Liquid Compute’s Response to the CFTC Request for Comment on Compute Derivatives

Liquid Compute
45 Main Street, Ste 422, Brooklyn, NY 11201
liquidcompute.com

October 20, 2026

Christopher Kirkpatrick
Secretary of the Commission
Commodity Futures Trading Commission
Three Lafayette Centre
1155 21st Street, N.W.
Washington, D.C. 20581

Re: Request for Comment on the Listing of Compute Derivatives Contracts (RIN 3038-AF77)

Dear Mr. Kirkpatrick:

PMEX Markets, Inc. (“PMEX”) appreciates the opportunity to submit comments in response to the Commodity Futures Trading Commission’s (the “Commission”) Request for Comment on the Listing of Compute Derivatives Contracts (the “Request”).¹

I. About PMEX

PMEX has applied for designation as a contract market (“DCM”) alongside its affiliate PMEX Clearing, Inc., which has applied for registration as a derivatives clearing organization (“DCO”).

PMEX intends to operate under the name Liquid Compute² and is developing an exchange where market participants can trade, hedge, and discover prices for compute resources. GPU and memory pricing can move 20–40% in months, and those moves are discontinuous rather than gradual. Prices step on hardware announcements and model releases: a next-generation accelerator or a materially more efficient model architecture can reprice existing capacity in a single day. Startups, chipmakers, and infrastructure providers are therefore structurally exposed to a risk that cannot be managed by averaging into a position over time. Unlike traditional commodities, compute has no forward curve, no standardized hedge, and no benchmark contracts to support the trillions in artificial intelligence (“AI”) capital expenditures expected over the next decade. We treat compute and memory like commodities, beginning with standardized GPU contracts (H100, A100, B200, and successors) and expanding into other critical AI inputs such as power.

Liquid Compute’s founders met at the University of California, Berkeley while studying electrical engineering and computer science and have experience implementing large-scale AI systems at Fortune 500 companies. Our team combines deep technical and market-structure experience, including former staff engineers from Meta Platforms, Inc. and NVIDIA Corp., with electrical engineering PhDs.

II. Overview

PMEX strongly commends the Commission, and Chairman Michael S. Selig in particular, for opening this inquiry now, before market structure calcifies around whatever happens to have been built first. In our view, the timing of a regulator’s first question is often more consequential than the content of its eventual rule, and the Commission has rightly asked this one early.

PMEX also endorses the framing. The Chairman’s observation that “America cannot win the AI race without a robust derivatives market for compute” matches what we see daily in the market, and the historical analogy he draws is the right one.³ American markets set the global standard for the commodities that powered the industrial economy, and they did so not by accident but through a specific institutional sequence: a standardized deliverable unit, a neutral venue, transparent price formation, and then a derivatives complex layered on top. That sequence is the substance of this letter.

We summarize our principal recommendations here and develop them below:

• Anchor regulatory analysis in cash-market and settlement-reference integrity. A cash-settled compute contract is only as sound as the index it settles against, and several indices now in use draw materially on indicative rather than executed prices.
• Encourage clear, standardized compute specifications that improve comparability and fungibility. We identify below the specification elements we believe are irreducible.
• Establish expectations for robust index governance, transaction support, contributor diversity, methodology transparency, and regulatory access to material settlement inputs.
• Recognize physical deliverability, where practicable, as one potential convergence mechanism, while requiring robust alternative safeguards for cash-settled contracts.
• Tailor surveillance and risk controls to compute-specific attributes, including hardware class, region, tenor, and concentration.
• Require contract-specific disclosure of material compute risks, including settlement-reference, hardware-obsolescence, power/interruption, and export-control risks, as applicable to ensure customer protection.
• Apply heightened scrutiny to perpetual products because their funding mechanism depends continuously on reference-price integrity.
• Keep the registered domestic path fast enough to be the attractive path, so that innovation in this asset class does not default to de minimis exemptions or offshore venues.
• Recognize the financing channel. Transparent compute prices are a precondition for non-dilutive debt financing of domestic AI infrastructure, which is where the Chairman’s framing translates most directly into national capacity.

III. The Compute Cash Market Today

This Section responds to Question 1 of the Request (Compute Cash Markets: Size, Liquidity, and Other Considerations). Each subsection below identifies the specific question(s) it addresses.

A. Structure

The Request asks about the size and liquidity of compute cash markets, what proportion of compute transactions occur at publicly disclosed prices, and whether a publicly observable transaction record exists.⁴ The most honest answer is that the market’s size is not reliably measurable with currently available data, and that this fact is itself the most important finding we can offer the Commission.

Compute capacity today is transacted almost entirely through bilateral contracts negotiated between a provider and a single counterparty. These contracts are typically long-dated, non-transferable, non-standard in their performance definitions, and confidential. There is no central record of transacted price, no standard unit against which prices could be compared even if they were disclosed, and, in most cases, no contractual right for a buyer to resell capacity it has purchased and cannot use. A buyer who over-procures does not sell; it strands. A provider with unsold inventory does not clear it into a market; it idles.

The scale of that inefficiency appears to be substantial. Because no compute market has existed, compute contracts are generally not assignable. A buyer who over-procures cannot resell, transfer, or hedge the excess, and a buyer who finds itself short cannot readily acquire capacity from a counterparty holding more than it needs. The error therefore runs in one direction: anyone who fears being short will over-procure, and that capacity will sit idle when the forecast proves wrong. At a moment when compute is becoming a scarce input across the economy, the aggregate volume of stranded capacity represents a real loss.

That is a market design failure, not a technology failure, and it is precisely the kind of failure that exchange infrastructure exists to solve. A venue that trades a standardized, assignable unit of capacity would allow a buyer holding excess capacity to resell it and a buyer that is short to acquire it, converting stranded capacity into available supply and producing an observable record of transacted prices. We describe the specification such a unit requires in Section III.C and the role of physical delivery in Section IV.C.

