Cross-Commodity Swaps
In the AI era, a new economic signal has emerged: the Watt-Bit Spread – the margin between cheap electrons and the valuable intelligence they produce. To capture and hedge this spread, analysts have proposed the Watt–Bit Swap, a cross-commodity derivative. This concept parallels how metals producers use cross-commodity swaps to hedge the spark spread (the difference between electricity value and fuel cost). Our analysis is informed by Graeme Harrison’s Watt-Bit Swap concept and a previous Ventures Edge analysis of the Watt-Bit Spread.
Payoff = γN(PB-aPW)
a = Power Co. Participation
N = MWh provided from Power Co. to Compute Co.
PB = GPU Index Price
γ = Notional conversion rate
PW = Price of Watts
The Spark Spread and Watt-Bit Spread
Energy markets long used the spark spread to measure power-plant profitability. In essence, a spark spread represents selling electricity minus the cost of fuel (e.g. gas) to produce it. For example, if natural gas costs $4/MMBtu and a power plant needs 7 MMBtu per MWh, fuel costs $28/MWh; if power sells at $60/MWh, the spark spread is $32/MWh. Traders hedge this margin in commodity markets.
By analogy, the Watt-Bit Spread measures how much value AI computation (bits) gets per unit of electricity (watts). It highlights a key insight: AI workloads can earn many times more revenue per kWh of power than traditional uses. As our previous article notes, the Watt-Bit Spread “captures a fundamental disconnect between the cost of a watt and the value of the bits created by those watts”. In other words, turning electricity into AI services yields huge profit potential, and the Watt-Bit Swap is designed to let companies tap into that potential without committing to a fixed price.
Cross-Commodity Spread Trading
Spread trading is about exploiting differences rather than absolute prices. Instead of betting whether power prices will go up or down, traders focus on relative moves. For example, a cross-commodity swap between aluminum and copper might exchange cash flows to lock in the margin between the two prices. The underlying principle is simple: stabilize a producer’s operating margin by tying it to the market price of its output and input.
Key points about such spreads:
They require liquid price references (e.g. futures) for both commodities.
They lower volatility exposure, since the spread tends to move less than each price alone.
They reveal fundamentals: a wide spread signals tight supply or strong demand on one side.
In recent years, the explosion of data centers means electricity (once a minor cost) is now often the main bottleneck. By structuring a Watt-Bit Swap, tech companies and utilities can apply the same relative-value logic to electricity vs. compute.
Watt-Bit Swap Mechanics
The Watt-Bit Swap is effectively a contract linking electricity payments to AI compute pricing. Under this swap:
A data center (or Compute Co..) takes power from the grid as usual.
Instead of paying a fixed tariff for each MWh, they enter a swap with a power provider.
The swap pays out based on AI compute value (e.g., a market index of GPU rental rates) versus a reference power price (e.g., a power purchase agreement).
As Graeme explains, this “takes the fixed price of energy out of the equation” and ties the power plant’s revenue to compute value. In effect, when GPU prices rise, the power plant gets more money per MWh; when compute value falls, the power plant’s payoff falls too. This aligns the incentives of both sides. The data center can ramp up GPUs without ballooning electricity bills, and the utility earns a share of the AI profits rather than just a flat fee.
Another way to see it: the power plant’s revenue profile “looks exactly like the GPU cluster’s” once linked by the swap. They rise and fall together, which allows tech financiers (who understand GPUs) to help finance power infrastructure. It essentially creates a single intelligence infrastructure asset, rather than forcing old power PPAs onto fast-moving AI business models.
Notional and Participation
Two key swap terms are notional amount and participation rate:
Notional conversion rate (a): Converting GPUh to MWh requires multiplying the duration of compute by the physical power draw of the processors to bridge the gap between operational time and total energy consumed. This calculation captures the work performed as electricity is dissipated through the silicon, transforming abstract data processing into a measurable volume of thermodynamic heat and utility-scale power. (Please note in the model we used an assumed number)
Participation rate (γ) is the fraction of that energy volume covered by the swap. A γ of 100% means full exposure to the AI value spread; a lower γ means only part of the load is linked. This lets parties share risk: e.g. the utility might only take 50% participation to limit its exposure.
If a deal covers 100 MW and γ = 50%, then effectively 50 MW is on the floating swap, and 50 MW remains on a normal tariff. By adjusting γ, the Power Co. and Compute Co. can negotiate how much upside/downside each bears.
Numeric Example
Suppose a data center needs 100,000 MWh/year (about 11.4 MW continuous). If an average GPU cluster yields say $3 per GPU-hour in computing value, and each GPU-hour uses about 2 kWh of electricity, then every MWh can create roughly 500 times its energy cost in AI revenue. Numerically, $3/GPUh * (1,000/2 kWh) = about $1,500 value per MWh vs. only say $60 cost per MWh – an enormous spread.
