The EFP Seam.

Callback

I put money into the physical rebuild of the economy, energy, compute, robotics, the materials under all of it. So I have been watching the compute-futures market form with more than passing interest: it is paper being written on top of the physical supply I help fund. I also chair a company building in the same terrain. From both seats one pattern keeps surfacing, a market goes machine-speed years before anyone builds the record of who authorized the machines. Compute is the sharpest version I have seen yet.

In #018 I described a three-leg trade: hedge a compute position with a futures contract, convert that paper position into actual GPU capacity via an Exchange-for-Physical transaction, settle on-chain. The question I left was: who authorized that?

I said the EFP moment, the hand-off from paper to physical, was the highest-value governance surface in the stack. Turns out someone else read the same setup.

On approximately July 7, Architect Financial Technologies and Compute Desk announced ComputeConnect: the US financial industry's first compute EFP network, linking CFTC-regulated compute derivatives to physical GPU capacity delivery. Not a thesis. An announced piece of infrastructure.

The seam has a name now. That makes the governance question harder to defer.

The Mechanics

EFP, Exchange-for-Physical, is one of the older mechanisms in commodity markets. The logic is simple: a party holding a cash-settled futures position and a party holding the underlying physical commodity privately negotiate a swap. The futures leg closes. The physical leg opens. The exchange does not execute the transaction, but it receives the report. The trade is off-market, bilateral, and, in any regulated commodity, fully subject to the exchange's position-limit and reporting rules at the moment of execution.

In energy markets, EFPs are routine. A natural gas producer running a short hedge on NYMEX and a utility running a long physical purchase privately agree to net them. The paper positions disappear. The physical delivery obligation is established. The CFTC's Part 38 rules govern what gets reported, when, and to whom.

The compute version works the same way in principle. A firm hedging forward GPU spend holds a cash-settled position on, say, the Ornn OCPI-H100 contract or the NATIVX COIL index. At some point, that hedge needs to convert: the firm does not want cash settlement, it wants actual H100 hours. An EFP transaction pairs the futures close with a matched spot-capacity purchase from a provider in the ComputeConnect network. The futures leg reports to the American Innovation Exchange. The delivery obligation is now against a real counterparty with a real SLA.

That is the seam. It looks like a settlement mechanism. It is actually a governance event.

Why the Seam Carries the Load

Four things happen simultaneously at the EFP seam that happen nowhere else in the trade:

Counterparty risk materializes. Before the EFP, the position is marked-to-market against an index. Default risk sits with the clearinghouse. After the EFP, there is a named physical counterparty on the other side of a delivery obligation. If CoreWeave, a neocloud (Fluidstack or General Compute), or any capacity provider in the ComputeConnect network fails to deliver committed GPU hours, that is not a mark-to-market loss, it is a delivery default. The credit profile of the physical counterparty becomes load-bearing at the moment the EFP executes.

Scope is consumed. An agent operating inside a delegated mandate, "hedge up to X notional in compute futures, convert to spot capacity within Y window, against cleared counterparties", uses up scope at the EFP. Before that moment, the mandate is intact. After it, a specific counterparty relationship has been established, a delivery date has been set, a capacity provider has committed. The scope is no longer potential. It is spent. Whether the consumption matched the mandate, in size, counterparty, timing, location, is a question that must be answerable after the fact.

Position-limit and reporting obligations attach. CFTC-regulated EFPs in traditional commodities generate a reportable event. The CME and ICE filings, when they clear, will define what position-limit rules apply to compute futures. But the EFP is the moment when a cash-settled instrument begins to function as a physical-delivery commitment. That is the moment regulators in every other commodity market treat as highest-risk for manipulation, wash trading, and unauthorized excess. The architecture will import that scrutiny.

The authorization question becomes unanswerable without a prior record. Who told the agent to execute the EFP? At what size? Against which counterparty? Within what tolerance on basis? The futures leg had a timestamp and an exchange record. The EFP is negotiated off-market. Without a separately maintained record of the mandate that authorized the physical conversion, pre-execution, signed by the principal, scope-bounded, the answer to "was this trade authorized?" is reconstructed after the fact from logs, email threads, and oral testimony.

