Not An Offer  ·  Informational Only  ·  All Figures Hypothetical And Illustrative
The Firm

Compute Is A Traded Commodity.  We Trade The Spread, Not The Direction.

A specialist relative-value desk in compute and power.  We do not take a view on whether GPU rental rates rise or fall.  We trade the spreads between them, the way an energy merchant trades gas against power.
Silicon Data H100 and B200 Rental Index Futures list on NYMEX by CME Group from 05 October 2026, subject to regulatory review: the first regulated market in the input every AI system consumes.
--Days
--Hrs
--Min
--Sec
To First TradeNYMEX compute futures, 05 October 2026, subject to review
The Compute Forward Curve
Maturity runs left to right, chip generation runs into the screen, price runs up.  The surface is the shape the Firm expects each regime to produce.  Drag it.
R3  ·  Power-Constrained
Price ↑
Maturity →
Generation ↘
Drag to rotate
Illustrative surface shapes, not quoted prices, not a forecast and not a trading recommendation.  The instrument has no price history, so nothing here is derived from traded data.
The Thesis

Three Drivers Behind The Trade

Compute is financed, traded and hedged like energy infrastructure.  Each driver is observable in public disclosure.
DRIVER 1

Compute Is A Commodity

Listed futures from 05 October 2026, with over-the-counter forwards and residual value swaps already trading.  A new asset class is forming on an exchange timetable.

Proved in The Market
DRIVER 2

Capital Needs A Price

NVIDIA and six asset managers are financing compute like energy infrastructure.  Lenders, lessors and guarantors carry the exposure and must mark, hedge and transfer it.

Proved in the NVIDIA case
DRIVER 3

Builders Earn A Spread

A developer that controls time-to-power and hedges the energy-to-compute spread adds roughly ten points of equity return, and every project can carry that overlay.

Proved in the developer case
Therefore

An independent desk with physical-layer fluency, no chip inventory and no index affiliation.  It grows with every financing and every project as the listed market deepens.

The Arbitrage

The Spread Between Power And Compute

One megawatt-hour of facility energy becomes a compute margin.  That distance is the compute spark spread, and it is set by utilization, rental rate and useful life.
01

Time To Power

Large-load interconnection runs three to four years in ERCOT and approaches seven in PJM.  Time-to-energization is the binding constraint, and it anchors the forward curve.

02

The Margin

Compute revenue per megawatt-hour of facility energy, less power, operations and full capital recovery.  Power is under three percent of it.  GPU amortization is about thirty-five.

03

The Spread

Utilization, rental rate and useful life set it, which is why it is marked and hedged like a commodity margin rather than booked as a real estate yield.

A year of earlier energization is worth far more than a cheaper power contract.  Speed is the arbitrage, and the developer who hedges the spread earns a margin rather than a yield.
The Book

Five Sleeves, One Risk Budget

Every position is a spread.  8% to 10% annualized volatility, net index delta inside plus or minus 0.25x.
SleeveWhat It TradesWhy The Spread Exists
S1 · Curve Relative ValueH100 and B200, first to third maturitiesThe curve reprices faster than the physical delivery schedule standing behind it
S2 · Cross-Generation RatioB200 against H100 as a ratio spreadThroughput parity anchors the ratio to silicon rather than to sentiment
S3 · Average-Settlement DecayVariance decay inside arithmetic average settlementAveraging kills variance on a cube law, and the option surface prices it late
S4 · Index Against PhysicalListed index against public operator economicsThe desk transacts on both sides of that basis and therefore sees both prices
S5 · Gas And Power CalendarsNYMEX natural gas and Micro NGLiquid ballast, and the other half of the spark spread
No sleeve requires GPU prices to rise.  No sleeve requires them to fall.
The Business

What The Firm Does

01

Trading Book

A relative-value book in H100 and B200 compute futures, with natural gas and power calendars as liquid ballast.

02

Research And Advisory

Regime research, index diligence, and mapping a client’s realized rental rate to the settlement index before it can hedge with it.

03

Hedging Structuring

Operator hedging, lender risk transfer and offtake collars for financing platforms, special purpose vehicles and neoclouds.

04

Development Overlay

Energy-to-compute arbitrage for new data center projects: time-to-power, a hedged spread, cascading offtakes and a two-vehicle capital stack.

Shared Platform

One regime model, one metric stack, three derived indicators, systematic execution, independent risk, compliance and an information barrier.

Model Portfolio

Two Models You Can Run

Every figure is illustrative.
Model 1

The Compute Spark Spread

What one MWh of facility energy earns after power, operations and full capital recovery.

$799per MWh of facility energy, after capital recovery
Revenue$1,768
GPU amort
Facility
Power, ops
Spread
Illustrative.  788 GPUs per MW, $11,000 per kW of facility over twelve years
Model 2

Implied GPU Life

The index level is a joint statement about useful life and operator margin.  Solve for the floor.

$1.92index floor
Hardware / hr
Against a $2.64 spot index.  Illustrative
Why Now

NYMEX Lists The First Regulated Compute Futures On October 5

05 OCTFirst regulated compute futures list on NYMEX, subject to regulatory review
3Exchanges with announced compute derivatives products
6Exchange-traded fund issuers with registered compute products
$500BN+Of compute financing platforms announced by NVIDIA and six asset managers
$105BNResidual value guaranty NVIDIA disclosed on 4.25 GW of leased AI campus capacity
Where The Inefficiencies Sit
SourcesBuilt from public disclosures, exchange and index provider announcements, regulatory filings and licensed market data as of September 2026.  Company, exchange and index names appear for illustration only and no affiliation with or endorsement by any of them is implied.
Positioning

What The Firm Is, And What That Implies

Compute Capital Partners is a specialist commodity manager: a research house with a trading book attached and a structuring business alongside it.  The closest analogue is the trading arm of a physical commodity merchant, as Vitol and Glencore are in oil and metals.
Regulatory Identity
Commodity trading advisor and, once pooled, commodity pool operator.  CFTC registration and NFA membership, or a documented exemption under Rule 4.13(a)(3) or Rule 4.7, is a condition precedent to accepting third-party capital.  The strategy holds no securities, so the Firm is not an investment adviser.
Vehicle
Separately managed accounts and structuring mandates first.  Capacity is the binding constraint, so managed accounts let each client fund its own margin and take capacity pro rata under a published formula, with no commingling.  A pooled vehicle follows when daily volume supports its cost base.
Fee Model
Budget-based and capped rather than a flat percentage of assets.  The strategy carries a published capacity ceiling and the manager is paid to respect it.
Net Exposure
Structurally neutral.  Every position is a spread, and the book targets a net index delta within plus or minus 0.25x capital.  Any directional tilt is a residual of relative value, measured and reported daily.
Alignment
Sponsor capital funds the trading book through the track record period, with a general partner commitment of at least five percent of committed capital.

An Independent Desk Between Capital And Price

Capital providers need a price to lend against, and the price layer needs hedgers to become liquid.  If you are financing, building, operating or buying AI compute, we should talk.

Start A ConversationRisk And Governance