Search the market

Jump to a section, a GPU or a provider

Compute economy

Compute is turning into a commodity.

Scarce supply, volatile prices, big regional differences and a unit that can be standardized: compute now has every ingredient commodity markets grow from. The chart is today's evidence - the same chip, priced several times over depending on who sells it.

B200 4.0× spreadH200 6.0× spreadH100 8.0× spreadA100 3.8× spreadL40S 6.0× spread

Not all tokens are equal

You pay one price - but behind the API, requests are quietly sorted into tiers. Flat price, variable product.

PromptSummarize the termination clause in this contract.

What you think you buy

Full frontier model

Either party can terminate with 30 days' written notice. If the customer leaves early, a fee of two months' committed spend applies, and data must be returned within 14 days.

Quality100%

What you often get

Same price - nobody told you

You can end the contract if you give notice. There may be a fee.

Quality34%

The next utility

Every century adds infrastructure nobody can opt out of. Once compute is essential, how it's priced and rationed becomes a question of policy - and fair access starts with transparent prices.

  1. 1900sClean waterPlumbed into every home
  2. 1920sElectricityAlways on, metered by the hour
  3. 2000sThe internetDebated as a basic right
  4. NowAI computeIntelligence on tap - priced like a utility

Compute today looks like electricity in the early 1990s

Right before power became one of the largest traded commodities in the world.

Electricity1993Before the power exchanges
=Same stage
Compute2026Before the compute exchanges
  1. How it is bought
    Electricity, 1993

    Bespoke, bilateral power contracts

    Compute, 2026

    Bespoke compute contracts between clouds, AI labs and enterprises

  2. How prices are set
    Electricity, 1993

    Prices set administratively, not cleared by a market

    Compute, 2026

    Spot prices set by provider algorithms, not cleared independently

  3. Transparency
    Electricity, 1993

    No public reference price - every tariff is its own

    Compute, 2026

    The same chip is listed at several times the price depending on the seller

  4. Price risk
    Electricity, 1993

    Large users absorb price swings with few tools to hedge them

    Compute, 2026

    Buyers carry GPU price exposure with no standard way to hedge it

  5. What comes next
    Electricity, 1993

    The first competitive power auctions arrive within a few years

    Compute, 2026

    The first compute price indices and futures venues are launching

What followed for power: exchanges and a public day-ahead price, then forwards, futures and options. The rest of this page is what that path looks like for compute.

The market is being assembled

Supply and demand are already at scale. What's missing in Europe is the layer in between: an exchange, an honest benchmark and the intermediaries that make prices tradeable.

Booming
Supply
  • Neoclouds
  • Hyperscalers
  • Decentralized networks
Missing in Europethe layer in between
  • ExchangeNone in Europe
    Transparent marketplaces exist mainly in the US
  • BenchmarkFirst curves
    GPU price indices and forward curves are appearing
  • IntermediariesGap
    Market makers, banks and funds
At scale
Demand
  • AI companies
  • Corporates
  • Financial institutions

In place, at scale Missing or only forming in Europe

From FinOps to a compute treasury

Most organizations sit on the first two steps - reporting spend after the fact. The top of the ladder treats compute like any other currency the business holds. From “we really pay a lot” to “we pay just enough”.

  1. Level 1Cloud & modern stacksAdopt cloud to scale fast.CTO
  2. Most organizations are here
    Level 2FinOpsCost reports used after the fact to monitor and trim spend.CTO
  3. Level 3Value-based allocationCompute enters budget cycles with controls before spend.CTO + CFO
  4. Level 4Compute risk managementPrice exposure is monitored and hedged.CTO + CFO + CRO
  5. Level 5Compute treasuryA central compute pool, laddered contracts and internal transfer pricing.Board-level

Benchmarks first, then hedging

Once there's a public price, the familiar treasury tools follow - each one built on the step before it.

Hover or tap a step to see what it builds on.

A compute swap

Fixes the cost of compute you already buy at the floating price.

Fixed leg - you pay
€2.00 / GPU-hour
Floating leg - you receive
Dynamic price

When the grid is expensive the swap pays you; when it is cheap you pay it. Either way the all-in cost stays at the fixed rate.

A regional spread option

Protects against one region getting expensive - priced on the gap between grids, net of data movement, latency and portability limits.

The swap on today’s curve

Illustrative: one GPU-hour notional at a €2.00 fixed leg against the Dynamic floating price.

An honest public price is step one.

Soventra publishes what every major GPU cloud charges, hour by hour, in one currency - with the energy behind the price and the jurisdiction behind the data.

Open the market