B. Compute Behaves Like Power, Not Like Oil

The Request asks whether particular commodity markets are especially relevant or similar to the compute cash market, and how compute differs from cash markets the Commission regulates, including whether the commodity is storable.⁵ Much commentary describes compute as “the new oil.”⁶ We think that analogy is misleading in ways that matter for regulatory design. Compute is not storable. A GPU-hour that goes unsold at 2:00 p.m. is not inventory; it is gone. That distinction is not merely descriptive; it determines how a forward curve can be priced at all. In a storable commodity, the forward is disciplined by cash-and-carry arbitrage: a trader can buy spot, store, and deliver later, so the forward cannot drift far from the spot plus the cost of carry. Compute has no such trade. No one can buy GPU-hours available for use today, store them, and deliver those same GPU-hours six months later. The compute forward is therefore expectation plus risk premium with no arbitrage tethered to spot. The Commission has long supervised this same structure in electricity, and that is the cleanest reason compute belongs analytically with power rather than with oil.

These characteristics also bear on fungibility across producers⁷. Compute is location-specific because it is delivered through constrained power and interconnect infrastructure and is subject to data residency and latency requirements. Compute is heterogeneous along more dimensions than most commodities: two clusters with identical GPU counts can differ by an order of magnitude in usable throughput depending on interconnect topology and network fabric. And compute’s marginal cost is dominated by electricity, tying it structurally to a commodity complex the Commission and the Federal Energy Regulatory Commission (“FERC”) already understand well.

The closer analogue is wholesale electricity. We would respectfully direct the Commission’s attention to the restructuring literature: Joskow’s work on wholesale power market design;⁸ Stoft’s Power System Economics;⁹ and Hunt’s Making Competition Work in Electricity.¹⁰ The literature addresses the central design problem compute now faces: how to define a deliverable unit of something that cannot be stored. FERC Order No. 888 (1996)¹¹ and Order No. 2000 (1999)¹² are instructive precedents for the proposition that open, standardized physical access is a precondition for competitive markets rather than a refinement of them.

Bandwidth capacity trading is a second relevant comparison, and a cautionary one. Between 1999 and 2002, Enron Broadband Services, Williams Communications, RateXchange, and Band-X each attempted to create a traded market in telecommunications bandwidth, and each failed for identifiable reasons: a non-fungible underlying with no accepted normalization, no standardized delivery points at which capacity could be handed off, no neutral clearing, and a one-sided market in secular price decline. The first three are addressed by the design choices discussed in this letter: a standardized specification, real delivery infrastructure, and a neutral venue and clearinghouse that is not also the dominant trader. The fourth is a condition compute does not share; compute demand is growing, and the market has natural interest on both sides between providers who are structurally long capacity and buyers who are structurally short capacity.

The successful comparisons point in the same direction. Forward freight agreements became liquid only after the Baltic Exchange published route indices built from assessed transactions in standardized vessel classes; cleared weather derivatives rely on an underlying measured by a neutral third party against a published specification; and the hashrate forward market developed after a hashprice index gave miners and buyers a common reference. In each case, the reference came first and the derivative followed.

C. The Missing Primitive: A Deliverable Unit

The Request asks the extent to which units of compute are fungible across producers without adjustment and, if not, what quality or grade adjustments are necessary to foster fungibility.¹³ No compute derivatives complex can be robust until market participants converge on what a unit of compute is. We do not suggest that the Commission should specify this by rule; contract design is properly the province of registered entities under Core Principle 3. But we would encourage the Commission, in evaluating any compute contract, to ask whether the specification addresses the following elements, each of which we have found to be load-bearing:

• Hardware class and configuration, specified to the SKU level, including memory configuration and any relevant firmware or driver constraints.
• Quantity and contiguity, including whether the delivered capacity must be co-located and interconnected as a single fabric, since a thousand scattered GPUs are not economically equivalent to a thousand contiguous ones.
• Interconnect and network fabric specification, stated explicitly rather than assumed. Networking is among the most frequently omitted and among the most economically significant: identical hardware provisioned at 400G versus 800G commands materially different prices, because fabric bandwidth, not GPU count, is the binding constraint on usable throughput for distributed training workloads. A contract silent on fabric has not specified the good being sold.
• Tenor and delivery window, including start date, duration, and any permitted interruption.
• Prepayment and payment structure, including the percentage of contract value payable in advance and the schedule for the remainder. The same capacity, on the same hardware, for the same tenor prices differently at 0%, 20%, and 30% prepayment, because prepayment shifts both financing cost and counterparty credit exposure between the parties. Any index that aggregates transactions without normalizing for prepayment terms is averaging prices that are not commensurable.
• Delivery location and jurisdiction, including data residency and the identity of the underlying data center.
• Power and cooling arrangements, including whether power is firm, interruptible, or curtailable.
• Availability and performance standards, expressed as measurable service levels rather than commercial best-efforts language.
• Remedies for degraded performance, silent substitution, or non-delivery.

A contract that leaves these terms to bilateral negotiation is not a standardized instrument, and an index built from such contracts is not comparing like with like, however many transactions it observes.

D. Purchase Modes, Observable Volume, and Price Behavior

The Request asks to distinguish among purchase modes and to state, for each, the volume transacted, the number of distinct counterparties, and whether any transaction price is published.¹⁴ PMEX does not presently have reliable market-wide data sufficient to quantify the total volume or distinct counterparties by purchase mode. The descriptions below therefore reflect market structure and PMEX’s observations, rather than market-wide estimates.

• On demand: Capacity billed by the hour or minute and available without commitment. This is the purchase mode in which a price is most routinely published, although posted rates are generally provider-administered rather than transaction-formed, and negotiated transactions may occur at discounts to posted rates.
• Spot and secondary: Short-tenor capacity transacted at negotiated prices through marketplaces, brokers, or direct bilateral arrangements. Transaction prices generally are disclosed only to the counterparties. This segment may become an important source of transaction-based index inputs as observable activity develops.
• Reserved: Capacity committed for a defined term on provider-standard terms, typically without material prepayment, at a discount to on-demand rates that scale with tenor and volume. The buyer selects from a published menu of commitment options rather than negotiating them. Transaction-level prices generally are not publicly disclosed, but the terms governing them are administered and therefore comparatively uniform across buyers.
• Committed and contracted: Longer-dated capacity agreements in which the commercial terms themselves are negotiated bilaterally, often with prepayment, minimum-draw obligations, and confidentiality provisions. Prepayment is a defining feature of this model and a key driver of price dispersion. These arrangements appear to represent a substantial share of the market’s economic value, but transaction prices are generally not publicly observable.

