Deep Investment Research16 min read
Workers Are Training Their Replacements: The Hidden Royalty in Robot Data
Workers film dishes, welding and packing for one-time pay. Robotics companies can reuse those motions indefinitely. The investment question is whether cheap collection becomes a durable moat—or remains purchasable gig labor.

One Wage, Permanent Skill
Bloomberg reports workers in India earn an extra 150 rupees an hour to film physical tasks. Their motion happens once; the dataset can train machines indefinitely.
Physical AI is creating motion copyright without royalties. The worker gets labor income while the buyer acquires a long-lived intangible asset.
Why Internet Data Is Not Enough
Robots need friction, occlusion, weight and failure recovery—not merely the next token. Micro1 employs thousands across more than 50 countries at $15 an hour, while Scale AI has collected over 100,000 hours.
The moat is not video volume. It is exclusive rights, rare failures and measurable transfer across robots and environments.
A New Data Supply Chain
DoorDash Tasks can turn more than eight million Dashers into distributed sensors. Mecka raised $60M and claims $100M of contracted run-rate revenue; XDOF raised $70M.
Figure raised more than $1B at a $39B post-money valuation. Investors must test whether each deployment improves every robot or merely creates another custom integration.

Cash Flow Is the Counterexample
Symbotic reported $721M of quarterly revenue, $55M of net income and roughly $22.5B of backlog. It shows that robotics value ultimately rests on measurable throughput and cash flow.
Use the DCF calculator to separate hardware, deployment, software and data economics, then stress test intervention rates, data reuse and customer concentration.
The Masters’ Test
Buffett asks whether proprietary data creates pricing power. Munger inverts incentives between one-time worker pay and permanent data rights. Graham refuses to capitalize an unproven flywheel at full value. Howard Marks looks for picks-and-shovels while humanoid consensus drives prices higher.
The opportunity is not the largest video library. It is the company that lawfully controls differentiated data, reduces failures and converts copied human motion into free cash flow.
Recent posts








































































%20-1.jpg?width=1600)
Frequently asked questions
Can robot data be a moat?
Only if it is exclusive, lawful, reusable and demonstrably lowers failure and deployment costs.
What is Figure’s reported valuation?
$39 billion post-money in its last publicly reported funding round.
Value the intervention rate, data rights and cash burn—not the demo video.
Download Valuefocus