Compute Capital Markets

Spot Index Derivative

Futures

A perpetual, cash-settled futures contract priced against the Qi energy index. Hedges and discovers the price of compute today.

FormPerpetual future UnderlyingQi energy index SettlementQUAI
Fixed-Income Instrument

Bonds

A zero-coupon bond collateralized by locked coinbase emissions, tokenized for liquid secondary trading. Finances and discounts the price of compute tomorrow.

FormZero-coupon bond CollateralLocked coinbase SettlementQUAI

What each solves

The two instruments answer different questions. Futures answer "what is compute worth right now?" Bonds answer "what is the future cost of compute today?"

Futures

Spot price discovery for inference

A cash-settled, perpetual futures contract on a standard unit of inference compute. The underlying is the Qi energy index — what one unit of compute costs in raw energy terms. No hardware delivers; positions settle financially against the index.

The problem
There is no observable, non-manipulable price for inference compute. Existing benchmarks are survey indexes — they aggregate what vendors charge, not what compute costs. They reflect margin and capacity politics, not physics.
What it solves
A futures market produces a forward-looking, manipulation-resistant price for the cost of computation. AI agents, model operators, and infrastructure providers can hedge their compute exposure the same way airlines hedge fuel.
Who needs it
Inference-heavy operators who carry compute as a variable cost. Hyperscalers managing capacity risk. Treasuries hedging future training runs. Speculators expressing views on the energy cost of intelligence.
Read the Futures spec →
Bonds

Fixed-income against locked emissions

A zero-coupon bond. Locked coinbase rewards form the collateral. The bond is issued at a discount to face value and tokenized — holders can hold to maturity for full face value, or sell into a secondary market at any time. Yield is fixed at issuance by the protocol's lock multiplier.

The problem
GPU mining networks lack long-term capital commitment. Miners can switch chains in hours, leaving networks vulnerable to rental-hashrate attacks. ASIC networks solve this with hardware lock-in; GPU networks need a financial mechanism instead.
What it solves
Locked rewards create ASIC-like commitment without specialized hardware. Tokenized principal preserves capital flexibility — miners can sell the bond for immediate liquidity without breaking the lock. The protocol gets security; the miner gets cash flow.
Who needs it
Miners optimizing between yield and liquidity. Capital allocators seeking real-yield exposure to compute infrastructure without operating hardware. DeFi protocols building structured products on top of a defined yield curve.
Read the Bonds spec →

Two instruments, two markets

Together they form a complete capital market for energy-denominated computation. Futures address the derivatives market; Bonds address the fixed-income market. Every other instrument prices off one of these two surfaces.

Dimension
Futures
Bonds
Asset class
Derivative — perpetual index future
Fixed-income — zero-coupon bond
Time horizon
Spot price, no maturity
Fixed term: 3, 6, or 12 months
Underlying
Qi energy index per compute unit
Locked coinbase emission, payable at maturity
Yield mechanism
Funding rate, basis trades
Discount to face value, lock multiplier
Primary market addressed
Compute hedgers, agents, speculators
Miners, real-yield investors, DeFi treasuries
TradFi analog
E-mini S&P 500 future, Brent crude future
Treasury bill, brokered CD, Pendle PT
What it produces
Forward price surface — the spot curve
Discount curve — the term structure

Why start with energy

Both instruments share a single 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.

Computation is, fundamentally, the orderly conversion of energy into change of state. Every floating-point operation, every hash, every inference token is paid for in joules. The cost floor of any computation is the electricity it consumes — everything above that is margin, capacity premium, or speculation.

Proof of Work is the only consensus mechanism that bonds tokens to physical reality. A miner cannot produce a valid block without expending verifiable electrical work. The block reward is, by construction, a receipt for energy spent. This makes the coinbase emission the cleanest possible reference price for one unit of compute — denominated in nothing but joules.

Stake-based systems cannot do this. Their unit of account is the token itself, which is circular: the price of compute is whatever holders agree it is. Energy is exogenous. Energy is non-arbitrary. Energy is the same physical quantity in Texas, Iceland, and Sichuan.