Two features of this distribution are particularly relevant to derivatives design. First, the most observable prices are often provider-posted rather than transaction-formed, while a substantial amount of economic activity occurs through confidential bilateral agreements. Second, the purchase modes are not fully interchangeable: an on-demand rate and a committed rate may reflect different tenor, availability, interruption, and performance terms. An index therefore should avoid combining non-comparable transactions without appropriate normalization.

Observed compute prices may also reprice discontinuously around hardware-generation changes, large capacity additions, and changes in power availability. Publicly cited price series often rely heavily on posted on-demand rates, so care is warranted before treating those series as representative of the broader bilateral market.

E. Measuring the Compute Market: A Proposed Methodology

The Request asks what data sources and methodologies should be used to assess market size, liquidity, transaction volume, participant types, and supplier concentration, and specifically asks commenters to distinguish public from non-public sources.¹⁵ PMEX believes the market should be evaluated using both categories, with the limitations of each source disclosed rather than embedded in a single headline estimate:

• Public source: installed accelerator base, derived from hardware-vendor shipment disclosures and public financial reporting, converted to a capacity measure using expressly stated duty-cycle assumptions.
• Public source: data-center capacity and power-interconnection information, including utility interconnection queues, regional transmission organization filings, and announced site capacity. These data can help bound deliverable compute because power and interconnection constraints may limit usable capacity.
• Public source: public-cloud disclosures, including revenue attributable to AI infrastructure and disclosed remaining performance obligations, where available, as indicators of committed-mode activity.
• Public source: colocation and independent-operator disclosures, including contracted capacity, utilization, and customer concentration where reported.
• Non-public source: marketplace, broker, and bilateral transaction data. These data are among the most useful sources for actual transaction prices, but they are not generally available on a market-wide basis.
• Non-public source: transaction data generated by registered or otherwise observable physical venues. As such venues develop, transaction-level data could become an important source for regulatory analysis, subject to appropriate confidentiality protections.

Each source is imperfect. The Commission should therefore require transparency regarding the inputs and assumptions supporting any market-size or liquidity estimate. Supplier concentration also warrants particular attention because it can affect both reference-price representativeness and deliverable supply.

With respect to producers and market share, PMEX’s qualitative observation is that concentration should be considered at more than one level.¹⁶ Accelerator manufacturing is highly concentrated, while capacity operation is more dispersed but may still be concentrated within particular hardware, region, or tenor buckets. The public data presently available are not sufficient for PMEX to provide a reliable market-wide concentration estimate.

On whether a price reporting agency or enforced standardized methodology exists¹⁷: we are not aware of any price reporting agency for compute operating under an enforced standardized methodology comparable to those the Commission is accustomed to in energy or agricultural markets. Several commercial index and price-assessment products exist, and we address them in Section IV. We are not aware of any that is subject to an external assurance regime.

The Request also asks whether any audits, studies, or independent verifications of compute transaction data have been conducted and what they concluded.¹⁸ We are not aware of any. In the absence of such work, we suggest that the absence is itself a material fact in evaluating any contract that settles to a compute price series, and that the Commission should be slow to treat publication as evidence of verification.

F. Settlement Against Data the Commission Cannot Observe

The Request asks whether it would be appropriate to permit trading in a derivative contract settling to a price computed from data the Commission may not be able to observe, verify, or surveil.¹⁹ PMEX does not believe a listed contract should settle to a reference price whose material inputs cannot be made available, on an appropriate confidential basis, to the listing DCM and the Commission for verification and surveillance.

Core Principle 4 requires a designated contract market to have the capacity and responsibility to prevent manipulation, price distortion, and disruption of the cash-settlement process. Appendix C similarly emphasizes the reliability of a cash-settlement price and the integrity of the underlying cash market. Those objectives are difficult to satisfy where material settlement inputs cannot be reconstructed or reviewed.

PMEX suggests distinguishing between data that are confidential and data that are unobtainable. Confidential transaction data may still be usable if the index administrator and contributing venues or parties are subject to arrangements that make material input data available to the listing DCM and the Commission for surveillance, verification, and reconstruction of the settlement calculation. Inputs that cannot be made available on those terms should not be relied upon as material settlement inputs.

G. Deliverable Supply, Position Limits, and Accountability

The Request asks what an appropriate deliverable supply estimate methodology for purposes of Core Principle 5 would be,²⁰ and separately asks for an estimated deliverable supply for compute at the relevant pricing points.²¹ We take the two together.

PMEX cannot presently provide a reliable numerical estimate of deliverable supply because the answer depends on the contract specification and on data that are not available on a market-wide basis. Until the specification fixes hardware class, contiguity, interconnect, tenor, delivery location, and power arrangements, the set of capacity capable of satisfying a delivery obligation is not sufficiently defined. Subject to that limitation, PMEX proposes the following methodology:

• Begin from installed capacity conforming to the contract specification at the delivery location or locations, rather than from total regional compute.
• Reduce to capacity that is uncommitted during the delivery window. Capacity subject to an existing committed-mode agreement generally would not be economically deliverable during that period, and this adjustment may be significant.
• Reduce further to capacity whose operator has been onboarded to the venue and verified as capable of delivery. Unverified capacity is not economically deliverable and treating it as deliverable supply overstates the market’s capacity to absorb a squeeze.
• Apply a power-availability constraint at the site level. A site may hold conforming hardware and lack firm power for the delivery window, particularly in regions where interconnection is constrained.
• Measure at the pricing point rather than in aggregate, where a pricing point is the combination of hardware class, delivery region, and tenor bucket. Aggregate deliverable supply is not a meaningful constraint on a position concentrated in one bucket.
• Re-measure at a defined frequency, monthly at a minimum. Deliverable supply in this market moves on hardware deployment and contracting cycles rather than on a production or harvest cycle, and an annual estimate will be stale.

