Rillor

Home / Markets / Rillor Compute Index

Rillor Compute Index

What AI compute costs, to rent and to own. The index tracks the prices and availability neoclouds and hyperscalers publish through their pricing APIs, and the prices of GPU systems themselves.

Request index dataContribute pricing data

Defined by Methodology v1.0, published 9 Oct 2026.

Four series, one method

Rental and availability describe renting compute. Hardware describes owning it, and rent-to-value connects the two.

SeriesMeasuresBroken out byUnit
RCI RentalPublished rental prices for GPU capacityGPU class, region, term and pricing type (on-demand, spot or preemptible, reserved). Hyperscaler and neocloud segments are shown separately.USD per GPU-hour
RCI AvailabilityShare of tracked offers that providers report as available to rentGPU class and segmentPercent
RCI HardwareGPU system prices by comparable gradeSystem grade. Air-cooled and liquid-cooled builds are separate grades.USD per system
RCI Rent-to-ValueAnnualized rent over hardware price for the same GPU class, with payback and residual-value contextGPU classRatio
Hyperscaler and neocloud segments
Hyperscalers are large general-purpose cloud providers. Neoclouds are cloud providers built around GPU capacity for AI. Each source in the register carries one segment, and the two segments are shown separately in every series that uses them.
Pricing types
On-demand is an hourly price with no commitment. Spot or preemptible is capacity the provider can reclaim at short notice. Reserved is a committed term, recorded with its length in months.
How rent-to-value is built
A year of rent for one GPU at the published rate, divided by one GPU's share of a comparable system's price. It relates what a GPU earns on rent to what it costs to own.

What the index covers

Eleven GPU classes, from the A100 to rack-scale NVL72 systems, across cloud rental and hardware.

GPU classes

GPU classMakerArchitecture
H100NVIDIAHopper
H200NVIDIAHopper
B200NVIDIABlackwell
B300NVIDIABlackwell Ultra
GB200 NVL72NVIDIAGrace Blackwell, 72-GPU rack
GB300 NVL72NVIDIAGrace Blackwell Ultra, 72-GPU rack
MI300XAMDCDNA 3
MI325XAMDCDNA 3
MI355XAMDCDNA 4
L40SNVIDIAAda Lovelace
A100NVIDIAAmpere

Scope

Segments
Hyperscaler and neocloud, shown separately
Pricing types
On-demand, spot or preemptible, and reserved terms
Dimensions
GPU class and variant, region, term, pricing type
Hardware
GPU systems by comparable grade, with air and liquid cooling graded separately
Availability
As each provider reports it at capture time
Capture
The methodology captures prices and availability from provider pricing APIs and published pricing pages
Index coverageChanges to classes and scope are recorded in the change log

Where the data comes from

Each series names its sources. Each observation records its source and its capture time.

Rental and availability

RCI Rental and RCI Availability use providers' official pricing APIs and published pricing pages, within each source's terms. Availability is what a provider reports as available to rent, recorded as reported.

Hardware

RCI Hardware uses published listings, quotes and reported sales of GPU systems. Each observation is tagged with its evidence class, and the classes stay separate in calculation.

Contributed data

Operators, resellers, buyers and lenders can contribute pricing data under a data agreement. Contributions enter the source register with their evidence class.

Contribute pricing data

Evidence classes

Evidence classWhat it recordsUsed in
SpecificationWhat a system is: GPUs, memory, networking and cooling. It carries no price.Grade definitions
ListingAn asking price a seller publishes, including the rental prices providers publish through their pricing APIs and pagesRCI Rental, RCI Availability, RCI Hardware
QuoteA price offered for a stated configuration, quantity and delivery dateRCI Hardware
Reported saleA completed sale, as reported by a party to itRCI Hardware
Verified transactionA completed sale checked against its documentsRCI Hardware

How a price becomes a series

Every observation passes through the same five steps. Rillor's own quotes and agreements stay outside the index.

