Catalog floors held across the 9 tracked providers from Nov 2 to Nov 8 (per GPU-hour: H100 SXM $3.90 and H200 $4.29 at Crusoe, B200 $6.69 at Lambda, A100 SXM4 $2.30 at Crusoe), while the OCPI market index rose for H100 SXM, H200 and B200. The index now sits well above the cheapest published H200 and B200 prices, and Crusoe, which sets the H100 SXM and H200 floors, has announced December 1 increases. These are catalog observations, not live quotes.
Crusoe has announced higher on-demand prices starting December 1: H100 from $3.90 to $4.50 per GPU-hour, H200 from $4.29 to $5.40 and B300 from $8.96 to $9.90, plus an $8.50 B200 price where it now says “Contact sales”. Its A100 prices are unchanged. If other providers’ prices hold, the H100 SXM floor would then move to Lambda at $3.99, and the cheapest H200 price would rise to $5.40, matching Nebius.
Across the November 2 to November 8 GPU pricing comparison, 0 matched public on-demand GPU prices across the providers we track changed by more than 1%. We compared 13,904 of this week’s 13,935 GPU price listings directly with last week’s prices for the same provider, region and GPU configuration. The other 31 listings had no exact prior match: 24 were newly visible and 7 had appeared in an earlier catalog.
Over the 7 days from November 1 to November 8, 2026, Ornn’s daily OCPI index (US dollars per GPU-hour) moved as follows: H100 SXM moved from $3.02 to $3.37 (up 11.6%); H200 moved from $6.21 to $6.82 (up 9.8%); B200 moved from $9.44 to $9.94 (up 5.3%); A100 SXM4 moved from $1.06 to $1.03 (down 2.8%).
The market index is catching up to the cheapest listed H100 SXM price, and pulling further ahead of the cheapest H200 and B200 prices. Crusoe's $3.90 H100 SXM price is now just $0.53 above the index, down from $0.88 a week earlier. The cheapest H200 price ($4.29, Crusoe) is now $2.53 below the index, up from $1.92, and the cheapest B200 price ($6.69, Lambda) is $3.25 below, up from $2.75. For all three, November 8 is the highest weekly index reading on our chart, which starts on May 18: H200 has climbed from about $4.50 in late August and early September, and B200 from about $6 at the end of August. Crusoe’s announced December 1 increases, above, point in the same upward direction.
The index and our catalog measure different things. The OCPI (Ornn Compute Price Index) tracks the prices at which GPU time actually traded, while our catalog shows the prices providers publish for specific regions and configurations. Read the comparison as market context, not as a quote for the same offer.
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 dated market evidence, use GPU Audit. Verify current price, capacity, availability, orderability, terms, fees and taxes directly with the provider before acting.
Ornn OCPI context
OCPI is the Ornn Compute Price Index (Ornn x ICE), a third-party index that tracks the traded price of compute rather than quoted or listed prices, shown as market context. Ornn measures on-demand rentals only; reserved-capacity, long-term and forward contracts are outside the scope of the indices. These are daily readings through November 8, 2026. Ornn index source.
GPU family
November 1
November 8
7-day change
October 9
30-day change
H100 SXM
$3.02
$3.37
+11.6%
$2.85
+18.2%
H200
$6.21
$6.82
+9.8%
$5.68
+20.1%
B200
$9.44
$9.94
+5.3%
$7.47
+33.1%
A100 SXM4
$1.06
$1.03
-2.8%
$0.94
+9.6%
All readings are US dollars per GPU-hour. Catalog and OCPI prices are displayed rounded to cents. Spreads and percentage changes are approximate and may differ slightly when recomputed from the displayed rounded values.
Lowest observed catalog price: The lowest comparable public on-demand price found in our November 8 catalog across the nine providers we track.
Ornn OCPI context
OCPI and Lowest Observed Catalog Prices
Saved public catalog snapshot · November 8, 2026
Exact GPU cohort comparison
H100 SXM
Public on-demand catalog prices for the same GPU group, normalized per GPU-hour. The lowest price may come from a different region or instance size than the offer you need.