On position limits and accountability, PMEX believes that, given current data limitations, position accountability may be more readily calibrated than a hard numerical limit. Accountability levels could be set at the relevant pricing-point level rather than solely on an aggregate basis and could permit the exchange to request position, delivery-capability, and underlying-exposure information when a threshold is reached. As reliable deliverable-supply data develops, the appropriate framework can be reassessed.

Non-storability changes the nature of squeeze and corner analysis. It may reduce the relevance of classic inventory-accumulation scenarios, but it can increase delivery-window risk because new or substitute capacity may require materially more time to deploy than the delivery window permits. The Commission therefore should focus its Appendix C analysis on supply adequacy at the relevant pricing point and delivery window.

H. Anticipated Effects of Listing on the Compute Cash Market

The Request asks whether a listed contract settling to a published compute index could change provider incentives.²² PMEX believes it could, including through the channels identified by the Commission.

• Publication of posted rates. Once a provider-administered rate becomes an input to a settlement index, it acquires significance beyond its ordinary commercial function. A provider whose posted rate becomes an index input may have incentives to alter the timing, granularity, or level of that rate. This is one reason transaction-based inputs are preferable where available.
• Pricing and structuring of bilateral reservations. A listed contract may provide an alternative route to price exposure and could influence bilateral pricing. Providers and purchasers may also respond through non-price terms, including minimum-draw, substitution, curtailment, and prepayment provisions, which a price-only index may not capture.
• Disclosure of utilization and committed capacity. Utilization and committed-capacity information is commercially sensitive. As derivatives markets increase the value of such information, voluntary disclosure incentives may change, reinforcing the importance of clearly defined regulatory and index-data arrangements.
• Allocation of capacity among purchasers. A provider that both supplies capacity and holds derivative exposure may have an incentive to direct volume toward or away from observed venues or to alter transaction timing relative to an observation window. Surveillance and index methodology should be designed with that possibility in mind.

The parties best positioned to influence a reference price may overlap with the firms supplying capacity, particularly where a small number of providers contribute a material share of observable transactions or posted rates. Similar issues arise in other commodity benchmarks, but compute’s developing cash-market transparency and concentration make contributor identification, transaction-based inputs, and conflict management especially important.

IV. Market Oversight and Manipulation Concerns

This Section responds to Question 2 of the Request (Market Oversight and Susceptibility to Manipulation). Each subsection below identifies the specific question or questions it addresses.

A. The Reference-Price Problem

The Request asks what features a compute derivatives contract that cash settles to an index calculated over predominantly bilateral and privately priced transactions would need to demonstrate to satisfy Core Principle 3.²³ Section 5(d)(3) of the Commodity Exchange Act and DCM Core Principle 3 require that listed contracts not be readily susceptible to manipulation. For a cash-settled compute future, the integrity and manipulability of the settlement reference are central components of that analysis. Where the reference price can be materially influenced through thin, concentrated, or non-transactional inputs, derivatives-market surveillance alone may not address the source of the risk.

We respectfully suggest that this is the area of greatest present risk in compute markets. Based on PMEX’s review of publicly available methodology materials, some compute price series use a mix of executed transactions, posted rates, indicative quotes, or assessed inputs. The relative weight, transaction count, and contributor concentration may vary by hardware, region, and tenor bucket. The Commission therefore should evaluate the robustness of each settlement reference based on its actual methodology, observable transaction support, governance, and susceptibility to influence rather than assuming that publication alone establishes robustness.

B. Recommended Index Standards

This subsection responds to Questions 2(a) and 2(b), including the Request’s question regarding protections that would prevent a capacity provider from adjusting a posted rate, directing capacity onto or away from an observed venue, or executing or declining to execute transactions during the observation window. The final two standards below address those mechanisms directly.²⁴ We encourage the Commission to articulate expectations along the following lines for any index used to settle a listed compute contract:

• Inputs should be executed transactions wherever available, with any indicative or assessed component separately disclosed as a percentage of the index.
• Governance should conform to the IOSCO Principles for Financial Benchmarks, including a documented methodology, a change-control process, and independent oversight.
• The methodology should specify minimum transaction counts, minimum notional volume, and minimum contributor diversity per published bucket, with a defined fallback when thresholds are not met.
• Contributor concentration should be measured and disclosed because a reference price may become more susceptible to influence when one seller or contributor accounts for a large share of observed volume.
• In PMEX’s view, an administrator of a settlement reference generally should not take positions in contracts that reference it. This approach is consistent with existing international standards: IOSCO Principle 3 requires benchmark administrators to identify and address conflicts arising from their own or their affiliates’ trading activities and to maintain appropriate separation between benchmark administration and business lines affected by the benchmark. Where an administrator trades contracts settled by its own index, it may have a direct financial interest in the index outcome, and internal policies alone cannot fully address that conflict. Where an administrator owns, or is affiliated with, a venue that contributes input data, the relationship should be disclosed in the methodology itself rather than in a separate governance document. While the concentration of multiple roles within a single corporate group is not inherently impermissible, market participants that rely on the benchmark to price or settle contracts would benefit from transparency regarding those relationships. For its part, the parent company of PMEX publishes indices and does not take positions in contracts that reference them.
• The methodology should be robust to a contributor directing capacity toward or away from the venues whose transactions it observes. We would suggest minimum contributor-diversity thresholds per published bucket, disclosure of each contributor’s share of observed volume, and a documented review where a contributor’s share changes materially between periods.
• The observation window should be designed so that a contributor cannot materially move the published price by executing, or declining to execute, within it. Extending the window, calculating on a volume-weighted rather than last-price basis, trimming outliers, and randomizing the window within a disclosed range are each established techniques in other benchmark contexts, and each is available here.

