About
Touchmark builds an AI inference marketplace that scores model outputs on quality, efficiency, and value, then converts those scores into platform pricing. It sells to AI platforms and model providers, differentiating through quality-adjusted pricing and a live order book for reserved inference capacity rather than token-based billing.
Market
Touchmark competes in AI inference infrastructure and compute procurement, combining model-serving access with an exchange-like market for future inference capacity. Unlike direct providers such as Fireworks AI, Together AI, Groq, and Replicate, or routers such as OpenRouter, Touchmark lets buyers lock in posted prices for future delivery, resell unused commitments, and lets providers sell capacity forward to finance GPU buildout. Its differentiation is the two-sided order book, forward pricing, and planned benchmark- or quality-settled contracts rather than ordinary pay-as-you-go token pricing.
Touchmark serves AI application companies—from startups to larger engineering organizations—with predictable, high-volume model-inference needs and infrastructure or procurement teams seeking lower, more predictable future capacity costs. Its other customer segment is model-serving and GPU-infrastructure providers that want to sell capacity forward, lock in revenue, and plan GPU spending against committed demand.
At a Glance
Problem
Touchmark is addressing two linked inefficiencies in AI infrastructure. AI inference capacity is typically bought through private, inflexible negotiations: companies commit ahead of time to obtain discounts, but remain liable for unused capacity if demand changes. Providers face the opposite problem: they must finance GPU buildouts years ahead while revenue arrives gradually, without a transparent view of forward demand. At the application layer, token pricing also charges customers whether an output is useful or not, making AI spend difficult to explain and vulnerable to waste from poor model selection, runaway agents, and bloated prompts.
The clearest killer use case is a company that wants the economics of a forward capacity commitment without being trapped by it. It can reserve a discounted block of future model inference, then resell the unused commitment if plans change. Touchmark’s broader thesis is that AI should ultimately be priced by delivered capability or quality rather than raw token volume.
Product / Service
Touchmark presents a two-sided marketplace for reserved AI model-inference capacity. Buyers fund a balance, select a model and delivery month, and purchase a fixed block at the live best ask on a two-sided order book. The block is accessed through an OpenAI-compatible key and delivered through the provider’s endpoints; buyers receive deeper discounts for committing further ahead and can offload the contract before it is exhausted. Providers set their price and floor, sell capacity forward, and use visible orders to plan GPU spending and lock in revenue.
The company also describes a quality-based pricing layer delivered through an SDK. Outputs can be evaluated against common benchmarks and product-specific graders, with customer charges adjusted to the quality of each result; prepaid funds are held in escrow and released only when deliveries clear the agreed quality bar. In the company’s examples, high-quality output attracts a surcharge, average output remains at the reference price, and low-quality output receives a refund, shifting the economic unit from model tokens toward verified work.
Market
Touchmark sits in AI infrastructure and the expanding AI inference market, with a differentiated focus on financial-market-style trading of future inference capacity and quality-adjusted AI output. It is not simply another model-hosting provider: its proposed market mechanism connects buyers seeking discounted, flexible capacity with providers seeking forward demand and financing visibility. Adjacent alternatives include GPU marketplaces such as Vast.ai, reserved-GPU and forward-contract marketplaces such as Compute Exchange, and conventional inference platforms including Baseten, Fireworks AI, FriendliAI, Modular, DeepInfra, and Novita AI. The evidence does not identify a direct competitor offering the same combination of a secondary market for model-inference contracts and output-quality pricing.
Touchmark appears to be an extremely early-stage company rather than a scaled commercial operator. Y Combinator lists it as an active Summer 2026 company in artificial-intelligence infrastructure, with two employees in San Francisco, while the company site shows a signup path and illustrative order-book contracts for future delivery months. The reviewed evidence does not disclose customers, revenue, transaction volume, or a specific funding amount, so the strongest supported conclusion is that it is in early product or market-launch stage and that commercial traction has not yet been publicly demonstrated.
Founders & Leadership
Funding History
Y Combinator
Recent News
Touchmark appears in a Y Combinator Summer 2026 batch-company directory under touchmark.ai. The listing identifies the company as an AI startup focused on solving AI pricing.
A third-party company profile describes Touchmark as pricing AI products by evaluating output quality rather than counting tokens and reports that the company is backed by Y Combinator. No specific funding amount is stated in the available coverage.
Y Combinator’s company profile presents Touchmark as an active San Francisco AI-infrastructure startup in the Summer 2026 batch. It lists co-founders Ilia Bolgov and Roman Yanushevskyi and describes the company’s focus as solving AI pricing.
Touchmark’s official product page describes a live order book for reserved AI-model capacity, allowing buyers to purchase discounted inference capacity and providers to sell capacity forward against visible demand. The platform also supports reselling contracts before capacity is consumed.
Active Roles
0No active roles right now.
Get notified when they postBusiness Model
Touchmark monetizes AI inference through quality-adjusted pricing: platforms are billed based on the evaluated value of model outputs rather than token volume. Its marketplace also uses live order-book pricing for reserved capacity, connecting buyers seeking discounts with providers selling future capacity.