI context

OCPI, the Ornn Compute Price Index, tracks the traded price of compute rather than quoted or listed prices, separate from the public on-demand catalog prices tracked by AIForge Works. Ornn measures on-demand rentals only; reserved-capacity, long-term and forward contracts are outside the scope of the indices. See Ornn’s methodology. These are settled daily readings through October 11, 2026.

All readings are US dollars per GPU-hour. Percentage changes are approximate because the public history endpoint returns prices rounded to cents. Orange marks price increases; green marks decreases.

Lowest observed catalog price: The lowest comparable public on-demand price found in our October 11 catalog across the nine providers we track.

Cohort qualifications: The A100 comparison uses the 80 GB SXM4 cohort. The Azure A100 listing is the 80 GB Standard_ND96amsr_A100_v4 at $4.10/GPU-hour; the 40 GB Standard_ND96asr_A100_v4 is excluded. The AWS H100 source identifies H100 80 GB but does not explicitly establish the SXM form factor, so that listing retains a cohort-specific source qualification. Some Lambda listings lack a published capacity region. The saved CoreWeave A100 label does not establish SXM4, so that entry is not included as a comparable price. OCPI and observed catalog prices are not live quotes, proof of capacity or orderability, or purchase recommendations.

Week at a glance

We compared 14,021 of this week’s 14,032 GPU price listings directly with last week’s prices for the same provider, region and instance. 11 listings did not have an exact prior match; the details below explain those differences. 0 verified per-GPU prices changed by more than 1%.

Collection-method correction: Azure RTX PRO 6000 (NCv6) sizes are now identified from Microsoft’s size table; whole-GPU sizes are compared per GPU-hour in every Azure region where they are listed; ¼- and ½-GPU, retired preview, announced and undocumented sizes stay in the catalog but outside per-GPU comparisons; last week’s comparison was restated under the same rule.

Collection-method correction: Crusoe location labels follow its published locations, including US East (Virginia) and US West (Nevada). Both saved weeks use the corrected labels, so the label correction is excluded from catalog movement.

Provider roll-up

No reportable exact GPU catalog movement was found for Azure, GCP, OCI, CoreWeave, Crusoe, Nebius or Vultr.

Newly visible listings

These five AWS records describe operating-system variants of the same eight-GPU instance in London, all with shared tenancy. Each has a distinct AWS SKU; they are not five different GPU models.

Catalog changes in detail

Restored listings

• Lambda: gpu_8x_a100 in Germany at $1.99 per GPU-hour.
• Lambda: gpu_1x_a100_sxm4 in India at $1.99 per GPU-hour.
• Lambda: gpu_8x_a100 in Israel at $1.99 per GPU-hour.
• Lambda: gpu_8x_a100 in Osaka, Japan at $1.99 per GPU-hour.
• Lambda: gpu_1x_h100_pcie in Utah, USA at $3.29 per GPU-hour.
• Lambda: gpu_8x_a100_80gb_sxm4 in Region not published at $2.79 per GPU-hour.

Listings no longer present

• Lambda: gpu_8x_a100 in Arizona, USA at $1.99 per GPU-hour.
• Lambda: gpu_8x_a100_80gb_sxm4 in Illinois, USA at $2.79 per GPU-hour.
• Lambda: gpu_1x_h100_pcie in Region not published at $3.29 per GPU-hour.

Catalog visibility does not establish a product launch, retirement, or available capacity.

Evidence and methodology

We compare public on-demand prices for the same provider, region and GPU configuration. Per-GPU prices and spreads are calculated from unrounded catalog rates, then displayed to the nearest cent. Recalculating from the rounded figures shown can therefore differ slightly. The lowest observed comparable catalog price excludes spot, reserved or committed terms, software-only charges and fractional GPUs. Catalog prices are saved evidence, not live quotes or guarantees of orderability.

Azure comparison recount: Applying the same GPU-specification rules to the October 4 catalog makes 100 previously excluded listings comparable and removes 77 previously included listings from the comparison, affecting 177 listings in total. Last week’s published 14,001 is therefore restated as 14,024 (14,001 + 100 − 77). This week’s 14,032 is eight higher than that restated baseline. The adjustment changes comparison eligibility; it is not a provider price change.

Vultr comparisons: The October 11 saved catalog yielded no confirmed comparable public on-demand price for any of the four GPU cohorts shown here. The A100 evidence includes a listing whose public on-demand status is unproved and entries without the GPU specifications needed for this comparison. “N/A” means that we could not establish a comparable price from this capture; it does not establish whether Vultr currently has capacity or offers the GPU under another purchasing arrangement. Missing location information is a separate limitation.

Lambda availability: The six restored and three absent records are counted by provider, region and instance. They include the H100 PCIe listing at $3.29 per GPU-hour, which was reported without a region last week and in Utah this week, and the A100 80 GB SXM4 listing at $2.79, which was reported in Illinois last week and without a region this week. The eight-GPU A100 listing at $1.99 was absent from Arizona and restored in Germany, Israel and Osaka. Matching SKU and price across regions does not establish that these are the same capacity, a physical relocation or a labeling error. The counts describe feed visibility at the two collection times, not distinct product launches, retirements or capacity available now.

AWS G7e review: 18 AWS G7e listings in US East (Ohio) were reviewed. Their GPU configurations match AWS’s published specifications. The apparent changes came from new “without licenses” Windows with SQL Server listings that share an identifier with existing license-included listings, whose prices did not change. They are not counted as price changes.

Next Pulse: Week of October 19, 2026