The Request asks what volume, transaction-frequency, and contributor-concentration data characterize the population from which a compute reference price may presently be calculated.²⁵ PMEX’s observations suggest that some hardware-region-tenor buckets may be supported by limited transaction counts and contributor diversity, but PMEX does not have market-wide data sufficient to quantify those characteristics reliably.

For the same reason, PMEX does not believe the public data presently available support a universal numerical threshold applicable to all compute reference prices. The Commission could instead require each methodology to establish and disclose minimum transaction, notional, and contributor-diversity thresholds appropriate to the relevant bucket, together with a defined fallback when those thresholds are not met:

• A minimum number of executed transactions per published bucket per observation period, from a minimum number of distinct contributors, with no single contributor exceeding a defined share of observed notional.
• A published fallback where thresholds are not met — extension of the observation window, aggregation to a broader bucket with disclosure, or suspension of publication — rather than silent substitution of indicative inputs, which is the response we would most expect and least want.
• Disclosure, with each published price, of the transaction count, notional, contributor count, and indicative share actually supporting that print. A participant should be able to assess the robustness of the specific number against which its position settles, not only the methodology in the abstract. We regard this as the single most valuable index reform available and the cheapest to implement.

C. Physical Delivery as an Anti-Manipulation Anchor

This subsection further responds to Question 2(a) and Question 2(i), which invites other considerations relevant to a DCM’s compliance with the Core Principles. Physical deliverability, where commercially practicable, can provide an additional convergence mechanism between the derivatives and cash markets. For cash-settled contracts, comparable discipline must instead come from a robust settlement methodology, sufficient transaction support, transparent governance, and the DCM’s ability to obtain underlying data for surveillance and settlement reconstruction.

We recognize that physical delivery of compute is operationally demanding — it requires provider onboarding, capacity verification, an access-control layer, and a defined remedy when delivery fails. That difficulty is precisely why we chose to build the physical venue first rather than the index-and-futures layer, which is materially easier. We note that the Commission has precedent for authorizing a venture-backed entrant to operate a trading venue and affiliated clearinghouse for fully collateralized, physically settled contracts in a novel asset class. In 2017, the Commission granted LedgerX, LLC registration as a derivatives clearing organization to clear fully collateralized digital currency swaps,²⁶ and in 2019 the Commission designated LedgerX as a contract market.²⁷ We believe that precedent is instructive here.

D. Surveillance Data

The Request asks what surveillance capabilities would be necessary to satisfy Core Principle 4 for compute derivatives, whether those capabilities are presently feasible, and whether a DCM should be required to maintain an information-sharing arrangement with each compute venue and capacity provider whose data enter a settlement reference.²⁸ Effective surveillance in this market will require fields that do not exist in the Commission’s current reporting taxonomies. Position and large trader reporting for compute should, in our view, capture hardware class at SKU granularity, delivery region, tenor bucket, and contiguity, because economically identical-looking positions in different buckets carry entirely different risk. Reporting that aggregates “GPU compute” into a single line item will obscure the concentrations that matter.

The Commission should also consider information-sharing expectations between physical compute venues, index administrators, and derivatives markets that reference their data. Where those functions sit in separate firms, effective cross-market surveillance may depend on contractual or other arrangements that permit access to relevant transaction information.

The Request asks whether these capabilities are feasible from a technological, operational, and legal perspective.²⁹ Technologically and operationally, PMEX believes the relevant data fields can be captured where a venue transacts physical capacity. The legal dimension requires additional care because compute capacity providers generally are not CFTC registrants and therefore are not subject to the routine reporting, recordkeeping, and supervisory obligations applicable to registered derivatives-market participants. Providers also may be subject to confidentiality obligations or located outside the United States.

A venue-level information-sharing arrangement may be appropriate where transactions from a particular physical venue materially enter a settlement reference. Requiring a separate agreement with every individual capacity provider may be impracticable for an index with a broad and changing contributor base. One workable alternative is for the index administrator to require contributors, as a condition of inclusion, to permit relevant transaction-level data to be made available to the Commission and the listing DCM on request, subject to appropriate confidentiality protections.

E. Whether a Conforming Reference Series Exists Today

The Request asks whether there is a cash price series for compute that could serve as a reference price satisfying the Appendix C criteria — reliable, acceptable, publicly available, and timely, computed from a cash market that is sufficiently liquid and not itself readily susceptible to manipulation — and asks us to describe any such series and to say what the Commission should do if none is available.³⁰

Based on publicly available methodology documents and the information presently available to PMEX, we do not have sufficient information to conclude that an existing compute price series satisfies all of the Appendix C criteria for use as a listed contract’s settlement reference. The Commission should evaluate each series directly based on its methodology, transaction support, contributor concentration, governance, and ability to provide access to material inputs.

This does not mean that existing compute price series lack informational value. It means that usefulness as a market indicator and suitability as the settlement reference for a listed derivatives contract are different standards, and the latter warrants a more demanding review.

As to what the Commission should do in the absence of a conforming series, we would offer the following suggestions:

• Articulate clear expectations for compute settlement references before market convention solidifies around a particular series, including expectations regarding transaction support, contributor diversity, governance, methodology transparency, and access to material inputs.
• Consider contract structures that reduce reliance on opaque third-party inputs. Depending on the product, that may include physical delivery or settlement to a transaction-based price generated by a venue with sufficient observable activity and appropriate governance.
• Use a seasoning mechanism of the kind described in Section VI. A series that has published continuously with adequate transaction support over a defined period has demonstrated something a methodology document cannot.

F. Market Safeguards, Risk Controls, and Liquidity Protections

The Request asks what market safeguards, risk controls, and liquidity protections may be appropriate for compute derivatives.³¹ Depending on contract design and liquidity conditions, a DCM could consider safeguards such as the following:

• Price bands calibrated to the behavior of the underlying market, with parameters capable of accommodating both relatively stable periods and discontinuous repricing around hardware-generation or capacity events.
• Velocity logic and trading pauses calibrated to prevailing liquidity, with pause periods sufficient to permit orderly repricing where the book is thin.
• Maximum order-size controls and self-match prevention, particularly where related entities may participate in multiple roles across the physical and derivatives markets.
• Minimum listing conditions tied to the reference index, so that a contract is not listed against a bucket in which the index does not meet its own volume and contributor thresholds.
• Settlement-period protections, including use of a sufficiently representative observation window and transparent fallback provisions where the reference price is anomalous or unavailable.
• Position accountability at the pricing-point level as described in Section III.G, with escalation to the exchange’s market regulation function on breach.
• Published escalation protocols for the case in which the reference index produces an anomalous print or fails to publish, including the contract’s fallback settlement provision.

