ComputeCapital Markets

Three instruments on one physical anchor: the energy cost of computation, measured by proof of work.

The instruments

Three instruments, three surfaces

Together they form a complete capital market for energy-denominated computation. Futures produce the spot curve, bonds the term structure, options the volatility surface. Every other instrument prices off one of these three.

What each solves

Futures answer what is compute worth right now? Bonds answer what is the future cost of compute today? Options answer how uncertain is that cost?

Dimension
Futures
Bonds
Options
Asset class
Derivative — perpetual index future
Fixed-income — zero-coupon bond
Derivative — convertible exchange option
Time horizon
Spot price, no maturity
Fixed term: 3, 6, or 12 months
Fixed expiry: 3, 6, or 12 months
Underlying
Qi energy index per compute unit
Locked coinbase emission at maturity
QUAI/Qi conversion ratio, struck in Qi
Yield mechanism
Funding rate, basis trades
Discount to face, lock multiplier
Premium, paid for by reduced accretion
Primary market
Compute hedgers, agents, speculators
Miners, real-yield investors, treasuries
Miners writing convexity, operators buying it
TradFi analog
E-mini S&P 500, Brent crude future
Treasury bill, brokered CD
Convertible bond, FX exchange option
What it produces
Forward price surface — the spot curve
Discount curve — the term structure
Implied volatility — the price of uncertainty
The structural problem

It measures what vendors charge
not what compute costs.

A rental rate aggregated from commercial platforms reflects vendor margins, market positioning, capacity utilization, and competitive dynamics. It answers what is the current ask — not what it costs to produce the computation.

The distinction collapses the moment markets thin out, vendors coordinate pricing, or a new hardware generation reshapes the cost curve overnight. Survey indexes are backward-looking by construction: they report what was charged, not what the physics demands.

What a futures market needs

Non-manipulable, anchored to physical reality, forward-looking by structure.

Rental-rate surveys provide none of these. An index built on delivered work per unit of energy provides all three, because the underlying quantity is metered rather than reported.

The common anchor

All three share one anchor.

The energy cost of computation, measured by proof of work. This is not an arbitrary design choice — it is the only honest unit of account for compute pricing.

Why not price the machine

A benchmark denominated in silicon expires on the next tape-out

  • An H100 is not an H200, and training and inference depreciate on different curves
  • A hardware-tracking benchmark must be rewritten every generation
  • That rewrite lands exactly when collateral values and financing paths need it to hold still
  • Fungibility fails: an hour of a named chip is not comparable across clusters or terms
Why price the work

Work does not decay

  • Beneath every token there is an irreducible thermodynamic cost to erasing a bit
  • The ratio of useful output to energy spent is measurable in any generation, on any architecture
  • It holds for architectures not yet designed — the invariant a durable benchmark needs
  • One H100 in Virginia is approximately one H100 in Frankfurt, once you price the watts