The on-demand public rental rate of AI compute, captured in a fixed window across a twenty-seven provider panel and reported in full. The anchor configuration is the H100 SXM 80GB. United States dollars, per GPU-hour.
| Index | Value | Basis |
|---|---|---|
| SVCI Public GPU Rate Index | 103.8 | Headline composite, base = Issue 001; weights 0.45 / 0.45 / 0.10 |
| Compute Opacity Index | 28.5 | 100 times (1 minus CompositeCoverage) |
| Hyperscaler Premium | 2.7352x | Tier 1 median over the pooled non-hyperscaler median |
| Neocloud Discount | 0.6344 | Tier 2 median relative to the Tier 1 median |
| Price Floor | 2.8200 | Lower-quartile headline-rate floor marker |
| Coverage Score | 0.7150 | Weighted panel coverage |
The Confidence Score remains unpublished pending the per-row confidence audit. All other headline readings are calculated from the locked Issue 003 data layer.
Receive the benchmark instead of hunting for it. The public issue stays open; the subscription delivers the Monday Compute Price Brief, the market reading that explains it, and buyer tools as they launch.
| Tier | Weight | Panel slots | N in window | Median USD |
|---|---|---|---|---|
| Tier 1, Hyperscalers | 0.45 | 4 | 4 | 10.5306 |
| Tier 2, GPU clouds and neoclouds | 0.45 | 18 | 9 | 3.8500 |
| Tier 3, Marketplaces and aggregators | 0.10 | 5 | 2 | 4.6200 |
Panel slots are the locked twenty-seven provider composition across three tiers. N in window is the count of in-window observations that resolve to a comparable public rate. Medians are computed from the official capture.
SVCI Issue 003 reviews all twenty-seven headline-panel providers. Fifteen published comparable fixed public H100 SXM 80GB rates that cleared the inclusion gate. Twelve were captured and excluded because the public surface was a floor, marketplace, aggregator, sales gate, signed-in-only rate, or incompatible product mode.
The headline index rose to 103.8 because Tier 2 public-price coverage fell from ten providers to nine. The provider that dropped from the comparable set was Voltage Park, whose launch-issue fixed rate of $1.99 no longer appeared as a public fixed price.
| Status | Count |
|---|---|
| Headline-eligible public rates | 15 |
| Captured and excluded | 12 |
| Pending | 0 |
| Total headline panel | 27 |
| Tier | Resolved | Coverage |
|---|---|---|
| Tier 1, Hyperscalers | 4 / 4 | 100.0% |
| Tier 2, GPU clouds | 9 / 18 | 50.0% |
| Tier 3, Marketplaces | 2 / 5 | 40.0% |
| Weighted Coverage Score | 0.7150 | |
| Compute Opacity Index | 28.5 |
The Rule I-16 publication gate is met: coverage is non-zero in all three tiers. The confidence audit remains separate from the public rate inclusion decision.
Among the nine Tier 2 providers that published a comparable fixed rate in both Issue 001 and Issue 003, the median was unchanged at $3.85. Together AI repriced lower, but remained above the median. The canonical rise to 103.8 is therefore a coverage and composition event, not a broad matched-provider price increase.
| Metric | Issue 001 | Issue 003 |
|---|---|---|
| Matched Tier 2 median, N=9 | $3.8500 | $3.8500 |
| Matched-panel SVCI | 100.0 | 100.0 |
Tier 3 remains a two-of-five fixed-rate observation. RunPod and Paperspace enter the tier median. Vast.ai, TensorDock, and Shadeform do not because dynamic marketplace medians, starting floors, and aggregator pass-through prices are not comparable fixed provider rates. The 0.10 weight remains locked.
Every published rate traces to a recorded source and evidence artifact. The public evidence manifest identifies the provider folders, screenshots, source notes, and SHA-256 hashes retained in the sealed package. The canonical data layer, calculation report, capture store, and reconciliation ledger are published with the issue.
The full screenshot archive remains the sealed public evidence package of record. Its manifest is published here so every retained artifact can be identified and hash-checked. Corrections remain permanently available in the corrections record.