Because the settlement reference may be a principal source of risk in a new asset class, fallback provisions should be specified in the contract terms in advance and disclosed to participants rather than determined solely through post-event discretion.

G. Observed Conduct in the Compute Cash Markets

The Request asks whether there are types or patterns of trader or intermediary conduct in compute cash markets that raise risks warranting close monitoring, and whether any of it requires additional Commission action.³²

The Commission may wish to monitor incentives associated with strategic over-procurement, secondary resale, related-party transactions, provider-administered posted rates, and bilateral conformity disputes. PMEX does not suggest that these practices are unlawful; rather, each may affect the representativeness, transparency, or observability of cash-market pricing. Where such activity contributes to a settlement reference, contributor identification and access to underlying transaction data become particularly important. This discussion also responds to Question 3(e), which asks about trader or intermediary conduct particular to compute cash and derivatives markets.

V. Customer Protection

The Request asks about customer protection.³³ We would distinguish two categories: the traditional protections concerning customer funds, which are well developed and translate directly to this asset class; and a set of protections specific to compute, which do not yet exist anywhere. Existing customer-funds, clearing-member financial standards, default-management, and disclosure protections should apply to compute derivatives without dilution. Compute-specific protections should additionally address risks arising from the underlying market and settlement mechanism, as applicable to the particular contract. For physically delivered contracts, provider verification and delivery-capability standards may be important where the provider population ranges from large institutional operators to smaller capacity providers.

• Hardware substitution and conformity. For physically delivered contracts, the specifications should address hardware substitution and define how conformity with the required hardware and configuration will be measured.
• Performance degradation. Delivery terms should also address material performance degradation, interruption, and remedies where delivered capacity does not conform to the contract specification.
• Index transparency for customers. For cash-settled contracts, customers should have access to the settlement methodology, material information about input composition, and sufficient information to understand the transaction support for the reference price.

A. Anti-Money Laundering and Customer Identification

The Request asks whether compute presents heightened Anti-Money Laundering (“AML”) or customer-identification concerns and what challenges intermediaries may face.³⁴ A futures commission merchant carrying compute futures generally faces the same core customer-identification obligations applicable to other futures products, but compute may introduce additional diligence considerations where derivatives activity intersects with the physical delivery market.

Those considerations may include the identity and location of capacity providers, the jurisdiction in which compute is delivered, sanctions and export-control restrictions applicable to advanced accelerators, and payment structures associated with longer-term capacity agreements. The Commission and the National Futures Association may wish to evaluate whether existing AML, sanctions, and customer-identification guidance adequately addresses these compute-specific delivery and geographic risks.

B. A Geopolitically Sensitive Underlier

The Request asks what customer-protection and disclosure considerations arise from a contract settling against a geopolitically sensitive commodity, and how they differ from a contract settling against a commodity such as oil.³⁵ The difference, in our view, is one of mechanism and speed rather than of kind.

Oil carries substantial geopolitical risk, but that risk is transmitted through supply and demand: a disruption removes barrels and the price responds. Compute is exposed to a regulatory channel that operates directly and discontinuously. An export-control determination, a licensing decision, or a restriction on the deployment of a hardware generation in a jurisdiction can reprice compute without any change in physical supply or demand and can do so on the basis of government action that is not announced in advance, is not forecastable from market data, and in some cases is not public when taken. Three consequences follow for disclosure:

• Some export-control or jurisdictional events may affect multiple hardware classes or delivery locations simultaneously, limiting the effectiveness of diversification within the asset class. That possibility should be disclosed as a characteristic of the underlying if material to the contract.
• Large accelerator purchasers and operators may have greater visibility into licensing posture, allocation, or capacity constraints than other market participants. Such informational asymmetries may be relevant to disclosure and surveillance design even where the information is obtained through ordinary commercial activity.
• Historical price data will understate tail risk. Realized volatility computed over a period containing no control event does not describe the distribution. Any disclosure presenting historical volatility for a compute contract should say so expressly.

PMEX therefore supports contract-specific disclosure of material export-control and supply-restriction risks where those risks could affect the underlying reference market, deliverability, or price formation.

C. Idiosyncratic Risks in the Terms and Conditions

The Request asks whether the terms and conditions of a compute futures contract should include information related to idiosyncratic risks.³⁶ PMEX believes they should. Because compute contracts may expose participants to risks not captured by generic futures disclosures, the contract terms and related disclosures should clearly identify material risks specific to the underlying and settlement mechanism, including as applicable:

• The settlement reference in full: its administrator, its methodology, the proportion of indicative or assessed input, the minimum transaction and contributor thresholds applied, and the fallback governing a failure to meet them or to publish.
• Any affiliation among the index administrator, the listing venue, contributing venues, and market makers in the contract.
• Hardware obsolescence, stated as a characteristic of the underlying: that the referenced hardware class has a finite economic life, that successor generations reprice it discontinuously, and that the contract does not adjust for generational transition unless the specification so provides.
• Power and interruption terms, including whether the capacity underlying the reference is firm, interruptible, or curtailable. A price series computed over interruptible capacity is not comparable to one computed over firm capacity, and a participant should not have to infer which it holds.
• Export-control and supply-restriction risk, as described in Section V.B.
• For a physically delivered contract, the conformity standard, the measurement method, the party performing measurement, and the remedies for substitution, degradation, and non-delivery.
• For a perpetual contract, the funding-rate calculation, its observation windows, any caps, and the circumstances in which it may be adjusted.