OCPI-H100 SXM
Ornn x ICE · November 8, 2026
$3.37/ GPU / hr
Lowest observed catalog price
Crusoe - Global pricing · November 8 catalog
$3.90/ GPU / hr
Above OCPI
Catalog checked
Nov 8
All 9 provider files scanned
Providers with a comparable catalog price 8 / 9
Lowest observed catalog prices by providerNovember 8 catalog · lowest price first within each group
ProviderLowest catalog priceSpread vs OCPIInstance / SKURecorded region
Hyperscalers4/4 matched
AWSHyperscalers
$6.88/GPU/hr
+$3.51/ +104.2%
p5.48xlarge8 GPUs
US East (N. Virginia)
OCIHyperscalers
$10.00/GPU/hr
+$6.63/ +196.7%
B98415-us-ashburn-18 GPUs
us-ashburn-1
GCPHyperscalers
$11.06/GPU/hr
+$7.69/ +228.2%
gcp-machine-a3-highgpu-8g8 GPUs
us-west1
AzureHyperscalers
$12.29/GPU/hr
+$8.92/ +264.7%
Standard_ND96isr_H100_v58 GPUs
westus3
Neoclouds4/5 matched
CrusoeNeoclouds
$3.90/GPU/hr
+$0.53/ +15.7%
h100-80gb-sxm-ib.8x8 GPUs
Global pricing
LambdaNeoclouds
$3.99/GPU/hr
+$0.62/ +18.4%
gpu_8x_h100_sxm58 GPUs
Region not published
NebiusNeoclouds
$4.50/GPU/hr
+$1.13/ +33.5%
gpu-h100-sxm/1gpu-16vcpu-200gb1 GPU
eu-north1
CoreWeaveNeoclouds
$6.16/GPU/hr
+$2.79/ +82.6%
hgx-h1008 GPUs
global
VultrNeoclouds
N/A
No comparable price
No confirmed comparable price in this saved catalog.
N/A
Spread = catalog price minus the November 8 OCPI of $3.37/GPU-hour; percentage = spread divided by OCPI. Orange is above OCPI; green is below. Catalog and OCPI prices are displayed rounded to cents. Spreads and percentage changes are approximate and may differ slightly when recomputed from the displayed rounded values. This compares catalog prices with an index, not region-matched quotes.
Lowest observed catalog price: The lowest comparable public on-demand price found in our November 8 catalog across the nine providers we track.
Sources and limits: The check scanned 6,412,199 pricing records and found 478 comparable price listings across the four GPU groups. GPU specifications, source-file hashes and per-GPU arithmetic were checked. Prices exclude spot, reserved/committed and fractional-GPU offers. These are saved-catalog prices; they do not establish current capacity or orderability.
Exact GPU cohort comparison
H200
Public on-demand catalog prices for the same GPU group, normalized per GPU-hour. The lowest price may come from a different region or instance size than the offer you need.
