A commitment with room to change

A team reserves compute for a training run. The work finishes early, the next model needs a different configuration, or demand shifts toward inference. The team still has a valuable resource: the remaining capacity it has paid for or committed to buy.

A transferable contract gives that capacity a path to another user. The first buyer can recover value through a sale, and the next buyer can acquire capacity for the period it needs. This matters for a cluster already running as much as for a reservation at a facility still being built.

The beginnings are already visible

AWS permits eligible Standard EC2 Reserved Instances to be resold through its Reserved Instance Marketplace; Convertible Reserved Instances cannot be sold there. This provides a defined resale mechanism for a particular class of AWS commitment.

SF Compute brings resale into GPU procurement. It describes a model in which customers can sublease capacity on infrastructure it operates, including clusters it operates on behalf of other owners. Its published tooling supports buying capacity, selling unused allocations, and market-making strategies.

These models make the value of flexibility tangible. Tensors is building toward a provider-neutral market that extends the idea across qualifying supply, connecting capacity available now with commitments for future delivery.

Make the position fit the workload

With transferable contracts, procurement becomes a position a buyer can manage over time. A team could reserve a larger block to secure availability, then offer unused portions to other buyers where the contract permits subdivision. Another could sell an existing allocation and buy a different performance class or delivery window as its workload changes.

The contract needs to make these actions straightforward: what quantity and time period can move, who may receive it, and which payment obligations move with it. A defined process for consent, access, and release of the original buyer lets the next holder understand exactly what it is buying.

Give providers flexibility too

Providers can use the same market to manage the capacity behind their commitments. An operator with spare supply could sell it to another provider that needs qualifying compute to meet a delivery obligation. A contract that allows substitution creates a reason for operators to trade with one another as well as with end users.

Existing capacity then serves two purposes: it can run a buyer’s workload directly, or it can help fulfill a commitment made elsewhere in the market. A common specification makes that second use possible without asking the holder to accept a different service.

Build a market around changing needs

Buyers and providers bring different schedules to the market. A specialist market maker could bridge them by quoting both sides and holding capacity between transactions, earning a spread for that service. Published bids and offers would give participants a clearer view of the price of changing a position.

The value of transferability begins with the freedom to make another decision. A buyer that sees a credible resale path has more reason to reserve ahead. A provider can compete on flexible terms as well as price. Capacity already in service becomes easier to allocate, and capacity still to come gains a broader set of potential buyers.

Sources

Continue the conversation

Tensors is building the infrastructure for compute markets.

Talk to us