Nine of the ten information points in the story I read this week trace back to a single interested party. The tenth is the act of submission itself. That is not a data set. It is a press release wearing a lab coat.
The story, stripped to its ledger: HPC — a policy advocacy group whose name maps almost too cleanly onto Hyperliquid — has filed a comment letter with the U.S. Commodity Futures Trading Commission arguing that compute should be treated as a commodity and listed as perpetual futures on registered American venues. Six measures. Zero contracts. No testnet. No designated contract market filing. No CFTC response. Not even a clean timestamp: the letter is dated October 22, while one of its measures references a policy statement from 2026. Either it is a forward-looking document written a year early, or it is a document whose publication date nobody reconciled. Both possibilities change what the thing is worth.
I have spent five years building models that separate what a market says about itself from what its ledger says. Most of the time the gap is noise. Occasionally the gap is the entire story. This letter is the purest test case I have seen since the Terra autopsy — not because the idea is fraudulent, because it is not, and several of its measures are technically defensible — but because the evidentiary base is one stakeholder describing its own preferred future in the grammar of settled fact.
That distinction is the whole article. What follows is not a verdict on whether compute should trade. It is an audit of whether the thing being proposed can actually be built, who it would actually benefit, and what would have to be true for the narrative to survive contact with a settlement date.
The Only Verifiable Fact
Here is what is actually on the table, separated from what the framing wants you to assume is on the table.
HPC submitted a comment letter. That is the one verifiable event. Everything else — the six measures, the claim that compute qualifies as a commodity, the assertion that spot market maturity should not be a precondition for listing, the proposal to let competing indices settle in parallel, the self-admitted concern that early markets will attract manipulation and concentration, the two proposed sale formats of GPU hours and inference tokens, the principles for a reliable compute reference index — is advocacy. Advocacy is legitimate. Advocacy is also not evidence, and the difference matters enormously when the subject is a derivative that will eventually settle real money against a number nobody has yet defined.
I have not seen the letter text. I have not run live checks on total value locked, open interest, funding rates, or valuation multiples on any related token. Where a number should sit, I am leaving a hole rather than inventing one. In my 2022 Terra work I traced $2.3 billion in outflows across 50,000 wallets and published the dashboard in real time, and the single most important design choice in that dashboard was that every panel showed its own confidence level. Readers trusted the autopsy because it refused to guess. The same discipline applies here, at a much earlier stage of the lifecycle, where the temptation to fill gaps with narrative is far stronger.
So the honest framing is this: the object of analysis is not a regulatory fact. It is a policy position held by a party with an economic interest in the outcome. Everything downstream — every price, every inference, every 'compute is the new oil' thread — inherits that provenance.
The Machinery Being Proposed
Perpetual futures are the least mysterious part of this proposal, and it is worth saying so plainly, because the industry tends to over-credit novelty.
The mechanical structure is settled engineering. A perpetual contract has no expiry; it tracks a mark price and uses a funding rate to tether that mark to an underlying reference; a liquidation engine enforces margin. Hyperliquid and dYdX have run this machinery at scale for years. The funding mechanism exists because a perpetual has no delivery date to force convergence, so the protocol pays one side and charges the other at intervals, and arbitrageurs are supposed to close the loop by hedging against spot. That is the entire economic function of the funding rate: it is not a yield, it is a tether.
Everything that follows in this article is a consequence of that single sentence.
The regulatory path is equally unremarkable. To offer futures to U.S. participants, a venue must be a designated contract market. A DCM lists new products either through self-certification under Rule 40.2 or through CFTC review under Rule 40.3. Both routes are available only to registered entities. This is the detail that most of the excited commentary missed, and I will return to it, because it quietly reassigns the beneficiaries of this entire campaign.
The substantive question is not whether the contract can be engineered. It can. The question is whether the underlying can be defined precisely enough that a settlement price means something, and whether the resulting market has any force pulling it back toward physical reality. On both counts, compute is a harder asset than anything currently listed on a U.S. exchange, including electricity.
