The Commodity Analogy
Compute futures borrow patterns from established commodity markets — oil, electricity, semiconductor wafers. Each analogy illuminates something and breaks at specific points. Understanding both is essential to evaluating where compute futures will and won't work.
Oil analogy
Oil futures (NYMEX WTI, ICE Brent) are the most-developed commodity futures market. The structure compute futures imitates includes:
- Multi-month forward contracts.
- Standardized contract size and quality specification.
- Reference price index.
- Mix of financial and physical settlement.
- Speculative and hedging participants.
Oil's match to compute is partial. Oil has a single global reference grade (mostly); compute has many distinct GPU types. Oil storage moderates price spikes; compute can't be stored.
Electricity analogy
Electricity futures are more compute-like in several ways:
- Electricity can't be stored at scale; like compute, it must be consumed in real time.
- Pricing varies by location (different regional markets).
- Different "products" (peak vs off-peak) trade at different prices.
- Capacity (forward access to generation) trades separately from energy (kWh consumed).
The capacity-vs-energy distinction is particularly relevant. Compute futures may eventually distinguish between "reserved capacity" and "actual usage" similarly.
Semiconductor wafer analogy
Semiconductor wafers have informal forward markets but no major listed futures. The analogy:
- Like wafers, GPUs come in distinct generations with rolling obsolescence.
- Like wafers, GPU production capacity is concentrated in a few facilities.
- Like wafers, supply is mostly contract-priced rather than spot.
The semiconductor analogy highlights the obsolescence risk in compute — an H100 in 2028 has very different value than an H100 in 2024 because the alternatives have changed.
What the analogy gets right
The commodity-futures framework works for compute because:
- Compute is fungible enough within a GPU type to support standardized contracts.
- Forward pricing is genuinely useful for both hedgers and speculators.
- Reference indices can be constructed from observable transaction prices.
- Settlement mechanics (financial primarily) are well-developed.
Where it breaks
The analogy breaks in important ways:
- GPU generations. H100 in 2027 isn't the same as H100 in 2025 — the relative value vs alternatives shifts. Oil from 2027 is just oil.
- No storage. Compute is consume-or-lose. Storage smooths most commodity markets.
- Concentrated supply. NVIDIA controls supply more tightly than any single oil producer controls oil.
- Demand structure. AI compute demand can change architecture (different attention, custom silicon) in ways oil demand doesn't.
- Market depth. Compute futures markets in 2026 are thin; oil futures have decades of liquidity.
Takeaway
The commodity analogy is the right starting point but should be held lightly. Compute is genuinely novel in several ways. The next chapters look at the specific exchange offerings.