Index methodSources lead to capture, then normalization, then evidence class, then the four published series. Below a conflict firewall, Rillor's own quotes and agreements are blocked from entering the index and are excluded from every published series.SourcesPricing APIs, pages,listings, quotes, salesCaptureEach price, with itssource and capture timeNormalizePer GPU-hour, by class,region, term, typeEvidence classEvery observationtagged, classes apartSeriesRental, Availability,Hardware, Rent-to-ValueConflict firewallRillor's own quotes and agreementsGPU system quotes, sales, forward agreementsExcluded from every published seriesMethodology v1.0, part 2: administration and conflicts
Index methodThe same steps apply to every series
  1. SourcesOfficial pricing APIs and published pricing pages for rental and availability. Listings, quotes and reported sales for hardware. Contributed data under a data agreement.
  2. CaptureEach observation records one price or availability report, its source and its capture time.
  3. NormalizePrices are put on a per GPU-hour basis and mapped to GPU class and variant, interconnect, node size, region, term and pricing type.
  4. Evidence classEach observation carries its evidence class, so a listing never counts as a sale.
  5. SeriesA median for each provider, then the median across providers, produces each published series.

Methodology v1.0

The rules behind every series. Each version is dated and stays on record.

Rillor Compute Index

Methodology, version 1.0

This document defines the Rillor Compute Index. It sets out the four series, the sources the index may use, how observations are captured, normalized and classed, how each series is calculated, and how the index is governed.

Request the methodology (PDF)
Version
1.0
Publication date
9 Oct 2026
Administrator
Rillor Corporation
Series
RCI Rental, RCI Availability, RCI Hardware, RCI Rent-to-Value
Format
PDF

Contents

  1. DefinitionsThe terms the index uses, from GPU class and pricing type to segment, offer and evidence class.
  2. Administration and conflictsRillor administers the index, and its own quotes and agreements are excluded from published series.
  3. Eligible sources and the source registerWhich sources qualify, and the register that records each one with its segment and terms of use.
  4. CollectionHow prices and availability are captured from pricing APIs and published pages, and what each observation records.
  5. NormalizationHow observations reach a common basis: per GPU-hour, GPU variant, interconnect, node size, region, term and pricing type.
  6. Evidence classesHow specifications, listings, quotes, reported sales and verified transactions are defined and kept apart.
  7. CalculationThe median for each provider, then the median across providers, and how rent-to-value is built from rental and hardware prices.
  8. Data rules for publicationThe coverage a series needs before a value is published.
  9. Expert judgmentWhen judgment may be applied, who applies it and how each use is recorded.
  10. Publication schedule and correctionsWhen each series is published, and how errors are corrected and disclosed.
  11. Methodology changesHow changes are proposed, reviewed, announced and dated.
  12. Complaints and recordsHow to raise a complaint about a value or the method, and which records the index keeps.
  13. LimitationsWhere each series is reliable and where it needs care, such as thin coverage for a GPU class or region.
  14. Version historyEach version of this document, its date and what changed.

Governance

Rillor administers the index under its published methodology. Conflicts, changes, corrections and complaints follow written rules.

Administration

Rillor Corporation administers the index. It maintains the source register and the methodology, and calculates each series under the methodology.

Conflicts

Rillor also supplies GPU systems and arranges forward agreements. Its own quotes and agreements are excluded from every published series.

Changes

Changes to the methodology, the source register and the GPU class list are dated and recorded with the reason for each one.

Corrections

When a published value is wrong, we publish the corrected value, mark it as a correction and keep the original in the record.

Complaints

Send a complaint about a published value, a source or the method to contact@rillor.com. Each complaint is logged and answered in writing.

Records

Captured observations, calculations and judgments are kept, so any published value can be traced back to its inputs.

Change log

VersionDateChange
1.09 Oct 2026First publication of the methodology

Who uses the index

Teams that rent, buy or finance GPU compute, and the researchers who study what it costs.

Capacity planningCompare rental prices by GPU class, region, term and pricing type, and see how often providers report capacity as available to rent.GPU and neocloud pricingProcurementWeigh renting against owning, and check a GPU system quote against prices for comparable grades.GPU systemsFinancing diligence and collateral researchBorrowers and lenders review hardware prices and rent-to-value for the systems behind a lease or loan.GPU financingForecastingForecasting research uses the series as inputs, and as recorded outcomes to score forecasts against.Forecasting and market intelligence

Notices

Rillor Compute Index series are for research and planning use. They are not offered for settlement of any contract.

Rillor is not a registered investment adviser or commodity trading advisor and does not provide investment or trading advice. Research, forecasts, data and software described on this site are for research and engineering use. Nothing here is an offer or recommendation to buy or sell any security, commodity interest or digital asset. Past or simulated results do not indicate future results.

NVIDIA, AMD and related product names are trademarks of their owners and are used only to identify products. No affiliation or endorsement is implied.

Request the series you need.

Request index dataContribute pricing data