OCPI-H200
Ornn x ICE · November 8, 2026
$6.82/ GPU / hr
Lowest observed catalog price
Crusoe - Global pricing · November 8 catalog
$4.29/ GPU / hr
Below OCPI
Catalog checked
Nov 8
All 9 provider files scanned
Providers with a comparable catalog price 7 / 9
Lowest observed catalog prices by providerNovember 8 catalog · lowest price first within each group
ProviderLowest catalog priceSpread vs OCPIInstance / SKURecorded region
Hyperscalers4/4 matched
AWSHyperscalers
$7.91/GPU/hr
+$1.09/ +16.0%
p5en.48xlarge8 GPUs
US East (N. Virginia)
OCIHyperscalers
$10.00/GPU/hr
+$3.18/ +46.6%
B110519-us-ashburn-18 GPUs
us-ashburn-1
AzureHyperscalers
$10.60/GPU/hr
+$3.78/ +55.4%
Standard_ND96isr_H200_v58 GPUs
westus2
GCPHyperscalers
$10.60/GPU/hr
+$3.78/ +55.4%
gcp-machine-a3-ultragpu-8g8 GPUs
us-central1
Neoclouds3/5 matched
CrusoeNeoclouds
$4.29/GPU/hr
-$2.53/ -37.1%
h200-141gb-sxm-ib.8x8 GPUs
Global pricing
NebiusNeoclouds
$5.40/GPU/hr
-$1.42/ -20.8%
gpu-h200-sxm/1gpu-16vcpu-200gb1 GPU
eu-north1
CoreWeaveNeoclouds
$6.31/GPU/hr
-$0.52/ -7.6%
hgx-h2008 GPUs
global
LambdaNeoclouds
N/A
No comparable price
No confirmed comparable price in this saved catalog.
N/A
VultrNeoclouds
N/A
No comparable price
No confirmed comparable price in this saved catalog.
N/A
Spread = catalog price minus the November 8 OCPI of $6.82/GPU-hour; percentage = spread divided by OCPI. Orange is above OCPI; green is below. Catalog and OCPI prices are displayed rounded to cents. Spreads and percentage changes are approximate and may differ slightly when recomputed from the displayed rounded values. This compares catalog prices with an index, not region-matched quotes.
Lowest observed catalog price: The lowest comparable public on-demand price found in our November 8 catalog across the nine providers we track.
Sources and limits: The check scanned 6,412,199 pricing records and found 478 comparable price listings across the four GPU groups. GPU specifications, source-file hashes and per-GPU arithmetic were checked. Prices exclude spot, reserved/committed and fractional-GPU offers. These are saved-catalog prices; they do not establish current capacity or orderability.
Exact GPU cohort comparison
B200
Public on-demand catalog prices for the same GPU group, normalized per GPU-hour. The lowest price may come from a different region or instance size than the offer you need.
OCPI-B200
Ornn x ICE · November 8, 2026
$9.94/ GPU / hr
Lowest observed catalog price
Lambda - Region not published · November 8 catalog
$6.69/ GPU / hr
Below OCPI
Catalog checked
Nov 8
All 9 provider files scanned
Providers with a comparable catalog price 5 / 9
Lowest observed catalog prices by providerNovember 8 catalog · lowest price first within each group
ProviderLowest catalog priceSpread vs OCPIInstance / SKURecorded region
Hyperscalers2/4 matched
OCIHyperscalers
$14.00/GPU/hr
+$4.06/ +40.8%
B110978-us-ashburn-18 GPUs
us-ashburn-1
AWSHyperscalers
$14.24/GPU/hr
+$4.30/ +43.3%
p6-b200.48xlarge8 GPUs
US East (N. Virginia)
AzureHyperscalers
$27.04/GPU/hr
Not comparable
Reference only, not comparable: Azure lists no B200 VM. Its Blackwell offer is GB200 (ND GB200 v6: 2 Grace CPUs + 4 Blackwell GPUs), from $108.16/hr per VM on Linux on-demand, about $27.04 per GPU-hour.Standard_ND128isr_NDR_GB200_v6 · 4 GPUs
eastus, eastus2, westus2, westus3
GCPHyperscalers
$8.06 · $16.11*/GPU/hr
Not comparable
Reference only, not comparable: Google’s pricing page lists no on-demand price for its B200 machine (A4). $8.06 is its per-GPU B200 rate for DWS Flex-start (Dynamic Workload Scheduler); * see the note below the table.A4 B200 via DWS Flex-start · per GPU
$8.06 rate: same in 41 regions
Neoclouds3/5 matched
LambdaNeoclouds
$6.69/GPU/hr
-$3.25/ -32.7%
gpu_8x_b200_sxm68 GPUs
Region not published
NebiusNeoclouds
$8.50/GPU/hr
-$1.44/ -14.5%
gpu-b200-sxm/1gpu-20vcpu-224gb1 GPU
us-central1
CoreWeaveNeoclouds
$8.60/GPU/hr
-$1.34/ -13.5%
nvidia-b2008 GPUs
global
CrusoeNeoclouds
N/A
No comparable price
No confirmed comparable price in this saved catalog.