The headline panel contains twenty-seven provider slots. Fifteen comparable fixed public H100 SXM 80GB rates enter the medians. Twelve captured rows are disclosed as non-inclusions. No row remains pending.
| ID | Provider | Status | Median | Reason | Note |
|---|---|---|---|---|---|
| E001 | AWS | Priced | Yes | Public on-demand p5.48xlarge, normalized per GPU by 8 | |
| E002 | Azure | Priced | Yes | Public ND96isr H100 v5 on-demand rate, normalized per GPU by 8 | |
| E003 | GCP | Priced | Yes | a3-highgpu-8g on-demand rate, normalized per GPU by 8 | |
| E004 | Oracle | Priced | Yes | BM.GPU.H100.8 public rate, normalized per GPU by 8 |
| ID | Provider | Status | Median | Reason | Note |
|---|---|---|---|---|---|
| E005 | CoreWeave | Priced | Yes | HGX H100 on-demand, normalized per GPU by 8 | |
| E006 | Crusoe | Priced | Yes | H100 80GB HGX on-demand per GPU | |
| E007 | Lambda | Priced | Yes | H100 SXM on-demand per GPU; tax footnote resolved | |
| E008 | Nebius | Priced | Yes | H100 on-demand GPU-hour rate | Rate remained above the Tier 2 median. |
| E009 | Together AI | Priced | Yes | GPU Clusters H100 on-demand, explicitly per GPU | Together AI fell from the launch issue but remained above the median. |
| E010 | Hyperstack | Priced | Yes | H100 SXM on-demand public rate | |
| E011 | Verda | Priced | Yes | H100 SXM5 80GB pay as you go | DataCrunch now operates under the Verda brand. |
| E012 | Voltage Park | Captured, excluded | No | Starting-from floor and contact pricing, no comparable fixed rate | |
| E013 | Fluidstack | Captured, excluded | No | No public fixed rate; pricing route returned not found | |
| E014 | Nscale | Captured, excluded | No | No public fixed rate; reserve flow only | |
| E015 | IREN | Captured, excluded | No | H100 80GB SKU shown, price is contact sales | |
| E016 | Applied Digital | Captured, excluded | No | Infrastructure and AI-factory product, no public GPU rental rate | Panel-fit review is carried in the governance register. |
| E017 | GMI Cloud | Captured, excluded | No | Dedicated pricing pages show from $2.00, a floor rather than a fixed rate | |
| E018 | Cudo Compute | Captured, excluded | No | Exact H100 SXM 80GB SKU shown, quote on request | |
| E019 | Spheron | Captured, excluded | No | Live marketplace floor and spot pricing, not a fixed provider rate | |
| E020 | Jarvislabs | Priced | Yes | H100 SXM 80GB 1x on-demand rate | Daylight recapture confirmed the original $2.69 rate. |
| E021 | Hyperbolic | Captured, excluded | No | No public comparable fixed rate; public, enterprise, and signed-in modes differ | |
| E022 | Atlas Cloud | Priced | Yes | H100 SXM 80GB fixed on-demand rate, per GPU |
| ID | Provider | Status | Median | Reason | Note |
|---|---|---|---|---|---|
| E023 | RunPod | Priced | Yes | Secure Cloud fixed H100 SXM on-demand rate | Current official page verified June 25; visual reference is from June 15. |
| E024 | Vast.ai | Captured, excluded | No | Dynamic marketplace median and range, excluded under F016 | |
| E025 | Paperspace | Priced | Yes | Paperspace H100 on-demand documentation rate; 8x cross-check | Legacy $2.24 commitment and DigitalOcean $3.39 product were excluded. |
| E026 | TensorDock | Captured, excluded | No | Starting-from $2.25 floor, not a fixed comparable rate | |
| E027 | Shadeform | Captured, excluded | No | Aggregator pass-through rates, no Shadeform-owned fixed rate |
| Category | Count |
|---|---|
| Priced, enter medians | 15 |
| Captured and excluded | 12 |
| Pending | 0 |
| Total headline panel | 27 |
By the Sorso View Editorial. Full construction at sorsoview.com/methodology.
Issue 002 is preserved at an SVCI level of 100.0. Issue 001 remains the base issue.