GPU Hours Are Not Fungible
Start with the standardization problem, because it is upstream of everything else.
An H100, an H200, a B200, an A100 — these are not interchangeable units. The same model in different regions carries different power costs and different latency to the customer. The same model under different service levels — bare metal, reserved, spot — prices differently by design, because the vendor is selling different risk. Interconnect bandwidth changes the value of the same chip by a factor that can reach several times for training workloads, because a thousand GPUs on InfiniBand behave nothing like a thousand GPUs on commodity Ethernet. Cooling, uptime guarantees, and jurisdictional restrictions on where the workload may run all move the price.
A benchmark index for compute therefore faces a problem that crude oil solved with a grading system. West Texas Intermediate works because API gravity and sulfur content are measurable, standardized, and contractually bounded; a refiner knows what a barrel of WTI is to within a tolerance. There is no equivalent specification sheet for a GPU hour. To build one, the index compiler would need a quality-adjustment methodology spanning silicon generation, region, service tier, and fabric — and every one of those adjustments is a discretionary judgment that someone can dispute at settlement.
The alternative is to accept large basis risk, meaning the hedge only partially hedges. For the hedger that is a cost. For the speculator it is an opportunity. For the regulator it is a red flag, because an index with wide, judgment-driven adjustment bands is an index with wide, judgment-driven manipulation surface.
I ran into the same structural problem in 2020, in a much simpler setting. I built custom SQL against Ethereum mainnet and analyzed $45 million of Uniswap V2 liquidity flows over four weeks, hunting for a recurring arbitrage inefficiency in stablecoin pairs. The pairs looked homogeneous — a dollar is a dollar — and they were not. Fee tiers, pool depth, and routing latency created persistent micro-basis that the naive model treated as noise. I ended up documenting the geometric decay of impermanent loss instead, because the decay curve was the honest finding and the arbitrage was an artifact of my own aggregation choices. The lesson stuck: when your unit of account is heterogeneous, your index is a model, not a measurement.
Compute is that problem with three more dimensions.
The Commodity That Evaporates
This is the part of the proposal that receives the least attention and deserves the most.
Traditional commodity futures price discovery rests on the cost-of-carry model. Spot and futures are tethered by storage, transport, insurance, and financing. If futures trade too far above spot, someone buys spot, stores it, and delivers. If they trade too far below, someone draws down inventory. The tether is physical. It is why convergence happens, and it is why the funding rate on a perpetual has something to arbitrage against.
A GPU hour is perishable. Unused capacity does not sit in a warehouse waiting for a better price; it expires the instant the clock passes. There is no inventory to carry, no storage cost to model, no delivery pipeline, no short-side physical hedge. When supply outruns demand, the rental rate does not fall to a floor set by storage cost — it can fall to zero, and in principle below zero, with suppliers effectively paying buyers to keep utilization high rather than eating a total loss on idle metal.
Three consequences follow, and each one is a design constraint rather than a philosophical objection.
First, the futures price is driven almost entirely by expectation and positioning, with no physical restoring force. Convergence efficiency collapses. A market with weak convergence is a market where the basis can stay dislocated for months, which means the hedge fails precisely when the hedger needs it.
Second, the funding rate loses its anchor. Funding works because arbitrageurs connect the perpetual to spot. If the spot market is over-the-counter, opaque, and impossible to short, then funding is no longer a tether — it is an isolated variable that a large position can push around. In that regime, funding becomes a lever for extracting value from the counterparty rather than a mechanism for price integrity. Volatility exposes leverage, and a funding rate with no arbitrage backstop is leverage with the safety rails removed.
Third, the contract needs negative-price handling built in from day one. Anyone who watched WTI settle at a negative number in April 2020 understands that a physically anchored commodity can still invert when storage saturates. Compute has no storage to saturate. It has perishability instead, which is worse, because the negative region is not an edge case triggered by a logistical crisis — it is the normal state of oversupply.
