What exactly is in the GPU pricing dataset?

Two tiers. Tier 3 is raw instance-level GPU rental records with hardware specs, location, datacenter conditions, power efficiency, and workload types. Tier 2 aggregates those records to country level, broken out by contract type: spot, interruptible, 3-month and 6-month reserved, second-hand, and brand-new.

How often is the data updated?

Every business day, in step with the publication of the Silicon Data portal and indices. Historical records are retained, so the dataset supports backtesting as well as live monitoring.

Which GPUs and providers are covered?

Coverage spans 30+ rental platforms — roughly 95% of neocloud providers and all major hyperscalers — across the models tracked by the SiliconIndex family, including H100, H200, A100, B200, and MI300X.

Is this the same data behind the SiliconIndex benchmarks?

Yes. The published indices are computed from these records. Tier 3 is the raw input, Tier 2 is the country aggregation, and the daily index values themselves are available separately as the Index & Curve Datasets.

How is the data delivered?

As CSV downloads for research and spreadsheet workflows, or over API for systematic and production use. Both delivery paths carry the same records and update on the same daily schedule.

Can I redistribute the data or use it in a product?

Internal use is covered by an Enterprise subscription. Redistributing the data, publishing it to your own clients, or referencing it in a financial product is licensed separately — talk to sales and the terms will be scoped alongside the data.

GPU Pricing Dataset

GPU pricing data, from raw records to country benchmarks

Silicon Data's GPU pricing data covers instance-level and country-level rental rates, collected every day across 30+ platforms — from hyperscalers to neoclouds. The same records that power the SiliconIndex family, delivered as downloadable files or API for research, trading, and procurement teams.

Two tiers of granularity

GPU rental pricing at two levels of detail

Every tier is derived from the one below it, so the numbers reconcile all the way up to the published indices — what you see in a benchmark can be traced back to the raw records that produced it.

T-3

Instance-level pricing

The rawest tier: instance-level GPU rental records with full context — hardware specs, location, datacenter conditions, power efficiency, and workload types. The same records that feed everything above them, before any aggregation.

T-2

Country-level pricing

Aggregated from T-3: country-level GPU pricing broken out by contract type — spot, interruptible, 3-month and 6-month reserved, second-hand, and brand-new.

T-1 — the published indices and curves themselves — is available separately as Index & Curve Datasets.

How it works

How the GPU pricing data is built

Collected daily

Rental rates are gathered every business day across 30+ platforms — covering roughly 95% of neocloud providers and all major hyperscalers — with the hardware, location, and contract context attached to every record.

Standardized like-for-like

Records are normalized so an H100 hour in Virginia and an H100 hour in Singapore are comparable: same units, consistent contract-type labels, and datacenter conditions captured alongside the price.

Delivered your way

The same standardized records ship as CSV downloads and API feeds, and roll up into the country-level tier and the published SiliconIndex benchmarks — one pipeline, three levels of access.

Who it's for

Built for teams that need the numbers, not the charts

Quant & research teams

Backtest strategies on full price history, build custom indices and factors on raw records, and join pricing to your own supply and demand data.

Procurement & FinOps

Benchmark vendor quotes at the instance level, track reserved-vs-spot spreads by region, and negotiate with market-wide comparables.

Lenders & investors

Underwrite GPU-backed exposure with market data, monitor second-hand vs brand-new spreads, and feed valuation models with daily updates.

Delivery

GPU pricing data, the way your stack wants it

30+Platforms tracked globally

DailyRefresh cadence

CSV + APIDelivery options

Daily refresh

Updated every business day, aligned with portal publication.

CSV download

Flat files for research, spreadsheets, and backtesting workflows.

API access

Enterprise-grade delivery for systematic and production use.

GPU pricing data — common questions

Two tiers. Tier 3 is raw instance-level GPU rental records with hardware specs, location, datacenter conditions, power efficiency, and workload types. Tier 2 aggregates those records to country level, broken out by contract type: spot, interruptible, 3-month and 6-month reserved, second-hand, and brand-new.
Every business day, in step with the publication of the Silicon Data portal and indices. Historical records are retained, so the dataset supports backtesting as well as live monitoring.
Coverage spans 30+ rental platforms — roughly 95% of neocloud providers and all major hyperscalers — across the models tracked by the SiliconIndex family, including H100, H200, A100, B200, and MI300X.
Yes. The published indices are computed from these records. Tier 3 is the raw input, Tier 2 is the country aggregation, and the daily index values themselves are available separately as the Index & Curve Datasets.
As CSV downloads for research and spreadsheet workflows, or over API for systematic and production use. Both delivery paths carry the same records and update on the same daily schedule.
Internal use is covered by an Enterprise subscription. Redistributing the data, publishing it to your own clients, or referencing it in a financial product is licensed separately — talk to sales and the terms will be scoped alongside the data.

Get the data behind the benchmarks.

Dataset access is scoped with our team — tell us what you are building and we will put the right tier in front of you.