Touchmark

Companies that need guaranteed AI inference capacity currently buy it from individual providers at retail rates, without a way to compare bids or lock in performance guarantees like uptime, time-to-first-token, or throughput speed. Demand for inference also fluctuates, with baseline usage and unpredictable peaks, but buyers have no standardized way to cover peak demand separately from steady-state usage. Capacity purchased this way is typically locked in, non-transferable, and does not expire gracefully, leaving buyers overpaying or unable to resell unused capacity.

Touchmark created the 'inference contract,' a tradable instrument specifying a model, a volume of tokens (or an hourly throughput rate), weighted pricing for output, input, and cached-input tokens, explicit service levels (availability, time-to-first-token, tokens-per-second speed), and a duration window after which unused tokens expire. Buyers open a request for quotes and multiple providers bid against the same order, with Touchmark stating trades typically clear 10-60% below retail and that buyers pay no transaction fees. Touchmark meters usage and collects payment on behalf of providers. Buyers can cover demand in two ways: reserving a baseline through an RFQ with custom volume and terms, or reserving peak capacity by purchasing throughput up to three hours ahead of demand under the same SLA protections. Purchased capacity is also resellable to other buyers rather than being locked in.

Touchmark's stated vision is that AI inference has become a commodity but lacks matching financial infrastructure, which it positions itself to build, starting with capacity contracts.