None of this makes compute unhedgeable. It makes it a different kind of instrument, closer to a weather derivative than to a barrel of oil: an index-referenced, cash-settled contract on a non-storable variable. Weather derivatives work. They also trade thin, price off modeled rather than observed data, and settle against a third-party calculation nobody can arbitrage.
Outsourcing Integrity to the Index
Measure two is the most contested line in the letter, and it should be.
CFTC practice for listing a new contract has long included an assessment of whether the underlying cash market is not readily susceptible to manipulation. HPC's proposal asks the Commission to permit listing regardless of spot market maturity. Read charitably, this is a pragmatic acknowledgment that no mature compute spot market exists and waiting for one means waiting forever. Read precisely, it asks the regulator to abandon a core principle.
Either way, the technical consequence is unambiguous: contract integrity gets outsourced, in full, to whoever compiles the index.
That is not a small transfer of trust. In a physically settled commodity, the exchange's integrity rests on warehouses, assayers, and delivery logistics — multiple independent parties with their own incentives and their own audit trails. In a cash-settled index contract with no spot anchor, the exchange's integrity rests on one number produced by one methodology maintained by one firm. The clearinghouse is still there. The margin engine is still there. But the thing being margined is a calculation.
And the calculation is not audited to any standard I can find described in the proposal. There is no mention of an independent methodology review, no disclosure requirement for the index compiler's own positions, no fallback procedure for a data gap. In my own audit work I treat an unaudited pricing methodology as equivalent in risk to unaudited code, because the failure mode is identical: the system behaves correctly according to rules that were never checked against reality until the moment they matter.
Code is law; math is evidence. An index without a published, reproducible methodology is neither.
Competing Benchmarks and the Fragmentation Tax
Measure three — allowing multiple qualifying indices to settle in parallel — is the smartest technical idea in the letter, and it comes with a bill.
The logic is sound. If one index can be pushed, concentrate the attack surface and the payoff is large; require competing benchmarks and the marginal return to manipulation drops, because a manipulator would have to move several independent calculations at once. It is a de-single-point-of-failure design, and in isolation I would endorse it.
The bill arrives at the exchange level. If two venues list what is nominally the same product against different benchmarks, the contracts are not the same contract. Spreads open between them, cross-venue arbitrage breaks down because there is no shared settlement value to converge on, and liquidity fragments across venues that should be pooling it.
The historical precedent is instructive. U.S. natural gas futures built the deepest pool of liquidity in the world on a single benchmark at Henry Hub. Europe's early liberalization ran multiple hubs — TTF, NBP, and others — and spent years with liquidity scattered across them before consolidation. Fragmentation is not fatal. It is a tax, paid in wider spreads and thinner books, and it is paid by the participants who most need deep liquidity: the hedgers.
There is a second-order effect that matters more. The first index to achieve genuine volume becomes the de facto standard, and the compiler of that index acquires something close to price-setting authority over an asset class. That is a durable commercial moat built out of a regulatory accommodation. Anyone modeling the long-run value of index infrastructure should be pricing that outcome, not the contract fees.
Data Integrity Checks
Mandatory disclosure, in the format I use for every forensic piece.
Source provenance. The underlying report is a single-source summary of a single advocacy submission. Nine of ten information points originate with HPC. One is the submission event. There is no independent corroboration of any substantive claim.
Access limitation. I have not read the letter itself. Claims about what the letter argues are second-hand and should be treated as such.
Temporal inconsistency. The October 22 date conflicts with a reference to a 2026 policy statement. This is not a trivial typo. If the letter is prospective, its market relevance is a research note. If it responds to an actual 2026 statement, the relevance is materially higher and the timeline for DCM filings compresses.
