The launch problem was simple: public H100 prices that looked comparable often were not. Lambda itself published different per-GPU rates for different H100 SXM configurations. Its eight-GPU bundle was published at $2.99 per GPU-hour, while a single-GPU SXM price observed in April 2026 was $4.29 per hour. The difference was not evidence that one number was false. It was evidence that configuration and pricing basis had to travel with the number.
Two public Lambda-related references available around launch illustrated the problem:
The two figures differ by $1.30 per GPU-hour, about 43.5% relative to the $2.99 bundle rate, because they describe different configurations. That is a comparability problem, not a license to call either figure a universal market price.
That distinction became the operating premise of Sorso View. A public number is useful only when the hardware, unit, term, product mode, source class, and availability model are clear enough to compare it with another public number on the same basis.
What we are building.
Sorso View is a publication for the people who move and analyze the AI compute economy. The flagship product is the Sorso View Compute Index, a public benchmark across a twenty-seven-provider universe in three tiers. Under A1-R1, priced providers are observed fresh on weekly or biweekly issue cycles between monthly full-panel sweeps. Each issue publishes the benchmark readings, coverage state, evidence links, and the editorial interpretation of what changed.
The goal is not to be another lead-generation comparison table or an opaque institutional feed. The goal is to make the public reference layer of AI compute economics reproducible and readable: measurement and editorial in the same publication, with the source trail visible.
A benchmark is useful only when the reader can see what the number means, what it excludes, and how to reproduce it.
The four commitments.
A benchmark is a promise of method, not a promise that the market will move on a schedule. Sorso View publishes under four standards that apply across the Index and editorial work.
- i.Sourcing
Every official data point is traceable to a recorded public source. Estimates are labeled. If the evidence cannot support an official number, the number does not enter the benchmark.
- ii.Consistency
The same inclusion gate, anchor definition, normalization rules, and disclosed methodology govern each issue. Capture cadence follows the current methodology, and methodology changes are disclosed when they take effect.
- iii.Distinction
Measured data, derived data, exclusions, and editorial analysis are labeled separately. Inference economics, alternative silicon, and other research domains may be studied separately, but they are not silently blended into the SVCI.
- iv.Correction
When Sorso View is wrong, the error is corrected on the public record. The Corrections and Updates page preserves what was wrong, what is right, and how the error occurred.
What we will not claim to measure.
The SVCI does not observe confidential private contract pricing. It does not treat committed-use, reserved, spot, starting-floor, or account-gated figures as interchangeable with a fixed logged-out public on-demand rate. Providers that do not resolve to a comparable public rate remain visible as exclusions rather than being estimated into a median.
Revenue Intensity is a separate analytical proxy derived from public financial disclosures when the inputs permit it. It is not a realized customer rate and it is never substituted for an observed public price.
The public on-demand rate is therefore not claimed to represent most transaction volume. It is the most reproducible public reference layer available under the locked methodology, and that is the layer the Index measures.
Why now.
AI compute is a capital-intensive market with fast hardware turnover, constrained capacity, opaque enterprise contracting, and public rate cards that can describe different products or terms. That combination makes a historical public benchmark useful even in periods when the headline index does not move.
The record compounds by keeping the measurement basis stable and the evidence visible. The ambition is straightforward: become a reference that procurement teams, operators, investors, and reporters can cite because they can check it.
What happens next.
The Sorso View Compute Index publishes on recurring issue cycles under the current two-track cadence. Alongside it, Sorso View publishes periodic Essays, is developing the Reports program, and will activate the event-driven Sorso View Brief only through a separate public notice.
If you have made it this far, you are the audience. Subscribe below. The record continues issue by issue.
By the Sorso View Editorial
April 27, 2026 · Source-integrity update August 11, 2026