N/A
VultrNeoclouds
N/A
No comparable price
No confirmed comparable price in this saved catalog.
N/A
Spread = catalog price minus the November 8 OCPI of $9.94/GPU-hour; percentage = spread divided by OCPI. Orange is above OCPI; green is below. Catalog and OCPI prices are displayed rounded to cents. Spreads and percentage changes are approximate and may differ slightly when recomputed from the displayed rounded values. This compares catalog prices with an index, not region-matched quotes.
* In Google’s billing catalog as we captured it on November 8, a B200 SKU labeled OnDemand is priced at $16.11 per GPU-hour in its US regions (us-central1, us-east1 and us-west1), but Google’s public pricing page shows no on-demand price for its B200 (A4) machine. The DWS Flex-start rate, $8.06 after rounding to cents, is exactly half of that. Neither price enters this comparison.
Lowest observed catalog price: The lowest comparable public on-demand price found in our November 8 catalog across the nine providers we track.
Sources and limits: The check scanned 6,412,199 pricing records and found 478 comparable price listings across the four GPU groups. GPU specifications, source-file hashes and per-GPU arithmetic were checked. Prices exclude spot, reserved/committed and fractional-GPU offers. These are saved-catalog prices; they do not establish current capacity or orderability.
Exact GPU cohort comparison
A100 SXM4
Public on-demand catalog prices for the same GPU group, normalized per GPU-hour. The lowest price may come from a different region or instance size than the offer you need.
OCPI-A100 SXM4
Ornn x ICE · November 8, 2026
$1.03/ GPU / hr
Lowest observed catalog price
Crusoe - Global pricing · November 8 catalog
$2.30/ GPU / hr
Above OCPI
Catalog checked
Nov 8
All 9 provider files scanned
Providers with a comparable catalog price 6 / 9
Lowest observed catalog prices by providerNovember 8 catalog · lowest price first within each group
ProviderLowest catalog priceSpread vs OCPIInstance / SKURecorded region
Hyperscalers4/4 matched
AWSHyperscalers
$3.43/GPU/hr
+$2.40/ +233.1%
p4de.24xlarge8 GPUs
US East (N. Virginia)
OCIHyperscalers
$4.00/GPU/hr
+$2.97/ +288.3%
B95907-us-ashburn-18 GPUs
us-ashburn-1
AzureHyperscalers
$4.10/GPU/hr
+$3.07/ +297.7%
Standard_ND96amsr_A100_v48 GPUs
westus3
GCPHyperscalers
$5.07/GPU/hr
+$4.04/ +392.1%
gcp-machine-a2-ultragpu-1g1 GPU
us-central1
Neoclouds2/5 matched
CrusoeNeoclouds
$2.30/GPU/hr
+$1.27/ +123.3%
a100-80gb-sxm-ib.8x8 GPUs
Global pricing
LambdaNeoclouds
$2.79/GPU/hr
+$1.76/ +170.9%
gpu_8x_a100_80gb_sxm48 GPUs
Region not published
CoreWeaveNeoclouds
N/A
No comparable price
No confirmed comparable price in this saved catalog.
N/A
NebiusNeoclouds
N/A
No comparable price
No confirmed comparable price in this saved catalog.
N/A
VultrNeoclouds
N/A
No comparable price
No confirmed comparable price in this saved catalog.
N/A
Spread = catalog price minus the November 8 OCPI of $1.03/GPU-hour; percentage = spread divided by OCPI. Orange is above OCPI; green is below. Catalog and OCPI prices are displayed rounded to cents. Spreads and percentage changes are approximate and may differ slightly when recomputed from the displayed rounded values. This compares catalog prices with an index, not region-matched quotes.