That is fine for a human trader who made a phone call and left a voicemail. It is a structural gap for an agent that executed in milliseconds against parameters set six weeks ago by a committee that has since turned over.

The Stack as It Stands

Computeconnect is DESIGNED infrastructure against a pending exchange. The American Innovation Exchange is waiting on CFTC's DCM designation. CME x Silicon Data is pending regulatory review since May. ICE has two compute futures products announced (NATIVX COIL on July 1, Ornn index on May 19), both pending regulatory approval. None of these contracts are live.

But the scaffolding is moving faster than the governance layer.

VanEck estimates a $50B capital shortfall against the $70B in compute supply miners have already contracted. The capital to fund physical delivery is behind the paper commitments. When regulated compute futures go live and EFP becomes a standard mechanism for converting hedges into actual capacity, that gap becomes acute. Lenders and prime brokers who underwrite the physical delivery leg will want proof of authorization at the EFP moment, not retrospective documentation, but pre-existing, verifiable mandate scope.

The Analyst Read

I have watched this sequence run before, in other markets, and in companies I have funded. Every commodity market that has matured from paper trading to physical delivery has built the same governance layer at the EFP seam. In metals, it is the authorized-dealer regime. In energy, it is the Part 38 EFP reporting structure. In agricultural commodities, it is position-limit certification for delivery-eligible accounts. The layer is never the exciting part of the market. It is always the part that ends up mattering.

Compute will import this structure. The only open question is whether it arrives from the regulator or from the counterparty network itself.

In traditional markets, the governance layer was built reactively, after a delivery default, after a manipulation investigation, after a position-limit violation surfaced in an audit. Compute markets have the unusual advantage of watching the seam form before the first default.

The infrastructure that closes the seam at the EFP moment is portable, pre-execution mandate scope: a signed record that says this agent, acting under this authority, is cleared to execute a physical-delivery EFP against these counterparties, at this notional, in this window. That record does not live inside the exchange, no commodity exchange maintains principal mandates for its participants' automated agents. It travels with the agent, and it presents at the moment of execution.

That is what governs the seam. Not the clearinghouse. Not the delivery network. The mandate record that existed before the trade.

I have a position on how this resolves, a literal one. Accords, where I'm Executive Chairman, is building portable, pre-execution mandate scope for asset and capacity markets: the record described above. Disclosure: Accords is a portfolio position I chair. Its compute-market instrument is DESIGNED, not live, not a shipped product. But the seam is the same hand-off problem it already treats elsewhere, and it demands the same structure.

Falsifiable Close

Here is what breaks this.

If the American Innovation Exchange does not clear CFTC DCM designation, and if CME x Silicon Data and both ICE compute futures products similarly stall past Q4 2026, then compute EFP remains theoretical. No regulated EFP market, no regulatory scrutiny at the seam, no institutional pressure to build a pre-execution mandate layer. The governance gap exists but generates no friction that compels a solution.

The second failure mode is simpler: counterparty networks like ComputeConnect define their own authorization model, baking it into the onboarding flow for capacity providers. That closes the EFP seam mechanically, but closes it inside a single proprietary network. The portable, cross-platform mandate record does not get built. The governance layer becomes platform-specific, the same fragmentation problem that afflicts every silo-first approach to regulated infrastructure.

Watch: CFTC's clock on the American Innovation Exchange DCM application (the one ComputeConnect routes to). And whether ICE lists either compute product, COIL or Ornn-indexed, with real market-maker participation before year-end. If one regulated compute futures contract goes live with an EFP mechanism attached, the seam goes from architectural thesis to active compliance surface.

Max

Max Curnin has spent 11 years and 45+ investments on the physical rebuild of the economy, energy, compute, robotics, and the materials underneath. He is Executive Chairman at Accords, and writes The Agent Economy on the category he invests in and builds in.