D. Retail Participation

On retail participation³⁷, compute primarily is a wholesale industrial input, and PMEX does not presently identify a natural retail hedging use case. If retail-facing compute products develop, the Commission should consider the characteristics of the particular product, including leverage, liquidity, settlement methodology, and the complexity of the underlying reference market.

At a minimum, retail participants should receive clear contract-specific risk disclosure explaining the settlement reference and material compute-specific risks. Additional leverage or access controls may be appropriate depending on the robustness of the reference market and the design of the product.

VI. Perpetual Compute Futures

Perpetual compute futures could reduce roll costs and basis risk for participants with continuous exposure, which may make them useful for some commercial hedging strategies. As such, PMEX supports the Commission’s willingness to consider perpetual compute futures,³⁸ and we think the instinct behind them is sound. Compute exposure is continuous rather than calendar-bound, and forcing hedgers into quarterly expiries imposes roll costs and basis risk that a perpetual structure avoids. For a data center operator hedging a rolling exposure, a perpetual is a better instrument than a strip.

At the same time, a perpetual contract depends continuously on its funding mechanism and the reference price used to calculate that mechanism. Where the underlying reference market is thin or concentrated, that continuous dependency can create additional susceptibility to distortion and should be considered in the product-design and Core Principle analysis. A dated futures contract has expiry convergence as a backstop: however imperfect the reference index, the contract must ultimately settle against something, and the approach of expiry disciplines the price. A perpetual contract has no expiry. Its tether to the underlying is the funding rate, which is computed continuously from the spread between the perpetual price and the spot reference. This means the perpetual is more dependent on the quality of the spot reference than a dated contract is, not less, and it is dependent on it continuously rather than at a single point.

In a market where the spot reference is thin, that dependency is a live manipulation surface. The funding rate itself becomes a target: a participant able to move the reference index, even briefly, can extract funding payments from the other side of the market without ever needing the position to settle. This is not hypothetical; analogous dynamics have been observed in perpetual markets in other asset classes with thin or concentrated spot references.

Accordingly, if perpetual compute products are listed, PMEX believes the Commission should focus on the following safeguards.

• Require the underlying reference index to satisfy clearly articulated robustness standards appropriate to the product.
• Consider a seasoning requirement — a period during which the reference index has published continuously with adequate transaction support — before perpetual products referencing it are listed.
• Require transparency in the funding-rate mechanism itself: the calculation, the observation windows, the caps, and the circumstances under which it can be adjusted.
• Examine margin methodology for a continuously funded product that trades outside conventional hours, and the interaction with DCO Core Principle D risk management, including intraday margining and the treatment of funding obligations in default.
• Scrutinize any auto-deleveraging or socialized-loss mechanism proposed for a perpetual compute product, since such mechanisms transfer losses to profitable participants and sit uneasily with customer protection expectations in a regulated market.

These considerations do not argue against perpetual compute futures, but they support a more demanding review of reference-price integrity and funding-rate design before such products are listed.

VII. The Financing Channel: Where Market Structure Meets National Capacity

Greater transparency in compute pricing may also support financing of compute infrastructure by giving lenders more observable information regarding contracted pricing, utilization, and residual value. PMEX believes this is a potential secondary benefit of more transparent cash and derivatives markets, although additional empirical work would be needed to quantify the effect.

Companies purchasing GPUs to deliver inference are financed inefficiently today. The available capital is predominantly dilutive equity, or venture debt that functions less as asset-backed lending than as a bet on the borrower’s next financing round. Institutional credit investors are attracted to the yield, but generally cannot underwrite the asset class, because pricing, utilization, hardware obsolescence, and customer concentration are not observable. The result is a financing gap in the middle market, estimated to be roughly $50 million to $250 million per transaction, where deals are too small and too operationally specific for the bond market and the largest infrastructure lenders, and too specialized for generalist private credit.

This gap has a direct and quantifiable cost: American AI infrastructure is being built with equity that could be built with debt, which means domestic operators give up more ownership per unit of capacity deployed than they would in any mature asset class.

Transparent, observable compute prices change this. With live market data, a lender can test whether a borrower’s contracted rate sits above or below the clearing market, and therefore whether the contracted revenue is durable or vulnerable to renegotiation. It can mark hardware collateral against observable prices rather than a scheduled depreciation curve and calibrate amortization to stay ahead of a collateral value that can fall discontinuously when a new hardware generation arrives. It can estimate recovery value as a current market exercise rather than a static appraisal. And it can refresh all of these continuously over the life of a loan, identifying margin compression earlier than a quarterly reporting cycle would.

The hedging need is already large and already unmet. Lenders have extended multi-billion-dollar GPU-collateralized facilities to large compute operators, and those lenders now hold substantial residual-value and contracted-cash-flow exposure with no instrument available to hedge it. We would suggest that unhedged, concentrated residual-value exposure of this kind is worth the Commission’s attention on its own terms, independent of the market development question.

The policy conclusion is straightforward: a functioning compute derivatives market lowers the cost of capital for domestic compute buildout. That is the mechanism by which market structure translates into national capacity, and we think it is the strongest available answer to anyone who regards this inquiry as premature.

VIII. Consolidated Recommendations

We respectfully offer the following for the Commission’s consideration:

1. Anchor the framework in cash-market and settlement-reference integrity. Evaluate compute derivatives by reference to the quality, observability, governance, and manipulability of the cash market inputs supporting settlement.

2. Encourage clear and standardized product specifications. Contract terms should identify the hardware, region, tenor, performance, and other attributes necessary to compare economically similar units of compute. Without specifying contract terms by rule, the Commission can signal the specification elements it expects registered entities to address under Core Principle 3.

3. Establish expectations for robust settlement references. Methodologies should address transaction-based inputs, contributor diversity, concentration, governance, transparent fallbacks, published methodology, and regulatory access to material settlement data.

4. Build compute-specific surveillance capabilities. Surveillance and reporting should capture the attributes that drive risk, including hardware class, delivery region, tenor, and relevant concentration, and should support information sharing where physical-market data feed settlement.