If the Power Co. locked in a swap at 20% participation, the power provider would earn 20% of that multi-hundred-dollar margin on each MWh delivered. If participation was 50%, the power provider would get half of it. This illustrates how dramatically different compute economics can be.
The takeaway is: the swap payoff is (participation fraction) times (contract volume) times (the difference between compute value and Power Co.st per unit of energy). A higher γ or larger volume, or simply a wider gap between compute and power prices, all increase the payout to the power side.
How It Ties Watts to Bits
This swap makes electricity financing sensitive to AI demand and competition. If new AI chips or services raise the market value of computing, the power plant reaps more revenues immediately. Conversely, if GPU prices fall or efficiency jumps (meaning fewer kWh per computation), the swap reduces what the data center pays.
By decoupling from flat power rates, tech companies can treat power as a flexible input: they pay more when compute is most lucrative, and less when it isn’t. Utilities, meanwhile, gain a hedge against overcapacity – they’re not just taking on pure electricity risk, but sharing in growth by indexing to compute.
Metals vs AI: A Comparative View
The Watt-Bit Swap mirrors a familiar structure but with different underlyings. We can outline key contrasts:
Output/Value: Traditional swaps involve a hard commodity (like copper or aluminum) versus an energy price. The Watt-Bit swap uses intangible compute value (e.g. GPU rental rates) instead.
Correlation: In metals-power swaps, output price and power price often correlate through industrial cycles. In Watt-Bit, compute demand drives power demand – the two are even more tightly linked.
Notional basis: Metals notional is physical (tonnes), while Watt-Bit notional is in physics or equivalent computing capacity (MWh to GPUh) .
Volatility source: Metals spreads move with commodity booms/busts; Watt-Bit spreads move with token cycles, cloud demand and chip innovation.
Optionality: AI-driven spreads have more optionality (e.g. early mover advantage), whereas metals-plant capacity is more fixed once built.
Time-sensitivity: In AI, timing is critical (every year of extra capacity is huge). Metals markets evolve slower by comparison.
In short, the Watt-Bit Swap operates on the same principle as a metal-power swap (hedging a production margin), but the context is much more dynamic. Both simply exchange MWh payments for some index price, but in Watt-Bit that index reflects GPU market rates (equivalent to commodity price).
Cash Flows
The Power Company delivers electrons to the data center.
Under the swap, each month they don’t just invoice a flat power tariff. Instead, they invoice based on how much GPU processing happened.
The data center sells its AI services in the market and pays the swap settlement.
Put simply, money flows when compute value flows. If GPUs earn $X, then the Power Co. gets a related payout; if GPU economics falter, the Power Co.’s revenue falls in step.
A useful way to think is: the power plant’s income profile mirrors the data center’s revenue profile. This alignment means investors see both as one system. As Harrison observes, it lets chip financiers and energy financiers “speak the same language” – financing the whole AI-power system together.
Bringing it All Together
The Watt-Bit Swap’s payoff is extremely sensitive to technology shifts and market moves:
Compute Price Slippage: If competition or efficiency drives down the value of a GPU-hour, the swap’s margin shrinks, lowering what the Power Co. earns.
Efficiency Gains: If new GPUs use half as much power per calculation, the spread per MWh doubles – good for the data center’s cost, but it changes expected payoffs.
Buildout Uncertainty: Many cloud projects may queue for power, but not all be built. If the swap was signed on optimistic demand, overcapacity could leave a party with stranded commitments.
Correlation Breakdown: The model assumes GPU value tracks power availability. If AI demand plateaus or shifts, the assumed link might weaken, creating basis risk.
In practice, these swaps may include protective clauses: floors or collars on compute prices, caps on total volume, or termination provisions if a project halts. Harrison and others (Ornn, Silicon Data, etc.) mention measures such as availability adjustments and curtailment payments to manage such scenarios.
The Watt-Bit Swap could align interests and unlock financing in several ways:
For Power Companies: It converts uncertain merchant electricity risk into a more stable, index-linked revenue stream. By effectively selling a stake in GPU time, power plants can tap growth in the tech sector.
For AI/Cloud Providers: It eases huge capital hurdles. They can build more data centers knowing they only pay for power out of the revenue those centers actually earn, rather than fronting the cost themselves.
For Investors: It creates a new asset class, one that bridges utilities and high-tech. Funds that normally shy away from power projects might join deals with a Watt-Bit element, betting on AI growth.
Graeme aptly notes that this swap could “open up broader capital markets access for the power infrastructure” by linking it to silicon economics. It moves energy assets toward being “part of the larger AI system” rather than stand-alone utilities. In effect, Watt-Bit Swaps aim to make electricity financing as responsive as the algorithms that depend on it.
Ready to Dive Deeper Into the Mechanics?
Click here to download our open-source Cross-Commodity Swap Model and explore how these variables interact in a live trading environment.
Send michaelbrownyyc@gmail.com an email if you have any questions :-)