| Instrument | Venue | Role In The Book | Tenor | Status |
|---|---|---|---|---|
| Silicon Data H100 Rental Index Futures | NYMEX | Core: curve relative value and the ratio leg | Front 3 months, extending to 6 | Listing |
| Silicon Data B200 Rental Index Futures | NYMEX | Core: the cross-generation ratio against H100 | Front 3 months | Listing |
| Henry Hub natural gas and Micro NG | NYMEX | Ballast: carries the process and execution discipline | Front 12 months | Live |
| Power futures at relevant hubs | NYMEX, ICE | Regime indicator, later the siting basis leg | Front 24 months | Live |
| Compute futures on a second venue | ICE, others | Cross-venue basis, sized at zero until both legs clear | n/a | No date |
| Options on compute futures | NYMEX | Convexity for the reflexive tail | n/a | Future listing |
Single-name equities and equity options, including chipmakers, hyperscalers and neoclouds. Binary event contracts. Crypto. Naked directional positions on the index level. Physical GPU or capacity purchases. The equity exclusion is a compliance design choice as much as a strategy one.
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. Each compute contract is 730 GPU-hours, cash settled to the monthly average of the daily index.
Which input is scarcest. Silicon, capital, power or demand: whichever binds sets the price. The metric stack is monitored continuously and the regime is voted formally each month.
What the market asserts. Reconstruct the term structure, the cross-generation ratio and the implied butterfly, and compare against the pre-listing implied curve.
What the economics support. Solve the cost stack and derive the implied life, the compute crack and revenue per megawatt, then compare to disclosed accounting lives and resale prices.
Where the two disagree. A position requires a stated gap between what the curve asserts and what the physical economics support, plus a written mechanism for why it closes.
What would prove the call wrong. The reversing observation is registered in writing before the position is taken, and size is the lesser of the risk budget and the capacity formula.
Rest orders, then re-underwrite. Passive execution inside a 12 percent participation cap, with positions re-underwritten monthly and closed on an invalidation trigger regardless of the mark.
A regime call changes a position, and a position without a written invalidation test is never sized. Both rules are what an allocator diligences ahead of a track record.
| Sleeve | Expression | Holding Period | Economic Anchor |
|---|---|---|---|
| Curve relative value | H100 and B200, first to third maturities | 5 to 25 days | A new curve with no term-structure history |
| Cross-generation ratio | B200 against H100 | 15 to 40 days | Throughput parity, a fundamental anchor |
| Average-settlement decay | Front month convexity | 1 to 21 days | Cube-law variance decay, mechanical |
| Index against physical | Index versus public operator economics | 10 to 45 days | Public disclosure only, compliance-gated |
| Gas and power calendars | NYMEX natural gas and Micro NG | 3 to 30 days | Liquid today, and carries the process |
The listed curve spans the front few maturities at launch and the strategy is mean reversion rather than carry, so a short holding period keeps capital turning inside the capacity ceiling.
The strongest sleeve trades the ratio between the two listed chip generations, normalized for delivered compute. It has a fundamental anchor: the ratio reverts toward the price of a unit of delivered throughput, where most spreads revert only to a statistical mean. Both legs list on one venue on the same day with one settlement mechanic, which removes cross-venue and methodology basis, and margin offset between two correlated legs on one exchange improves capital efficiency. Decoupling is documented: January and March of 2026 each produced a double-digit move in one index with the other flat.
Throughput parity between chip generations is an engineering fact that can be measured against independent benchmarks, so the sleeve has a level to revert to rather than a mean to hope for.
The throughput ratio varies with precision, batch size, memory footprint and framework, and inference carries a wider gap than training. Supply timing dominates in the short run, so a large delivery tranche can hold the ratio away from parity for a quarter. The older generation has a longer amortization runway and a deeper installed base, which can make its floor sticky.
Energy desks have priced average-settled contracts for decades, while participants arriving from equity and crypto tend to price front-month risk as if it were terminal. The sleeve is sized for the first settlement cycles and scaled down as energy flow arrives.
The construction detail, the sleeve weights and the model behind each anchor are available to qualified counterparties under a confidentiality agreement.
Start A ConversationRisk And Governance