ThenAI

Marketplace and fulfillment infrastructure for AI training data. Connects AI labs with a distributed network of businesses that monetize the work they already do. All the data suppliers in this market work with each other. When a lab places an order, it chops it across multiple vendors. When a vendor wins a contract for a region where it has no supply, it quietly subcontracts a competitor who does. The valuable position isn't being another vendor in that network — it's being the neutral infrastructure the network clears through. And the only supply that scales with the market's churn is data from businesses doing work they already do. Buyer requirements flip every couple months; a supplier network built on existing operations can swap in and out with demand. One built on bespoke collection projects can't. Winning means signing up businesses as suppliers across geographies, which is a trust problem before it's a technology problem. The latest supplier is a family-owned factory in Taiwan — no cold email closes that deal. Closes because Chih speaks the language, understands how a family business there makes decisions, and can sit across the table and explain why cameras over their line are revenue, not a threat. Career accidentally optimized for this: Accenture (Canada), Zuora (China for 18 months, Australia), operating across Asia, Latin America, Africa, and the US. Those relationships are literally the supply map for the company. The Skyflow piece gives the compliance side — knows what a lab's compliance team will ask before they ask it. The combination of bridging a factory floor in Tainan and a robotics lab in San Francisco is genuinely hard to hire for. Three things converged in the last three to six months. 1) Demand crossed a threshold: robotics and physical AI labs are gated by data volume with no web to scrape for the physical world — labs are buying hundreds of thousands of hours of factory video a month. 2) The easy supply already commoditized: crowdsourcing hubs like India are falling out of favor because everyone has that data; what buyers pay for now is diversity (environments, regions, workflows), and that supply is fragmented across thousands of businesses — fragmented supply plus recurring demand is when a marketplace wins. 3) Provenance became a purchasing requirement between the EU AI Act and scraping lawsuits — labs need to show where data came from and that they had the right to use it, killing the gray-market shortcut. Physical AI hasn't specialized yet the way LLM data did; whoever builds the supplier network and trust infrastructure during the volume era becomes the default when it fragments into specializations.