5. Treat physical deliverability as a favored anti-manipulation feature. Convergence discipline is worth the operational cost, and the Commission should not inadvertently advantage the easier index-only path.

6. Require contract-specific risk disclosure. Disclosures should address material settlement-reference, hardware-obsolescence, power/interruption, export-control, and delivery risks, as applicable to the contract.

7. Apply heightened scrutiny to perpetual products. If listed, perpetual compute contracts should be supported by a robust reference price, transparent funding rate methodology, and risk controls appropriate to continuous funding and thin-market conditions.

8. Extend customer protection to delivery performance. Provider verification, conformity measurement, remedies for substitution and degradation, and index transparency for customers.

9. Keep the registered domestic path fast enough to be the attractive one. Meaningful activity in this asset class currently operates under de minimis exemptions or through offshore venues. We do not fault firms for making rational choices under regulatory uncertainty. But the Commission’s objective of establishing American leadership is served by ensuring that full domestic registration is a competitive choice rather than a costly one, and that new entrants can reach designation on a timeline the market will wait for.

10. Coordinate with FERC and the Department of Energy. Compute delivery is, at the margin, power delivery. The basis relationship between compute prices and regional power prices will become one of the most important exposures in this complex, and the two regulatory perimeters will need to meet.

11. Convene a staff roundtable. We would encourage the Commission to bring physical venues, index providers, clearing firms, compute operators, and lenders together in a public forum before proposing rules. We would attend and contribute data.

IX. Conclusion

PMEX appreciates the Commission’s early focus on compute-market structure. In PMEX’s view, durable compute derivatives markets will depend on transparent and observable cash-market inputs, clear product standardization, robust settlement methodologies, and surveillance tools tailored to the characteristics of compute. We are not asking the Commission for an exemption, a safe harbor, or a delay. We are asking for the opposite: clear rules, applied early, that make the standardized and deliverable version of this market more attractive to build than the fragile one. We have organized our company around the premise that the Commission will do exactly that, and we have applications pending that reflect it.

The United States has done this before. The institutional machinery that made grain, energy, and interest rate risk tradeable was not a natural feature of those markets; it was designed, and much of it was designed here. The Chairman is right that compute is the commodity that will power the intelligence economy, and right that the rules of the road should be American ones. We would be glad to help write them.

PMEX would welcome the opportunity to continue to meet with Commission staff, to provide transaction-level data from our physical venue as it develops, or to respond to any follow-up questions. Please do not hesitate to contact us.

Sincerely,

Ronit Jain
Chief Executive Officer, Liquid Compute
45 Main Street, Suite 422, Brooklyn, New York 11201
ronit@liquidcompute.com | (804) 380-4981

Notes

¹ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. 54259 (Aug. 21, 2026) (to be codified at 17 C.F.R. pts. 1 and 38).

² For more information, see Liquid Compute, https://liquidcompute.com (last visited Oct. 19, 2026).

³ Press Release, CFTC, CFTC Requests Comment on the Listing of Compute Derivatives Contracts (Aug. 19, 2026), https://www.cftc.gov/PressRoom/PressReleases/9286-26.

⁴ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Questions 1(b), 1(c), and 1(e)(ii)).

⁵ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Questions 1(a) and 1(e)(i)).

⁶ Yun Li, The new oil? Inside the effort to turn AI computing power into a tradeable commodity, CNBC (June 16, 2026), https://www.cnbc.com/2026/06/16/the-new-oil-inside-the-effort-to-turn-ai-computing-power-into-a-tradeable-commodity.html.

⁷ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. 54263 (Question 1(e)(iv)).

⁸ Paul L. Joskow, Challenges for wholesale electricity markets with intermittent renewable generation at scale: the US experience, 35 Oxford Rev. of Econ. Pol’y 291 (2019), https://academic.oup.com/oxrep/article-abstract/35/2/291/5477314.

⁹ Steven Stoft, Power System Economics: Designing Markets for Electricity (Wiley-IEEE Press, 2002).

¹⁰ Sally Hunt, Making Competition Work in Electricity (John Wiley & Sons Inc., 2002).

¹¹ Promoting Wholesale Competition Through Open Access Non-Discriminatory Transmission Services by Public Utilities; Recovery of Stranded Costs by Public Utilities and Transmitting Utilities, Order No. 888, FERC Stats. & Regs. ¶ 31,036 (1996).

¹² Regional Transmission Organizations, Order No. 2000, 65 Fed. Reg. 809 (Jan. 6, 2000), FERC Stats. & Regs. ¶ 31,089 (1999), order on reh’g, Order No. 2000-A, 65 Fed. Reg. 12088 (Mar. 8, 2000), FERC Stats. & Regs. ¶ 31,092 (2000).

¹³ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. 54263 (Question 1(e)(iv)).

¹⁴ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 1(a)).

¹⁵ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 1(b)).

¹⁶ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 1(e)(iii)).

¹⁷ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 1(e)(v)).

¹⁸ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 1(d)).

¹⁹ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 1(c)).

²⁰ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 1(g)).

²¹ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54264 (Question 2(h)).

²² Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 1(f)).

²³ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 2(a)).

²⁴ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Questions 2(a)-(b)).

²⁵ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 2(c)).

²⁶ Press Release, CFTC, CFTC Grants DCO Registration to LedgerX LLC (July 24, 2017), https://www.cftc.gov/PressRoom/PressReleases/7592-17.

²⁷ Press Release, CFTC, CFTC Approves LedgerX LLC as a Designated Contract Market (June 25, 2019), https://www.cftc.gov/PressRoom/PressReleases/7945-19.

²⁸ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 2(d)).

²⁹ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 2(d)).

³⁰ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263–64 (Question 2(g)).

³¹ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 2(e)).

³² Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54263 (Question 2(f)).

³³ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54264 (Question 3).

³⁴ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54264 (Question 3(a)).

³⁵ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54264 (Question 3(b)).

³⁶ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54264 (Question 3(c)).

³⁷ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54264 (Question 3(d)).

³⁸ Request for Comment on the Listing of Compute Derivatives Contracts, 91 Fed. Reg. at 54264 (Question 4).