The market is misjudging AI data companies. We are trying to analyze a shift in human knowledge through an old software lens. Frontier labs spend close to $1 billion a year each on human data. The annual market is already $10 billion. @HelloSurgeAI crossed $1 billion in revenue while remaining bootstrapped. @mercor is discussing a $20 billion valuation. @Meta paid $14 billion for half of @Scale_AI. For decades, tech budgets targeted productivity. Software automated tasks. Today, companies buy intelligence. This requires a completely new supply chain. The main input is human judgment turned into structured data. @aliansarinik at @micro1_ai estimates this opportunity at $1 trillion annually because it measures a core input to the economy. The big question is who buys this infrastructure next. The first wave is robotics. Language models grew on a free internet text subsidy. Robots have no equivalent. No archive shows a machine how to fold laundry. Every physical action must be captured and structured by hand. The second wave is enterprise agents. Traditional software testing cannot validate a probabilistic system. Engineers cannot write a standard test script for an agent in a dynamic corporate workflow. Humans must define the failure modes and build the evaluation datasets. At @Flybridge, we saw this pattern early. When we backed @Arcee in 2023, they built a platform to train small models inside a customer private cloud. The market ignored it then. Most thought everyone would just query a central model. We believed corporations would demand models they could own and inspect. @SatyaNadella noted that companies using external models pay twice, once with cash and once with proprietary data. Today, private model deployment is standard. The buyer base is scaling from 10 labs to over 1000 corporations. They all need private data loops. Skeptics see tech-enabled staffing agencies with thin margins and high customer concentration. But concentration is normal for early infrastructure. The first buyers of cloud computing were a small group of high-growth startups. The market diversified later. Margins also grow as the work moves up the stack. Basic labeling is cheap. Providing reinforcement learning environments and custom reward models generates software-like returns. The moat is operational execution. Building that muscle takes years. The long-term path relies on what the micro1 research team calls the no last mile framework. Better models do not eliminate the need for human input. They unlock more complex tasks. Each new tier of automation requires human judgment to supervise the next layer. Human expertise is becoming a primary input to the global economy. If you are a founder building infrastructure in this category, let us know. I would love to chat.