Sorso View
The Independent Benchmark for AI Compute Economics
SVCI Issue 006, July 21, 2026

The Sorso View Compute Index

By the Sorso View Editorial. Capture date Monday, July 20, 2026. Published July 21, 2026.
Publication resumed after a one-cycle interruption following Issue 005. Under Amendment A1 to Methodology v1.1, every panel observation in this issue rests on fresh screenshot evidence captured July 20, 2026, and provider movement is presented as change since Issue 004 rather than as a week-over-week measurement.
SVCI Composite103.5minus 0.3 points since Issue 004. Base 100.0 at Issue 001.
Tier 1 Median10.53065USD/GPU-hr, N of 4
Tier 2 Median3.85USD/GPU-hr, N of 9
Tier 3 Median4.47USD/GPU-hr, N of 2, thin coverage
Price Floor2.97median of bottom quartile
Compute Opacity28.5coverage 71.5 percent

The reading

The SVCI stands at 103.5 against the Issue 001 base of 100.0, a change of minus 0.3 points since Issue 004. The Companion Dollar Basket, the absolute-dollar secondary metric, is 6.92 dollars per GPU-hour. The Hyperscaler Premium is 2.74x and the Neocloud Discount is 63.4 percent. Composite coverage is 71.5 percent across the locked eligible panel of 27 providers, of which 15 priced and 12 were captured and excluded with evidence.

24681012USD PER GPU-HOURTier 1AWS: 6.88 USD per GPU-hourOracle Cloud Infrastructure: 10.0 USD per GPU-hourGoogle Cloud Platform: 11.0613 USD per GPU-hourAzure: 12.29 USD per GPU-hourTier 1 median: 10.53065 USD per GPU-hour10.53065Tier 2Jarvislabs: 2.69 USD per GPU-hourAtlas Cloud: 2.95 USD per GPU-hourHyperstack: 3.2 USD per GPU-hourVerda: 3.25 USD per GPU-hourNebius: 3.85 USD per GPU-hourCrusoe Cloud: 3.9 USD per GPU-hourLambda: 3.99 USD per GPU-hourTogether AI: 3.99 USD per GPU-hourCoreWeave: 6.155 USD per GPU-hourTier 2 median: 3.85 USD per GPU-hour3.85Tier 3RunPod: 2.99 USD per GPU-hourPaperspace: 5.95 USD per GPU-hourTier 3 median: 4.47 USD per GPU-hour4.47
Fifteen priced observations by tier, July 20, 2026 capture. Dots are individual providers; the gold diamond marks each tier median. Hover any mark for the provider and rate. The full table is on the Data page.

What moved since Issue 004

Two provider rates moved on identical bases. Every other priced observation matched its Issue 004 sealed value on an independent fresh read.

Hyperstack
Public H100 SXM on-demand, per GPU-hour
+33.3%
2.40 to 3.20, up 0.80. The largest movement in the series to date.
RunPod
Secure Cloud on-demand H100 SXM, per GPU-hour
-9.1%
3.29 to 2.99, down 0.30. Provider page stamped updated July 17, 2026.

Reading the index against Issue 005

Issue 005 printed 100.0 on a composition event: Hyperbolic entered the priced set at 1.39 and pulled the Tier 2 median to its base value. In Issue 006, Hyperbolic no longer publishes a public rate surface and returns to captured-excluded status, and Hyperstack repriced upward. The Tier 2 median accordingly returns to 3.85. The 3.5-point rise from Issue 005 is therefore a coverage-and-composition move combined with one genuine repricing, not broad market repricing.

Coverage and opacity

Twelve of 27 providers publish no comparable public fixed rate: quote-gated sales processes, from-floor figures, live host-variable marketplaces, and aggregator passthrough surfaces. The Compute Opacity Index stands at 28.5. Monitoring figures observed on excluded surfaces are permanently ineligible for every published metric.