Gross 1.5x to 2.5x, with net index delta within plus or minus 0.25x. Risk per trade capped at 40 basis points of capital. No position above the published capacity formula.
A maximum of 12 percent of daily volume on any trade. Resting orders only in compute for the first ninety days. The book liquidates within three sessions at target size.
No single spread above 25 percent of the risk budget and no sleeve above 35 percent. It is a two-contract market, so correlation is assumed high rather than measured optimistically.
A written invalidation test is registered before sizing. A invalidation trigger closes the position regardless of the mark. Every position is re-underwritten monthly against the regime vote.
| Drawdown, Peak To Trough | Action |
|---|---|
| -5% | Risk report to the investment committee, and every open position re-underwritten against the current regime |
| -8% | Forced review, gross exposure cut by half, and new risk only with written committee approval |
| -15% | Positions flattened and the strategy halted pending a full process review with the risk function |
The risk function is independent of the portfolio manager and kill authority sits with risk. Third-party administration and an annual audit are in place from inception.
A hedge is margined and the underlying asset is not. When price spikes, the hedge pays out immediately in cash while the offsetting gain arrives slowly as revenue, and for a neocloud the collateral is a depreciating GPU while the margin call is due tomorrow. Every client program is therefore sized on its liquidity facility first, and where a facility cannot be sized, margin-free structures are used instead: collars with thresholds, and physical offtake.
In securities, an insider trading on material non-public information about its own company acts unlawfully. The instinct is to assume a chip vendor or hyperscaler trading compute futures on its own allocation decisions commits the same offense.
CFTC Rule 180.1 was modeled on Rule 10b-5, but the Commission stated it would not reach trading on lawfully obtained non-public information absent a breach of a pre-existing duty or a misappropriation. Trading on proprietary information about one’s own business is lawful: it is the economic definition of hedging. A generator that knows its plant is going offline can lawfully buy power forward.
A paired trade, long the compute future and short a vendor’s stock into an unannounced allocation freeze, is a Rule 10b-5 case on the equity leg. The most important wall in this business runs between the futures desk and anything touching public securities, which is why single-name equities and equity options sit outside the mandate entirely.
| Guardrail | Legal Basis | What It Requires |
|---|---|---|
| The equity leg | Rule 10b-5 and Regulation FD | Allocation cuts and unpublished capital expenditure timing are material non-public information about listed issuers. Restricted lists around earnings, launches and export notices, and no instruments settling on a monitored issuer’s disclosures. |
| Index contributor status | Benchmark integrity | A panel contributor holding the contract is the benchmark-rigging pattern. Contributors are data providers under a published methodology, never oversight members. |
| Customer confidential information | Commodity Exchange Act misappropriation | Trading on a customer’s contract terms or capacity plans is misappropriation squarely inside Rule 180.1. This is the genuine exposure in a structuring business. |
| Bona fide hedge eligibility | Position limits | Position limit relief requires documented offset of commercial risk and separate hedge and speculative books. |
Walls will be built to stop insiders trading other people’s information and to keep the index clean. Insiders trading their own physical information is what a hedging market is for, so the asymmetry survives the compliance regime by design.
Pushing the index down is expensive, because the actor subsidizes its entire rentable fleet. Pushing it up is nearly free if the index reads posted list rates, and expensive only if it reads transacted prices. Posted versus transacted is the whole manipulation surface, and it is the precise question the CFTC has raised. On current evidence early manipulation is uneconomic against launch open interest, and the residual risk is incidental index distortion from ordinary commercial behavior.
| Filter | What It Does |
|---|---|
| Transacted versus posted | Tags every input as an executed rental, an OTC print or a posted list rate. Posted rates never set a signal alone |
| Listing persistence | Drops listings that stay unfilled across a full month, and repeated identical quotes with no evidence of capacity |
| Outlier bands | Excludes quotes beyond three median absolute deviations of their chip and region cohort |
| Provider concentration | Caps any single provider at 15 percent of the Firm’s reference series |
| Wash and round-trip detection | Flags OTC prints between related counterparties, and offsetting prints at matching size |
| Cross-venue reconciliation | Compares the settlement index, a second published index, OTC prints and the prediction-market implied curve every day |
| Settlement-window surveillance | Watches quote changes in the final days before each monthly settlement for unexplained moves |
The filtered inputs build the Firm’s own reference series for each contract. A signal fires only when the official index and the reference series agree within a set tolerance. When they diverge, new risk in that tenor stops and position size halves until the gap is explained. The same series supports independent marks for clients and a documented record for regulators and counterparties.
| Risk | Statement | Mitigant |
|---|---|---|
| Principal not yet appointed | The Partner and Chief Investment Officer is not yet under contract. | A shortlist and a conditional offer precede any trading; no interim trader is appointed. |
| No track record | Every figure published by the Firm is model output over stated assumptions. | A nine to twelve month administered record is the deliverable of the Prove stage. |
| Listing risk | The 05 October date is subject to regulatory review, and delay or specification change is possible. | The gas book, research and structuring continue through a delay; compute sleeves wait. |
| Index governance | The contract settles to a commercial index from a firm backed by a proprietary trading house. | A blocking diligence item. If it fails, the trading thesis fails with it, and the Firm says so. |
| Conflicts | Principals hold employment and board roles in power and AI infrastructure. | No trading authority for conflicted principals, and counsel clearance precedes any trade. |
| Capacity | Compute capacity absorbs only part of an account inside three years. | Undeployed capital stays with the investor, and energy calendars carry early capacity. |
| Ecosystem concentration | The index references one vendor’s chips, and that vendor’s exposures crystallize together in a demand shortfall. | A direction-neutral book, optionality in the demand-constrained regime, and marks that earn in both paths. |
| Overlay concentration | Most overlay revenue comes from a few project mandates, each dependent on financial close. | Mandates are sourced across developers and lenders, and operating capital is not dependent on overlay revenue. |
| Launch-window edge | Decay mispricing is richest in the first settlement cycles, before a principal is seated. | The listing months are recorded as data rather than traded by an interim appointment. |
The conflicts architecture, the index diligence file and the risk framework are available in full to qualified counterparties under a confidentiality agreement.
Start A Conversation