Quantitative gaps. No TVL, open interest, funding rate, or valuation multiple has been verified. Every such figure is absent rather than estimated. Readers should treat any number they see attached to this story elsewhere as unverified until a primary source produces it.
Conflict of interest. The submitting organization's name aligns with a major on-chain perpetuals venue. Its funding structure is not disclosed. In regulatory practice, that relationship is exactly the kind of interest that gets flagged during review of a comment letter, and it is the kind of interest that should be flagged by analysts too.
Model bias. My prior is that cash-settled derivatives on non-storable underlyings trade thin and price off models. That prior shapes my read of this proposal, and I am stating it rather than hiding it.
Who Actually Gets the License
The most useful sentence in this entire story is buried in the regulatory mechanics, and almost nobody discussing it has picked it up.
Listing a futures contract in the United States requires a designated contract market. Rule 40.2 self-certification and Rule 40.3 review are available only to registered entities. A venue that does not operate within that framework — that does not run a full know-your-customer program, that is not a registered DCM — cannot list the product, no matter how much advocacy its affiliated organization produces.
Which means the direct beneficiaries of a successful campaign are the incumbent licensed venues: the CMEs, the Coinbase Derivatives, the Kraken-adjacent and Bitnomial and Kalshi-type regulated marketplaces. They already hold the licenses. They already have the compliance infrastructure. They already have the institutional distribution. If compute perpetuals get approved, they list first, and they capture the flow.
The advocacy is not opening a door for itself. It is opening a category, and the category will be harvested by whoever holds the paper.
This is not cynicism; it is a structural read. In 2024 I analyzed daily inflow and outflow data from eleven spot Bitcoin ETF issuers against price action over six months and quantified a 0.85 correlation between institutional net inflows and price stability. The finding that mattered was not the correlation. It was that the anchor came from regulated wrappers with balance sheets behind them, not from the venues that had been trading the asset for a decade. Regulatory structure determines who captures the flow, and it usually is not the pioneer.
Narrative Versus Cash Flow
Here is the timing mismatch that will determine whether anyone makes money on this story, as distinct from writing about it.
If the CFTC adopted all six measures tomorrow, the path to a live, liquid contract runs through a policy statement, a DCM filing, product approval, market-maker onboarding, and a liquidity ramp. Realistically that is twelve to thirty-six months. Markets of this kind do not become deep on launch; they become deep when the hedgers arrive, and hedgers arrive when the basis is tight enough to be worth the operational cost.
The narrative, by contrast, prices in days. 'U.S. regulator recognizes compute as a commodity' is a sentence that can move a sector before the Commission has scheduled a meeting. My 2021 NFT work is the cleanest illustration I have: I processed 150,000 individual BAYC and CryptoPunks trade records and found that whale accumulation preceded floor price spikes by exactly seventy-two hours. The information existed on-chain, publicly, in real time. It was priced anyway. Narratives do not wait for confirmation, and they do not wait for cash flow either.
So the honest valuation framework separates two clocks. The narrative clock is short and volatile and entirely reflexive. The cash-flow clock is long and depends on approvals that have not been requested. Buying the narrative is a trade on attention. Buying the cash flow is a trade on a regulatory calendar that does not yet exist.
Anyone conflating the two is going to be confused about why the token went up on a letter and then gave it all back over the following quarter.
The Inference Token Is the Dangerous Part
Of the two sale formats HPC proposes — GPU hours and inference tokens — the second one is where this stops being a market-structure story and becomes a securities story.
A GPU hour is a service. It is consumed, it expires, it has no residual claim. Classifying it as a commodity is legally undramatic, and I will explain why shortly.
An inference token is different. If it is defined as a prepaid unit of model output purchased in advance, it is functionally a prepaid service voucher that can be made fungible, transferable, and settleable. That is a well-worn path. Cloud mining contracts, storage-credential derivatives, and prepaid compute vouchers have all at various points been assessed as investment contracts when sold with an expectation of appreciation derived from someone else's managerial effort. Tokenizing a prepaid voucher does not change its economic substance; it changes its distribution channel.
