A simple way to contract, finance, and allocate future compute.
Compute has quietly become one of the most important inputs in the modern world. It underpins artificial intelligence, scientific research, finance, logistics, media, education, and increasingly, national infrastructure. We still talk about it casually as “cloud” or “instances,” but underneath that language is something very real: data centers, power systems, cooling, networking, and highly specialized hardware.
For a long time, the dominant assumption was that compute would always be available. If you needed more, you spun it up. If demand increased, supply would catch up. That assumption is less reliable for the capacity that matters most. High-performance GPUs and AI-ready data center capacity are constrained by physical limits, supply chains, power availability, construction timelines, and geopolitics. When compute becomes scarce, access matters just as much as price.
As soon as access matters, the way compute is financed and allocated starts to matter too.
Compute Reality
Most compute today is sold in one of two ways. Either it is fully on-demand, where availability is assumed until it suddenly is not, or it is negotiated through bespoke enterprise agreements that take time, lawyers, and repeated reinvention. Both approaches work for many workloads. Neither is ideal when a customer needs a defined future claim on strategic capacity.
Other infrastructure sectors solved versions of this problem long ago. Power plants are built because someone commits to buying power in the future. Factories are financed because future production can be pre-sold. Aircraft are ordered years in advance because airlines make commitments before the planes exist. These agreements make future demand concrete enough to plan and finance supply.
Compute is increasingly in that category, but the contract structures around it are still developing.
Infrastructure Logic
A Simple Agreement for Future Compute, or SAFC, is an attempt to close that gap with a minimal, legible contract primitive.
At its core, a SAFC is straightforward. One party commits to a defined future compute requirement. ACF or another supplier agrees the entitlement, delivery window, commercial terms, dependencies, and operating responsibilities. That commitment can help capacity be planned, financed, allocated, and delivered with less ambiguity.
The simplicity is intentional. The goal is not to invent a financial trick. It is to reduce friction between people who need compute and the physical reality of supplying it.
A SAFC is designed to make future demand concrete enough that capacity can be planned, financed, allocated, and delivered without turning every transaction into a bespoke negotiation.
How It Works
- Define the requirement. Specify the workload, hardware class or acceptable equivalent, scale, geography, start window, duration, and any operating constraints.
- Agree the claim. The parties document the future capacity entitlement, pricing or pricing method, delivery conditions, dependencies, remedies, and responsibilities.
- Use present capacity where practical. Existing ACF-managed or qualified partner capacity can support integration and early workloads when it is available and appropriate.
- Use the commitment as demand evidence. A defined future requirement can inform procurement, site planning, equipment financing, project finance, and the timing of new infrastructure.
- Activate the entitlement. When the agreed conditions are met, the capacity is made available under the terms of the agreement and usage can be metered and reported through the operating layer.
Intended Use
The first group a SAFC is designed for are companies, teams, and institutions that genuinely need compute to operate. These include AI companies deploying models, researchers training systems, enterprises integrating AI into core workflows, universities running large experiments, and governments operating sensitive or sovereign workloads. For them, the value is operational: predictable access, clearer planning, and fewer surprises.
A SAFC gives them a way to secure future capacity while still being able to start integration through available capacity where practical.
The second group are organizations that treat compute as a strategic resource. These may include larger companies, research networks, venture firms, family offices, accelerators, or other institutions with multiple internal teams or portfolio companies whose needs shift over time. For them, a SAFC can function as an allocation right, allowing contracted compute to be directed to eligible workloads under agreed terms.
This is not equity and it is not a promise of financial return. In a separate Compute for Equity transaction, a company may pair compute access with equity or another negotiated form of consideration. The SAFC itself remains a compute agreement, not an equity instrument.
Time Alignment
One practical reality matters here. Infrastructure takes time to build, but teams often need to begin integration before new capacity is commissioned.
A SAFC can bridge those timelines. Where existing capacity is available, teams can integrate APIs, test workloads, validate performance, and build operational habits while longer-term capacity is prepared. As new capacity becomes available, delivery can transition into the longer-term structure defined by the agreement rather than forcing the customer to restart the relationship from zero.
This bridge between present access and future capacity is central to the idea. It acknowledges how systems are actually built and adopted without pretending that unbuilt capacity is already live.
Scope Limits
A SAFC is not a token, not a speculative vehicle, not equity, and not a promise about future prices or guaranteed availability before its conditions are satisfied. Compute markets evolve over time. Hardware improves, supply expands, power and site conditions change, and prices can move in different directions.
A SAFC does not eliminate uncertainty. It makes a defined future claim legible: what is being requested, what is being promised, what must happen before delivery, and what happens if agreed obligations are not met.
Clear definitions, shared expectations around delivery, and well-understood remedies matter more than complexity.
Open Standard
This idea is shared as research because markets tend to scale more smoothly when they have common primitives. In the absence of shared structures, similar agreements are repeatedly reinvented, adding friction and slowing progress. Publishing a Simple Agreement for Future Compute offers a reference point that customers, infrastructure builders, financiers, and other participants can read, question, and adapt.
This post is only an introduction. The more detailed structure, definitions, and design considerations are laid out in the accompanying white paper and template.
If the idea proves useful, it should evolve through practical application and better evidence. If not, it remains a documented attempt to clarify a problem that is likely to become more important as compute becomes more strategic.
You can read the SAFC materials below.