01

Non-arbitrary

Joules are the same everywhere. No vendor can manipulate the underlying; no token holder can vote it higher. The physics is the price.

02

Forward-looking

Energy markets already trade decades into the future. Anchoring compute to energy inherits a mature term structure, not a survey of recent rentals.

03

Verifiable

Proof of Work makes the energy expenditure cryptographically auditable. The block itself is the receipt — no oracle, no trust, no off-chain attestation.

Research dialogue · A compute capital markets competitor × Wolfram Blockchain Labs

The missing floor price for compute

Compute markets price from the top down — vendor ask rates, rental surveys, offtake negotiations. What happens when you price from the bottom up, starting at the thermodynamic cost of a hash?

The normalization stack

Each layer commoditizes the mess below it. Fuel is geographically non-fungible. Electricity is a first normalization. Standard GPU hardware is a second. The algorithm is a third. A random hash function is a fourth. And the final layer normalizes every compatible hashing algorithm into a single energy-denominated money output.

Layer 0 — Non-fungible

Fuel source

Natural gas in Texas, solar in Arizona, hydro in Quebec. Geographically non-fungible, compositionally dynamic, and impossible to standardize directly.

Layer 1 — First normalization

Electricity

A joule is a joule — but grid standards differ between ISOs, and behind-the-meter power isn't traded at all. PPAs further decouple electricity cost from compute price. This layer hides more than it reveals.

Layer 2 — Hardware normalization

GPU compute

One H100 in Virginia ≈ one H100 in Frankfurt. Standard hardware converts variable electricity into comparable computation. Performance deltas exist but are financially manageable — like light sweet crude in two ports.

Layer 3 — Workload normalization

Inference

Same model, same weights, different location. If you don't care about latency, inference is approximately fungible. A small delta, not a different product.

Layer 4 — Maximum fungibility

SHA-256 hash

Output is identically random regardless of hardware, location, or fuel source. No latency sensitivity. No model dependency. The most standardized unit of computation that exists — and it already has a deep, liquid market.

Layer 5 — Algorithm normalization

Merge-mined proof of work

SHA-256, Scrypt, ProgPoW, and any other compatible hashing algorithm all produce different kinds of irreducible computational work. SOAP merge-mines across these algorithms simultaneously, unifying heterogeneous proof-of-work hardware into a single security and economic layer.

Layer 6 — Monetary normalization

$Qi — Energy-denominated money

A universal denominator token payout normalizes them into a single energy-denominated money output. The token has adaptable economics to which algorithm produced the hash. It measures the energy expended, point blank. This is the final normalization: from many proof-of-work algorithms to one thermodynamic unit of account.

What competitors see

Competitors are building capital markets stacks for compute — regulated indices, pricing infrastructure, and derivative instruments. Here is what the leading teams have learned about compute pricing.

01 — Embedded electricity

Electricity is an input, not a readable output

The cost of electricity is embedded in GPU-hour and token prices. But solving the inverse problem — tracing a compute rate back to an electricity cost — breaks down because PPAs fix the underlying power cost while compute rates vary freely above it.

02 — The invisible market

Behind-the-meter power has no price discovery

Grid power trades on exchanges like Nodal. But as data centers move off-grid — natural gas turbines, on-site generation — the electricity they consume has no market. Compute pricing could become the only proxy for this invisible energy market.

03 — Regime dependence

The electricity signal depends on supply–demand balance

When compute is demand-constrained, price collapses toward a spread over electricity cost — high correlation, high predictiveness. When supply-constrained (now), price is bid up by what compute is worth to buyers, drowning out the energy signal.

04 — Unsophisticated pricing

GPU pricing is sales-driven, not quant-driven

Most compute providers don't dynamically reprice. Rates update weekly or biweekly. Sellers respond to high demand by extending contract durations rather than raising prices — collapsing the spot market and destroying price discovery.

Two regimes, two signals

Whether compute price tells you anything about underlying energy cost depends entirely on who has market power.

Demand-constrained

Price → electricity + spread

When GPUs sit idle, providers compete on price. Rates collapse toward marginal energy cost plus a thin margin. In this regime, compute price is highly correlated with electricity price.