Note also the rhetorical choice embedded in the proposal. The underlying is defined as compute — a service, a good, a commodity — rather than as a token. That is not an accident of drafting. It is a deliberate routing decision that steers the entire question toward the CFTC's commodity jurisdiction and away from the SEC's investment-contract analysis. Whether a CFTC commodity determination precludes an SEC action is a separate question, and the answer is no. Jurisdictions overlap. Precedent is not a shield.
If an inference token launches on the back of this advocacy, the first question any serious analyst should ask is not what the index says. It is who sold it, to whom, and what they were promised.
The Mirror Effect on DePIN
There is a consequence of this proposal that the compute-token sector will not enjoy, and it is worth stating before the marketing departments get there.
DePIN compute networks currently price their tokens against a reference set that is opaque. Real GPU rental rates are negotiated privately, quoted with volume discounts, and reported inconsistently. Token prices, meanwhile, are public and continuous. The gap between what a network's token implies about the value of its compute and what that compute actually rents for is not measured, because there is nothing to measure it against.
A credible, published, tradeable compute benchmark changes that permanently. Once a public reference price exists, every token in the category acquires a denominator. The divergence between implied and realized value stops being an unknown and becomes a chart. In the short run that will be packaged as price discovery and celebrated. In the medium run it is a stress test, and the networks whose tokens trade at a large multiple to their realized compute revenue will fail it visibly.
I am not predicting which ones. I am saying that the instrument this proposal creates is the instrument that will be used to grade them, and that the grading will be done by people who did not consent to being graded.
The Demand Side Is Real
The bear case writes itself. The bull case is narrower than the thread posters think, and it lives in one specific place.
Bitcoin miners that have pivoted into AI and HPC hosting have converted their balance sheets from self-mining into long-duration GPU lease contracts. Their revenue is now a function of rental rates over multi-year horizons. Their cost base is power, which they hedge; their capital base is metal, which they depreciate; and their revenue — the actual number that determines whether they service debt — is entirely unhedged.
That is textbook hedging demand. A miner with a three-year lease book and floating rental exposure needs a way to lock a floor. Today the only tool is a bilateral OTC forward with a counterparty who charges for the privilege. A listed, cleared, cash-settled compute contract would be a genuine operating risk management instrument for these firms, not a speculation.
Follow the gas. Always. The companies closest to the energy input are the ones whose economics are most exposed to the price of the output, and they are the ones who will show up as natural buyers of protection on day one. That is the fundamental support for this product existing at all. Everything else is positioning.
The Legal Question Is Not the One Being Argued
The proposal leads with the claim that compute qualifies as a commodity. That is the least controversial sentence in the document, and its prominence is a tell.
The Commodity Exchange Act defines a commodity with extraordinary breadth — goods, articles, services, rights, and interests in which contracts for future delivery are presently or in the future dealt in. The CFTC has asserted jurisdiction over Bitcoin, over ether, and historically over considerably stranger things, including box office receipts and temperature and whether it will rain. Compute, as a service sold forward, sits comfortably inside that definition. It is very close to settled law, and anyone treating it as the fight is arguing with a straw man the advocacy itself constructed.
Leading with the unarguable point is a standard rhetorical structure. Establish consensus on the easy claim, then carry the audience into the contested ones on the momentum. It works because readers do not distinguish between the parts they have been persuaded of and the parts they have merely been walked past.
The real fights are elsewhere, and there are three of them.
The manipulation-susceptibility requirement is the first, and it is the one measure two asks the Commission to waive. That is a request to abandon a principle, not to reinterpret it, and such requests have a low pass rate.
