The Index, Issue 008, August 16 2026

Issue 008.

A full fresh sweep of the locked twenty-seven-provider H100 SXM 80GB-class panel. United States dollars, normalized per GPU-hour.

Issue archivePrevious issueMethodology
SVCI Public GPU Rate Index
104.1
Up 0.6 points from Issue 007.
Compute Opacity Index
26.0
Down 2.5 points as one additional Tier 2 entry qualifies.
Coverage Score
0.7400
Sixteen priced and eleven freshly captured exclusions.
Companion Dollar Basket
$6.9445
Up from $6.9183.
The readings

Headline state.

Issue 008 index values
IndexIssue 008Issue 007Change
SVCI Public GPU Rate Index104.1103.5+0.6 points
Compute Opacity Index26.028.5-2.5 points
Coverage Score0.74000.7150+0.0250
Companion Dollar Basket$6.9445$6.9183+$0.0262
Price Floor$3.075$2.97+$0.105
Hyperscaler Premium2.7176x2.7352x-0.0176x
Neocloud Discount0.63200.6344-0.0024

The Tier 2 median rose to $3.875 after Lightning AI qualified under the successor-continuous E012 panel entry. Tier 3 returned to its Issue 001 base median as RunPod moved back to $3.29.

Free reader tools

Get the next SVCI issue when it publishes.

Receive the benchmark, the market reading that explains it, and buyer tools as they launch.

Check a GPU quote
Tier medians

The three tiers.

Median on-demand rate by tier, USD per GPU-hour
TierWeightPanel slotsN pricedMedian USD
Tier 1, Hyperscalers0.454410.53065
Tier 2, GPU clouds and neoclouds0.4518103.87500
Tier 3, Marketplaces and aggregators0.10524.62000
What moved

One admissible public rate changed.

Lightning AI, formerly Voltage Park. E012 now qualifies at $4.50 per GPU-hour for a regular single-GPU H100 80 GB offer. The prior excluded monitoring figure is not an admissible movement basis, so no price movement is stated for E012. Interruptible rates do not enter the panel.

RunPod. The Secure Cloud H100 SXM rate moved from $2.99 to $3.29, +$0.30 or +10.03%.

The SVCI increased to 104.1, while coverage improved because E012 moved from excluded to priced.

Full sweep

No carried exclusions.

Issue 008 is a Track 1 full sweep. All twenty-seven panel entries were freshly reviewed. Eleven providers were freshly captured and excluded. No Issue 007 exclusion disposition was reused as the Issue 008 evidence basis.

Vast.ai and Spheron remain outside the median as dynamic marketplaces. TensorDock is excluded because independent hosts set staggered prices and low-end availability is not guaranteed. Shadeform is excluded as an aggregator of third-party cloud prices. ChronoScale and other sales-led providers without a fixed public comparable rate are captured but not estimated.

Method in one paragraph

The governing rule.

The SVCI observes logged-out, public, fixed, on-demand rates for the H100 SXM 80GB-class anchor. Spot, interruptible, reserved, committed, starting-at, marketplace-floor, supplier-variable, quote-only and aggregator passthrough figures do not substitute for a comparable provider-owned rate. Full-machine prices are normalized to per-GPU in recorded arithmetic. Issue 008 requires no new methodology amendment.

Evidence and downloads

How to check this issue.

The structured data, calculation report, public ledger and selected evidence manifest are published below. The complete byte-level screenshot archive remains private.

September 14, 2026 correction: The linked data v1.1 and calculation report v1.2 remove the invalid E012 percentage movement and align with the issue narrative. Original v1.0 files remain available by their original URLs. The retained public capture ledger references the original data v1.0; corrected data v1.1 supersedes it for E012 movement and source metadata. No headline value changed. Correction record.

DataCSVDataJSONDerivationCalculation reportGovernancePublic capture ledgerEvidencePublic manifestMethodologyA1-R1
Together AI Issue 008 public evidence
NVIDIA H100 SXM 80 GB and public on-demand rate of $3.99.
Lightning AI Issue 008 public evidence
Single H100 80 GB regular rate of $4.50 on a logged-out public surface.
ChronoScale Issue 008 public evidence
Public site routes commercial engagement through a contact form and posts no fixed rate.
Shadeform Issue 008 public evidence
Marketplace aggregator displaying third-party provider prices.
Appendix A

Headline panel status.

Sixteen fresh priced observations enter the medians. Eleven fresh captured exclusions do not. No row is pending.

Tier 1, Hyperscalers, 4 of 4 priced
IDProviderStatusMedianRateReason
E001AWSPricedYes$6.88on_demand_fixed
E002AzurePricedYes$12.29on_demand_fixed
E003GCPPricedYes$11.0613on_demand_fixed
E004OraclePricedYes$10.00on_demand_fixed
Tier 2, GPU clouds and neoclouds, 10 of 18 priced
IDProviderStatusMedianRateReason
E005CoreWeavePricedYes$6.1550on_demand_fixed
E006CrusoePricedYes$3.90on_demand_fixed
E007LambdaPricedYes$3.99on_demand_fixed
E008NebiusPricedYes$3.85on_demand_fixed
E009Together AIPricedYes$3.99on_demand_fixed
E010HyperstackPricedYes$3.20on_demand_fixed
E011VerdaPricedYes$3.25on_demand_fixed
E012Lightning AI (formerly Voltage Park)PricedYes$4.50on_demand_fixed
E013FluidstackCaptured excludedNono_public_comparable_fixed_rate
E014NscaleCaptured excludedNono_public_comparable_fixed_rate
E015IRENCaptured excludedNono_public_comparable_fixed_rate
E016ChronoScaleCaptured excludedNono_public_comparable_fixed_rate
E017GMI CloudCaptured excludedNono_public_comparable_fixed_rate
E018CUDO ComputeCaptured excludedNono_public_comparable_fixed_rate
E019SpheronCaptured excludedNomarketplace_dynamic
E020JarvislabsPricedYes$2.69on_demand_fixed
E021HyperbolicCaptured excludedNono_public_comparable_fixed_rate
E022Atlas CloudPricedYes$2.95on_demand_fixed
Tier 3, Marketplaces and aggregators, 2 of 5 priced
IDProviderStatusMedianRateReason
E023RunPodPricedYes$3.29on_demand_fixed
E024Vast.aiCaptured excludedNomarketplace_dynamic
E025PaperspacePricedYes$5.95on_demand_fixed
E026TensorDockCaptured excludedNohost_marketplace_floor
E027ShadeformCaptured excludedNoaggregator_passthrough