This is designed for teams with variable or spiky compute demands that aren't well served by spot/on-demand market (price/availability uncertainty) or longer duration reserved capacity commitments (prone to underutilization). Examples could include inference providers that may have variable needs around new model launches, enterprises that are working on fine-tunes or eval runs, or any teams that deal with batch jobs or seasonal spikes. Right now these teams either overcommit to reservations that may be underutilized, or take on availability and price risk on the open market. The call sits in the middle of those two options. Today, you can pick a GPU, size, window, and strike and it quotes a fee off per-SKU rental indices (Ornn OCPI), a Bachelier option value, a basis adjustment for the provider you'd actually rent from, and a disclosed spread. Needless to say that two clusters of the same GPU are not the same product. Interconnect, memory per card, storage and networking, and region all influence the real number, so treat the price on the page as a basic starting point. Quotes today are indicative, and a real quote will reflect the actual spec plus each provider's own rate card and basis. Full methodology is linked on the site. This is early and the aim is to aggregate the bid side of the RFQ to earn some pricing power against a network of Neoclouds. In the fullness of time, this ceases to operate like a desk/broker style service and buyers/providers can match bilaterally. Again, note that today the quotes on the page are indicative, each provider will have its own basis that will need to be priced in along its rate card. Very curious for your thoughts.

Plainview

Capacity call options on compute.

Capacity call options on compute.