The second is position limits and concentration. Measure four concedes that early markets will attract manipulation and concentration, and proposes enhanced surveillance plus position limits as mitigation. That is substituting after-the-fact monitoring for before-the-fact depth. Regulators have consistently been reluctant to accept that substitution, because surveillance detects manipulation after the loss has been taken.
The third is the inference token's jurisdictional home. A CFTC commodity determination does not immunize an instrument from securities analysis. If the token is sold with appreciation expectations tied to an issuer's efforts, the analysis does not care what the Commission called the underlying service.
When Both Sides of the Book Are Bots
One more structural consideration, and it is the one I think will look obvious in hindsight.
A compute perpetual would launch into a market where an increasing share of volume is generated by automated agents. In my 2026 work on wallet clustering I analyzed one million transaction tags and found that roughly fifteen percent of what looked like organic trading volume was generated by coordinated AI agents, distorting the liquidity metrics that most participants use to size positions. That was a market with human participants and bot overlays.
A compute market inverts the ratio. The natural participants are GPU cloud providers running automated pricing engines, AI model developers hedging inference costs, and market-making bots optimizing funding capture. Humans will be a rounding error. The settlement index will be compiled by a firm whose own positions are unknown. The price will be discovered by machines trading against machines, tethered to a reference price that has no physical arbitrage anchor.
That is not a reason to forbid the market. It is a reason to demand three things before it exists: a published index methodology with an independent audit trail, disclosure of the compiler's economic interests, and surveillance infrastructure designed for agent-to-agent trading rather than retrofitted from a human order book. Without those, the market will function smoothly, clear correctly, and price something nobody can verify.
Correlation Is Not a Catalyst
The reflexive mistake in reading this story is to treat a comment letter as a market event. It is not. It is a document.
Nothing in the record shows the Commission has responded, scheduled a meeting, or signaled interest. The pricing of this news is therefore close to zero, and should be. The genuine volatility triggers sit several steps away: a formal CFTC response, a policy statement, or the first Rule 40.2 or 40.3 filing from a registered venue. Until one of those appears, the correct posture is a watch list, not a position.
And there is a subtler error available, which is to assume that because the advocacy comes from an organization aligned with a major on-chain venue, that venue is the beneficiary. It is not, for reasons that are purely mechanical: it is not a DCM. The people who will list this product are the people who already hold the licenses. The campaign may succeed and the sponsor may still capture almost none of the resulting flow.
Approval Is the Risk
Here is the contrarian read, and it is the one I would defend in a room full of people who want this to happen.
The dangerous outcome is not rejection. Rejection costs a press cycle and a few threads. The dangerous outcome is approval of a cash-settled contract on a perishable, non-storable underlying, settled against an index compiled by a party with a position in the outcome, in a market where the spot leg is opaque and cannot be shorted.
In that configuration, the funding rate stops being a tether and becomes a tool. The basis stops being a hedge and becomes a bet. The clearinghouse does its job flawlessly while the thing it clears drifts away from physical reality, because there was never anything physical to drift from.
I have watched this movie once. In 2022 I traced 50,000 wallets and $2.3 billion of outflows during the Terra collapse, and the mechanism was not a villain. It was a design that worked exactly as specified while the specification was wrong. The autopsy mattered more than the blame, because the specification error was the transferable lesson.
Compute is not Terra. But the failure mode — a system whose internal consistency is perfect and whose external anchor is missing — is the same shape, and it is the shape regulators are supposed to catch before the fact rather than after.
Takeaway
Watch four things, in this order. Any Rule 40.2 or 40.3 filing from a registered venue. A published index methodology with an independent audit trail and a conflict-of-interest disclosure. A formal CFTC response, which would convert a research note into a calendar. And the first funding-rate print on the first listing, which will tell you whether the tether exists or whether it is decoration.
If a GPU hour cannot be stored, cannot be delivered, and cannot be shorted, then what exactly settles when the contract expires — and who audits the answer? That is the only question that matters here, and the letter does not answer it.
Follow the gas. Always.
