AI compute market signals and learning
NewsletterContact
Market WatchBenchmarks
Enterprise Compute Cost$8.70 / H100e-hourH100e-hour means an H100-equivalent GPU hour, a normalized unit used to compare different GPUs and providers.Reliable GPU capacitySpot Compute Price$2.46 / H100e-hourH100e-hour means an H100-equivalent GPU hour, a normalized unit used to compare different GPUs and providers.Interruptible capacityCompute Market Rate$5.41 / H100e-hourH100e-hour means an H100-equivalent GPU hour, a normalized unit used to compare different GPUs and providers.Blended market rateEnterprise 7B Training Cost$73,235 / reference training runReference training costModel Serving Cost$3.51 / 1M tokensModel serving costGPU Availability78.1 / 100 scoreCapacity access

Last updated May 23, 2026 at 7:39 AM EDT

Market

Learn the AI compute market through benchmarks

AI compute pricing, workload cost, and capacity signals.

Learn the AI compute market through benchmarks.

How to read the market board

ComputeTape tracks directional AI compute market signals, not exchange-traded prices. Benchmarks combine public provider pricing, observed marketplace data, availability checks, and normalized H100-equivalent estimates. Use them as market indicators, not investment advice.

Compute Cost Benchmarks

Compute Cost Benchmarks

Live value cards with daily move and signal text. Last updated May 23, 2026 at 7:39 AM EDT.

AI Compute Market Rate

Value
$5.41 / H100e-hourH100e-hour means an H100-equivalent GPU hour, a normalized unit used to compare different GPUs and providers.
Daily move
n/a
Status
Stable
Signal
Headline benchmark for tracking whether AI compute is getting more expensive or cheaper.

Spot Compute Price

Value
$2.46 / H100e-hourH100e-hour means an H100-equivalent GPU hour, a normalized unit used to compare different GPUs and providers.
Daily move
n/a
Status
Looser
Signal
Estimated cost of cheaper interruptible GPU capacity. Useful for seeing whether excess GPU supply is available.

GPU Availability

Value
78.1 / 100 score
Daily move
n/a
Status
Healthy
Signal
Signal for how easy or hard it is to obtain AI GPU capacity.

Enterprise AI Compute Cost

Value
$8.70 / H100e-hourH100e-hour means an H100-equivalent GPU hour, a normalized unit used to compare different GPUs and providers.
Daily move
n/a
Status
Stable
Signal
Estimated cost of reliable enterprise-grade cloud GPU capacity.

Model Training Cost

Value
$73,235 / reference training run
Daily move
n/a
Status
Stable
Signal
Estimated infrastructure cost for a standard AI training workload.

Model Serving Cost

Value
$3.51 / 1M tokens
Daily move
n/a
Status
Stable
Signal
Estimated cost to serve AI model output at scale.

Providers

Provider Price Board

Snapshot values for normalized H100-equivalent pricing across tracked AI cloud providers.

ProviderSegmentGPUModeRaw $ / GPU-hrH100e $ / hrAdjusted H100e $ / hrAvailabilityConfidenceLast checked
AWSHyperscalersH100On Demand$12.29$12.29$7.22Quota GatedHighMay 23, 2026 at 6:00 AM EDT
AzureHyperscalersH100On Demand$12.29$12.29$12.90Quota GatedHighMay 23, 2026 at 6:00 AM EDT
CoreWeaveAI-native cloudsH100On Demand$6.16$6.16$6.34Available LimitedHighMay 23, 2026 at 6:00 AM EDT
CrusoeAI-native cloudsH100On Demand$3.90$3.90$3.90Available LimitedHighMay 23, 2026 at 6:00 AM EDT
LambdaDeveloper GPU cloudsH100On Demand$4.29$4.29$4.29AvailableHighMay 23, 2026 at 6:00 AM EDT
OracleHyperscalersH200On Demand$10.00$8.70$8.96Available LimitedHighMay 23, 2026 at 6:00 AM EDT
RunPod SecureDeveloper GPU cloudsH100On Demand$2.99$2.99$2.99Available LimitedMediumMay 23, 2026 at 6:00 AM EDT
NebiusAI-native cloudsn/an/an/an/a$5.41Available LimitedMediumn/a
Qualified Marketplace MedianQualified marketplace mediann/an/an/an/a$3.91Available LimitedMediumn/a

Access

Capacity Board

Availability and access signals for major AI compute providers. Updated weekly. Last refreshed May 23, 2026 at 6:00 AM EDT.

ProviderH100H200B200 / GB200MI300XLarge ClusterQuota FrictionStatus
AWSYesYesEmergingNoStrongHighConstrained
AzureYesYesEmergingSomeStrongHighConstrained
OracleYesYesEmergingNoStrongMediumLimited
CoreWeaveYesYesEmergingNoStrongMediumAvailable / tight
LambdaYesSomeNoNoModerateLow/MedAvailable
RunPodYesSomeNoSomeLowerLowVariable
CrusoeYesSomeEmergingNoStrongMediumExpanding
NebiusYesSomeEmergingNoStrongMediumExpanding