Independent market signal · methodology v1.0.0

GPU Cloud Price Index

A provider-balanced view of published GPU cloud prices. Every provider receives equal weight, every rate is normalized to USD per GPU-hour, and the historical line uses one fixed matched cohort so expanding coverage cannot masquerade as a price move.

  • 866 official, within-SLA and comparable offers at generation
  • 15 providers and 25 exact model labels
  • Source hashes and limitations travel with the API response

Fixed-cohort history

A price signal protected from catalog-growth bias

The baseline is 100 on 2026-08-02. Each point uses the same 61 offer identities across 5 providers. A flat line means the published provider-balanced index level stayed the same for that cohort at the displayed precision. Aggregation or rounding can hide changes in individual normalized rates; the index alone does not establish unchanged quotes or capacity.

SaaS Sentinel GPU Cloud Price Index historyProvider-balanced matched-cohort index from 2026-08-02 to 2026-11-01, with baseline 100.101.0100.099.008-0208-0408-0508-0608-0708-0808-0908-1008-1108-1208-1308-1508-1608-1708-1808-2108-2208-2308-2409-0709-0809-0910-0410-0710-0810-0910-1010-1310-2910-3010-3111-01
Snapshot days32
Raw observations29,679
Provider P25–P75$1.5186–$3.7315
Official / within-SLA at generation100% / 100%

Model-level market map

Provider-balanced medians by GPU model

These medians first collapse regions and SKUs within each provider, then give each provider one equal vote. The raw minimum and maximum remain visible so the summary never hides dispersion.

GPU modelHistoryAlertsOffersProvidersBalanced medianProvider P25–P75Raw rangeAvailability evidenced
RTX 4000 Ada—Follow alerts2971$0.74
$540.20 at 730 GPU-hours
$0.74–$0.74$0.52–$1.53100%
H100View historyFollow alerts1918$3.845
$2,806.85 at 730 GPU-hours
$2.955–$4.795$0.79–$23.3261.6%
RTX 6000—Follow alerts1321$1.50
$1,095.00 at 730 GPU-hours
$1.50–$1.50$1.50–$1.50100%
A100View historyFollow alerts1246$2.345
$1,711.85 at 730 GPU-hours
$1.9175–$2.945$0.6788–$8.19251.6%
L4—Follow alerts242$1.1649
$850.3748 at 730 GPU-hours
$1.0223–$1.3074$0.8798–$1.9225100%
L40SView historyFollow alerts186$1.6061
$1,172.4472 at 730 GPU-hours
$1.5175–$1.663$1.29–$1.8077.8%
Tesla V100S—Follow alerts181$0.8825
$644.225 at 730 GPU-hours
$0.8825–$0.8825$0.88–$1.5582100%
H200View historyFollow alerts136$4.14
$3,022.20 at 730 GPU-hours
$3.615–$4.425$0.79–$6.2561.5%
B200View historyFollow alerts74$7.645
$5,580.85 at 730 GPU-hours
$6.2038–$8.6675$0.99–$8.8757.1%
B300—Follow alerts61$5.245
$3,828.85 at 730 GPU-hours
$5.245–$5.245$0.99–$9.50100%
H100-PCIe—Follow alerts41$3.2025
$2,337.7914 at 730 GPU-hours
$3.2025–$3.2025$3.2025–$3.2025100%
A4000—Follow alerts31$0.76
$554.80 at 730 GPU-hours
$0.76–$0.76$0.76–$0.76100%
H100-SXM—Follow alerts31$3.5669
$2,603.8662 at 730 GPU-hours
$3.5669–$3.5669$3.5374–$3.6978100%
P4000—Follow alerts31$0.51
$372.30 at 730 GPU-hours
$0.51–$0.51$0.51–$0.51100%
P5000—Follow alerts31$0.78
$569.40 at 730 GPU-hours
$0.78–$0.78$0.78–$0.78100%
P6000—Follow alerts31$1.10
$803.00 at 730 GPU-hours
$1.10–$1.10$1.10–$1.10100%
RTX 4000—Follow alerts31$0.56
$408.80 at 730 GPU-hours
$0.56–$0.56$0.56–$0.56100%
V100—Follow alerts31$2.30
$1,679.00 at 730 GPU-hours
$2.30–$2.30$2.30–$2.30100%
A5000—Follow alerts21$1.38
$1,007.40 at 730 GPU-hours
$1.38–$1.38$1.38–$1.38100%
A6000—Follow alerts21$1.89
$1,379.70 at 730 GPU-hours
$1.89–$1.89$1.89–$1.89100%
RTX 6000 Ada—Follow alerts22$1.535
$1,120.55 at 730 GPU-hours
$1.5175–$1.5525$1.50–$1.570%
RTX PRO 6000—Follow alerts21$1.295
$945.35 at 730 GPU-hours
$1.295–$1.295$0.79–$1.80100%
B300-SXM—Follow alerts11$8.379
$6,116.67 at 730 GPU-hours
$8.379–$8.379$8.379–$8.379100%
P100—Follow alerts11$1.3641
$995.7937 at 730 GPU-hours
$1.3641–$1.3641$1.3641–$1.3641100%
RTX 5000—Follow alerts11$0.82
$598.60 at 730 GPU-hours
$0.82–$0.82$0.82–$0.82100%

Citation boundary

What you may conclude from this index

Comparable price level

The index describes published normalized compute rates. It is suitable for tracking direction and cross-model dispersion when the date, cohort and unit are preserved.

Not a transaction index

It is not weighted by provider size, customer volume or available capacity. Negotiated discounts, taxes, storage, egress and support can change an invoice materially.

Fail-closed history

Sparse dates are withheld. New providers do not rewrite old points, evidence-only refreshes do not become price alerts, and a listed price never proves stock.

Reuse with provenance

Build on the index, not a screenshot

Use the versioned API and retain the generated timestamp, exact validity deadline, at-generation scope and methodology link. At the deadline the snapshot is still current; one millisecond later it is historical.