↳ Compute price is informative about energy cost
Supply-constrained (current)

Price → willingness to pay

When demand exceeds capacity, buyers bid up the price of compute based on what inference is worth to them, not what it costs to produce. The electricity signal is drowned out by margin and urgency.

↳ Compute price reveals demand intensity, not energy cost

The PoW floor price

When inference demand drops, GPUs don't have to sit idle. Proof-of-work mining creates an always-available fallback — a thermodynamic floor beneath compute infrastructure.

High Low $/GPU-HR TIME → PoW FLOOR ELECTRICITY TRADEOFF POINT INFERENCE PREMIUM MINING ZONE
Inference value (market-driven)
PoW mining floor (thermodynamic)
Electricity cost (base layer)
The key insight

Mining is a hedge on compute infrastructure buildout

If the primary value creation is inference, and that creates a higher dollar value, but sometimes you cannot get full utilization — then mining allows you to recuperate something. GPU-friendly proof of work turns idle capacity into revenue instead of loss.

This only works if the PoW network is GPU-native. Bitcoin's SHA-256 runs on ASICs that can't do inference. A GPU-friendly proof-of-work system — like Quai Network — makes the tradeoff actually executable on the same hardware, which is what makes the floor real rather than theoretical.

The meta

There is a deeper structure hiding inside the tradeoff point.

Marginal opportunity cost

Each side precisely defines the unit economics of the other

The marginal economic tradeoff between inference and GPUs performing proof of work isn't just a hedge — it's a mutual pricing oracle. The opportunity cost of running inference is the PoW revenue you forgo. The opportunity cost of mining is the inference revenue you forgo.

This means inference and proof of work don't just coexist — they price each other. At the margin, one precisely defines the unit economics of the other. The tradeoff point is not a static floor. It is a continuously updated, market-driven signal that tells you the real cost of compute from two independent directions simultaneously.

No survey index can produce this. It emerges only when both workloads compete for the same hardware, in real time, with real capital at stake.

The tradeoff goes both ways

The marginal conversion point between PoW mining and inference isn't just a floor — it's a pricing signal that works in both directions.

Mining → Inference

Infrastructure already exists

Major miners like Marathon have already built facilities, secured electricity, and deployed cooling and power infrastructure for Bitcoin mining. They are rapidly transitioning toward inference because the economics now say "go to inference."

The physical plant — power, cooling, networking — was amortized by mining revenue. Inference inherits infrastructure that has already been paid for.

Inference → Mining

Downside protection for builders

If you are building a data center for projected inference demand, GPU-native mining provides a hedge in case that demand doesn't materialize as expected.

Instead of stranded capacity and write-downs, you have a thermodynamic fallback that generates revenue proportional to the energy consumed. The floor price is set by physics, not by market sentiment.

Why the floor matters for pricing

Competitors are building capital markets infrastructure for compute. The missing piece in every compute index is a non-manipulable, energy-anchored lower bound.

The problem with survey indexes

You can't build futures on a market where the floor is unknown

Current compute indexes aggregate what vendors charge. In a supply-constrained regime, that tells you willingness to pay, not cost of production. In a demand-constrained regime, prices collapse toward electricity plus margin — but nobody knows exactly where that floor is.

A GPU-native proof-of-work network provides the missing observable: the marginal revenue per GPU-hour from mining. This is a real, continuous, market-set price signal that establishes the floor beneath which no rational GPU operator would sell compute. It is the thermodynamic bid.

Without PoW floor

Futures settle against a survey index. Floor is estimated, not observed. Manipulation resistance depends on data coverage breadth.

With PoW floor

Floor is a live, on-chain, thermodynamically anchored price. No survey needed. Falsifying requires real energy expenditure.

What this enables

Credible lower bound for options pricing. Put spreads with real meaning. Infrastructure insurance products with an observable strike.

Try the Controller

See how the convertible controller keeps Qi indexed pricing and QUAI settlement equivalent — the engine behind Compute Futures.

Launch Demo