Lowest observed catalog price: The lowest comparable public on-demand price found in our November 8 catalog across the nine providers we track.
Sources and limits: The check scanned 6,412,199 pricing records and found 478 comparable price listings across the four GPU groups. GPU specifications, source-file hashes and per-GPU arithmetic were checked. Prices exclude spot, reserved/committed and fractional-GPU offers. These are saved-catalog prices; they do not establish current capacity or orderability.
OCPI vs Lowest Observed Catalog Prices by Week
OCPI index prices alongside the lowest observed catalog price for each GPU. Historical points retain their dated comparison basis.
OCPI indexCatalog price
OCPI indexCatalog price
H100 SXM
OCPI$3.37Catalog$3.90(Crusoe)
$0.53 above OCPI
OCPI indexCatalog price
H200
OCPI$6.82Catalog$4.29(Crusoe)
$2.53 below OCPI
OCPI indexCatalog price
B200
OCPI$9.94Catalog$6.69(Lambda)
$3.25 below OCPI
OCPI indexCatalog price
A100 SXM4
OCPI$1.03Catalog$2.30(Crusoe)
$1.27 above OCPI
The November 8 point pairs that day’s OCPI reading with the November 8 catalog minimum. Earlier points retain their published evidence.
The index’s daily path, November 1 to November 8
H100 SXM: it reached $3.53 on November 3, its highest reading in the three months of daily history Ornn publishes. It then swung between $3.10 and $3.52 before ending the week at $3.37.
H200: it also peaked on November 3, at $7.31, its highest reading in that history, then fell 12.6% the next day to $6.39 and ended the week at $6.82.
B200: it held between $9.41 and $9.53 through November 6, then rose to $9.94 on November 8, above its October 28 peak of $9.85 and its highest reading in that history.
A100 SXM4: it drifted down from $1.08 on November 2 to $1.03 on November 8, its lowest reading since October 19.
Published prices versus the index
The index reading and the catalog are both dated November 8 this week, but they are not matched by region, so treat these gaps as directional.
H100 SXM: the cheapest published price ($3.90) sits about 16% above the index, down from about 29% on November 1, because the index rose.
A100 SXM4: the cheapest published price is still more than twice the index.
H200 and B200: the gap runs the other way. Their cheapest published prices sit about 37% and 33% below the index.
Week at a glance
We compared 13,904 of this week’s 13,935 GPU price listings directly with last week’s prices for the same provider, region and GPU configuration. 31 listings did not have an exact prior match; the details below explain those differences. 0 matched prices changed by more than 1%.
Prior: 13,91813,904 matched + 14 no longer present = 13,918
Crusoe publishes one on-demand rate per GPU model, with no regional pricing. Crusoe prices are shown as global, and a Crusoe rate change is counted once.
Provider roll-up
Provider
Class
Newly visible
Restored
Newly absent
Price moves >1%
AWS
hyperscaler
24
0
0
0
Lambda
neocloud
0
7
14
0
Totals
24
7
14
0
No reportable exact GPU catalog movement was found for Azure, GCP, OCI, CoreWeave, Crusoe, Nebius or Vultr.
Newly visible listings
AWS
24 newly visible price listings in 4 groups
p6-b300.48xlarge in AWS GovCloud (US-West). B300 288 GB. 5 price variants. Public on-demand: from $170.90/instance-hour ($21.36/GPU-hour; OS: Linux; tenancy: Shared) to $174.76/instance-hour ($21.84/GPU-hour; OS: Red Hat Enterprise Linux with HA; tenancy: Shared).
p5.48xlarge in EU (Frankfurt). H100 80 GB. 5 price variants. Public on-demand: from $68.80/instance-hour ($8.60/GPU-hour; OS: Linux; tenancy: Shared) to $72.66/instance-hour ($9.08/GPU-hour; OS: Red Hat Enterprise Linux with HA; tenancy: Shared).
