# Sorso View Compute Index, Issue 009

**Publication date:** August 25, 2026  
**Capture date:** August 24, 2026  
**Status:** FINAL - SEALED FOR SITE UPLOAD  
**By the Sorso View Editorial**

## Sixteen prices, zero moves

The Sorso View Compute Index reads **104.1** for Issue 009, unchanged from Issue 008. Every one of the sixteen priced providers scheduled for the August 24 capture published the same controlling rate recorded in the prior issue. Sixteen fresh observations produced sixteen unchanged figures.

This was a priced-only capture under the two-track cadence. The eleven other panel members were deferred. They were not reviewed, carried, or newly excluded in Issue 009. Coverage Score and the Compute Opacity Index are therefore intentionally omitted.

## Index results

| Measure | Issue 009 |
|---|---:|
| SVCI | 104.1 |
| Tier 1 median | $10.53065 |
| Tier 2 median | $3.875 |
| Tier 3 median | $4.62 |
| Pooled non-hyperscaler median | $3.875 |
| Hyperscaler Premium | 2.7176x |
| Neocloud Discount | 0.6320 |
| Price Floor | $3.075 |

## The observed panel

| ID | Provider | Tier | Rate per GPU-hour | Movement |
|---|---|---:|---:|---:|
| E001 | AWS | 1 | $6.88 | $0.00 |
| E002 | Azure | 1 | $12.29 | $0.00 |
| E003 | Google Cloud | 1 | $11.0613 | $0.00 |
| E004 | Oracle | 1 | $10.00 | $0.00 |
| E005 | CoreWeave | 2 | $6.155 | $0.00 |
| E006 | Crusoe | 2 | $3.90 | $0.00 |
| E007 | Lambda | 2 | $3.99 | $0.00 |
| E008 | Nebius | 2 | $3.85 | $0.00 |
| E009 | Together AI | 2 | $3.99 | $0.00 |
| E010 | Hyperstack | 2 | $3.20 | $0.00 |
| E011 | Verda | 2 | $3.25 | $0.00 |
| E012 | Lightning AI, formerly Voltage Park | 2 | $4.50 | $0.00 |
| E020 | Jarvislabs | 2 | $2.69 | $0.00 |
| E022 | Atlas Cloud | 2 | $2.95 | $0.00 |
| E023 | RunPod | 3 | $3.29 | $0.00 |
| E025 | Paperspace | 3 | $5.95 | $0.00 |

## One legacy surface stopped publishing

Voltage Park's own pricing surface now routes all three displayed tiers to contact-for-pricing. The panel entry remains priced at $4.50 through the already approved successor and aggregator exception, attributed to Lightning AI, formerly Voltage Park.

## Different products can carry different H100 prices

Crusoe publishes $3.90 per GPU-hour for an H100 GPU instance and $5.50 per hour for an H100 self-serve managed deployment. The hardware label matches, but the product mode does not. Issue 009 applies the same rule previously used for Together AI: the raw GPU instance rate is included and the managed-inference rate is excluded.

The same product-mode distinction appears elsewhere in the observed set. Together AI publishes $3.99 for GPU clusters and $5.49 for dedicated inference. Verda publishes $3.25 for an instance and $3.58 for serverless. RunPod publishes $3.29 for a Pod and $4.79 for serverless. Public and self-serve does not automatically mean comparable.

## Advertised floors versus controlling rates

Three captured providers displayed starting or from language relevant to this review. Verda advertised $1.63 while its controlling on-demand H100 rate was $3.25. Atlas Cloud's page title advertised H100 from $2.49 while its table published $2.95. Jarvislabs advertised from $2.69, equal to its controlling $2.69 rate. Two floors were below the controlling rate and one matched it.

## Evidence and method

The controlling observations use public, fixed, hourly H100 SXM 80GB-class rates. Instance prices are divided by GPU count using decimal arithmetic. Spot, preemptible, commitment, reservation, quote-gated, starting-at, managed-inference, serverless, and other non-comparable product modes are excluded.

Issue 009 contains 92 registered frames covering all 16 scheduled providers. One registered duplicate was retained, producing 91 unique frame hashes. Thirteen signed-in frames were retained as flagged and superseded evidence: eleven for Google Cloud and two for Atlas Cloud. Logged-out recaptures control and reproduced the relevant figures. Five originals contain browser-bookmark privacy exposure and remain restricted to the private archive; cropped derivatives are used for any external evidence distribution.

## Scope disclosure

Deferred and not reviewed in Issue 009: E013 Fluidstack, E014 Nscale, E015 IREN, E016 Applied Digital, E017 GMI Cloud, E018 Cudo Compute, E019 Spheron, E021 Hyperbolic, E024 Vast.ai, E026 TensorDock, and E027 Shadeform.

Issue 009 makes no fresh disposition for those eleven entries and does not infer panel-wide coverage from the priced-only block.

## Seal record

The index was independently recomputed from the 16 listed rates. The internal result is `104.11993197289829799834847498758146279534651931875`, which displays as **104.1** under the locked rounding rule. The evidence archive and release package were rebuilt and verified on August 25, 2026.