Spotlight

What if AI compute had its own stock exchange?

Quick Pitch: Touchmark is building a marketplace for buying and selling reserved AI computing capacity, with plans to make it tradable like a commodity.

The Problem

• Opaque Pricing: AI companies negotiate computing capacity privately, with little visibility into what others pay.

• Locked Capacity: Buyers commit to future usage but have limited options to resell unused capacity.

• Changing Economics: Models and prices evolve quickly, making long term commitments potentially less attractive.

Why It Matters

AI companies are spending more on running models, with the global inference market projected to reach $255 billion by 2030.

Yet much of this capacity is negotiated privately, with limited price transparency or resale options.

As inference spending grows, businesses will increasingly need predictable costs, transparent pricing, and flexibility in managing capacity.

Snapshot

• Industry: AI Infrastructure / Compute Marketplace

• Headquarters: San Francisco, CA

• Founded: 2026

• Backed By: Y Combinator (S26), Pareto, Inception

• Funding: Raising $4M (Seed)

• Traction: Over 4 trillion tokens requested through RFQs, according to the company

Founder’s Edge

• Ilia Bolgov: Formerly worked on Wealth and Trading products at Revolut. Studied Mathematics and Finance at Imperial College London.

• Roman Yanushevskyi: Former quantitative researcher at Citadel Securities and early AI engineer at Lovable. Informatics Olympiad gold medalist.

Playing Field

• Inference Providers: Fireworks AI and Together AI sell compute directly, without independent price discovery.

• Cloud Platforms: AWS, Azure, and Google Cloud offer capacity within their own ecosystems.

• Routing Platforms: OpenRouter aggregates model access but lacks future contract trading.

Analysis

Bulls Case 📈 

• Early buyer interest suggests demand beyond listed supply.

• Founders combine quantitative trading expertise with firsthand inference buying experience.

• Providers gain access to more buyers and visibility into future demand.

• Future contract trading and hedging could expand revenue beyond brokerage fees.

Bears Case 📉 

• Differences in models and performance make inference difficult to standardize.

• Large cloud providers could build competing marketplaces.

• Exchanges need sufficient buyers and sellers to sustain trading.

• Repeat trading, essential to the larger revenue opportunity, remains unproven.

Verdict

AI companies are spending billions on inference, yet much of that capacity is still negotiated over email. Touchmark is betting inference will follow oil, with standardized contracts traded on an exchange.

The challenge is building enough liquidity to make that market work. If Touchmark succeeds, its value may have less to do with how much compute it trades and more to do with how much of the industry relies on its prices.

The Startup Pulse

Another happening week in startup funding. Three signals from this week:

1. AI infrastructure continues attracting massive capital

2. Investors are backing AI inside established industries

3. Public markets want clearer economics from capital intensive businesses

   • Manus AI: Raised $500M+ led by Boyu Capital and IDG after its $2B+ Meta acquisition fell through. The Singapore startup is now betting on independence.

   • Oxide Computer Company: Raised $445M Series D led by Eclipse, betting enterprises will prefer owning cloud infrastructure over renting from hyperscalers.

   • Vinci: Raised $250M at a $1.5B valuation, backed by Advent and Temasek. Its AI simulations aim to cut costly hardware design cycles.

   Read more

Supported By

Strategic breakdowns of how the biggest AI companies actually work — their business models, their moats, their vulnerabilities. This week: What Apple is actually doing with AI. It’s not what you think.

Written by Ashher

Update your email preferences or unsubscribe here

© 2026 AngelsRound

Austin Texas, United States