p5.4xlarge in EU (Frankfurt). H100 80 GB. 9 price variants. Public on-demand: from $8.60/instance-hour ($8.60/GPU-hour; OS: Linux; tenancy: Shared) to $15.34/instance-hour ($15.34/GPU-hour; OS: Windows; tenancy: Shared; preinstalled software: SQL Ent).
p5e.48xlarge in US West (Oregon). H200 141 GB. 5 price variants. Public on-demand: from $63.30/instance-hour ($7.91/GPU-hour; OS: Linux; tenancy: Shared) to $67.16/instance-hour ($8.39/GPU-hour; OS: Red Hat Enterprise Linux with HA; tenancy: Shared).
Catalog changes in detail
Restored or expanded regional observations
Lambda: gpu_1x_a6000 in Region not published at $1.09 per GPU-hour.
Lambda: gpu_1x_h100_pcie in Region not published at $3.29 per GPU-hour.
Lambda: gpu_1x_h100_sxm5 in Region not published at $4.29 per GPU-hour.
Lambda: gpu_2x_h100_sxm5 in Region not published at $4.19 per GPU-hour.
Lambda: gpu_4x_h100_sxm5 in Region not published at $4.09 per GPU-hour.
Lambda: gpu_8x_a100 in Region not published at $1.99 per GPU-hour.
Lambda: gpu_8x_a100_80gb_sxm4 in Region not published at $2.79 per GPU-hour.
Prior regional observations not seen this week
Lambda: gpu_8x_a100 in Arizona, USA at $1.99 per GPU-hour.
Lambda: gpu_8x_a100 in California, USA at $1.99 per GPU-hour.
Lambda: gpu_1x_h100_sxm5 in Central Texas, USA at $4.29 per GPU-hour.
Lambda: gpu_2x_h100_sxm5 in Central Texas, USA at $4.19 per GPU-hour.
Lambda: gpu_8x_a100 in Germany at $1.99 per GPU-hour.
Lambda: gpu_8x_a100_80gb_sxm4 in Illinois, USA at $2.79 per GPU-hour.
Lambda: gpu_8x_a100 in Israel at $1.99 per GPU-hour.
Lambda: gpu_1x_a6000 in North Texas, USA at $1.09 per GPU-hour.
Lambda: gpu_1x_h100_sxm5 in North Texas, USA at $4.29 per GPU-hour.
Lambda: gpu_2x_h100_sxm5 in North Texas, USA at $4.19 per GPU-hour.
Lambda: gpu_4x_h100_sxm5 in North Texas, USA at $4.09 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 Virginia, USA at $2.79 per GPU-hour.
Same SKUs under different region labels
The following SKUs appear in both regional observation lists at the same recorded per-GPU price, GPU label and GPU count. These are separate region-qualified observations, not evidence that the underlying SKUs disappeared. The records alone do not establish whether the locations changed or their labels changed; unknown regions remain unknown. The counts retain the original exact comparison keys.
Lambda: gpu_1x_a6000; prior labels: North Texas, USA; current labels: Region not published.
Lambda: gpu_1x_h100_pcie; prior labels: Utah, USA; current labels: Region not published.
Lambda: gpu_1x_h100_sxm5; prior labels: Central Texas, USA, North Texas, USA; current labels: Region not published.
Lambda: gpu_2x_h100_sxm5; prior labels: Central Texas, USA, North Texas, USA; current labels: Region not published.
Lambda: gpu_4x_h100_sxm5; prior labels: North Texas, USA; current labels: Region not published.
Lambda: gpu_8x_a100; prior labels: Arizona, USA, California, USA, Germany, Israel, Osaka, Japan; current labels: Region not published.
Lambda: gpu_8x_a100_80gb_sxm4; prior labels: Illinois, USA, Virginia, USA; current labels: Region not published.
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. Prices are normalized per GPU-hour and rounded to cents. 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.