SGPI Signwl GPU Price Index
The cost of cloud GPU compute, published as two daily series: an observation-weighted average level in USD per hour, and a chain-linked base-100 index for trend reading on a stable scale.
Methodology
View full methodology v1.0Two series. The SGPI is published as a level (observation-weighted average price in USD per hour, useful for the question “what does GPU compute cost today?”) and an index (chain-linked, base 2026-02-01 = 100, useful for the question “how has the price of GPU compute moved?”). The two can diverge when the mix of GPUs in the market changes — the index is designed to neutralise mix shifts so it reads as a pure price signal.
Scope. GPUs only. TPUs, FPGAs, custom silicon (Trainium, Inferentia) and other non-GPU accelerators are excluded.
Basket. Each (GPU model, pricing tier) pair is treated as a single basket item, with a per-day price obtained by averaging the underlying observations weighted by their observation count. The four pricing tiers — spot, on-demand, 1-year reserved, 3-year reserved — are blended at the item level rather than reported separately.
Weights. Items are weighted by the number of underlying price observations. Weights are a coverage and liquidity proxy — they are not consumption quantities or expenditure shares. The SGPI therefore measures the price movement of the observable cloud GPU pricing universe, weighted by how thoroughly each item is sampled. It does not claim to be a consumption-weighted index in the strict economic sense.
Index formula. The SGPI Index is chain-linked. Each day’s chain ratio is a weighted geometric mean of price relatives Pt / Pt-1 across items present on both days. Items entering (e.g. a newly-launched GPU) or exiting (e.g. a retired accelerator) drop out of the day’s ratio cleanly without breaking the series. Geometric (rather than arithmetic) means are used for symmetry — a 50% rise followed by a 50% fall returns the index to par — and outlier robustness. Fisher and Törnqvist superlative indices would require consumption-quantity weights, which we do not have, so are not used.
Update frequency. Daily. The pipeline runs each morning at 06:30 UTC from the previous day’s consolidated pricing data. Historical values may be revised when underlying source data is restated; revisions are noted in the change log.
Frequently Asked Questions
What is the SGPI?
The Signwl GPU Price Index (SGPI) is Signwl’s daily reference series for the cost of cloud GPU compute. It is published in two forms: a level (observation-weighted average price in USD per hour) and an index (chain-linked, base 2026-02-01 = 100).
What’s the difference between the Level and the Index?
The Level answers “what does GPU compute cost today?” in USD per hour. It moves both when prices change and when the mix of GPUs in the market changes — if cheaper inference GPUs become more heavily sampled, the Level can fall even though no specific GPU got cheaper. The Index answers “how have prices moved?” on a stable scale. It is chain-linked, so each day’s movement is computed only from GPUs present on both days — mix shifts drop out, and only genuine price changes affect the index. The Level and Index can diverge meaningfully when the basket of observed GPUs shifts.
How is the SGPI different from individual GPU pricing?
Individual GPU prices (like H100 at $6.43/hr) show the cost of one specific accelerator. The SGPI aggregates all tracked GPUs into a single market-wide indicator, weighted by observation coverage. This makes it useful for tracking the overall direction of GPU compute costs rather than any single GPU movement.
How often is the SGPI updated?
The SGPI is recalculated every morning at 06:30 UTC from the previous day’s pricing data. The full historical series begins 2026-01-08 (the first day with sufficient source coverage). Historical values are updated in place when underlying source data is revised; a change log on this page records any methodology revisions.
Explore the GPU Compute Market
Live pricing for 39 accelerators, GPU comparisons, regional analysis, and investment tools.