Gm gVangrid
Warehouses Are Solving the Wrong Half of Robot Data.
Humanoids went to work this year. Figure 02 spent eleven months on BMW's Spartanburg line. Figure 03 moved into logistics in June. Global shipments hit 19,100 units in H1 2026, up 272% year over year, with over 70% going into real industrial deployments.
The brains are consolidating too. Nvidia agreed to buy Hugging Face for roughly $12.9B on September 3. Compute is solved. Models are open. Money is chasing data hard. XDOF raised $70M to collect robot data by hand. Scale AI has logged 100,000 hours in its own labs. A facility in Sichuan opened to produce three million robot data entries a year.
All of them share the same limitation. A building, a staff, a rig. Data collected indoors, one session at a time.
A robot needs two kinds of data. How to move, and where it is. The first is being solved in warehouses. The second can't be. Simulation can't invent a city it's never seen, and no lab can send employees to capture every street on Earth every week.
The only sensor network big enough already exists. Three billion phones, in three billion pockets, walking through the world every day.
That's what
@vangrid_io _does. Contributors capture locations, get paid, and the captures become structured geometry anchored on Base. 1,024,912 captures anchored. 423,143 active nodes. $311K settled in USDC. Every number is a transaction you can open yourself.
Ground truth for physical AI isn't coming from a warehouse.