Remote operators at Apptronik's Austin facility guide every Apollo robot cycle, log it as training data, and underwrite a $5.5 billion humanoid bet.
A person stands off to the side of a 90,000-square-foot warehouse in Austin, eyes on a screen, hands on a controller. Across the floor, an Apollo humanoid waits beside a conveyor belt. The person issues a command. The robot lifts a box and stacks it. The cycle takes a few seconds. The next cycle starts immediately.
That loop is what Apptronik calls the future of work. The Texas robotics startup, now valued above $5.5 billion (a figure reported by Bloomberg following the company's February 2026 funding round), has named its facility Robot Park, a training ground where its Apollo humanoids practice warehouse and light-manufacturing tasks before any real-world deployment. A Business Insider facility tour, later republished by the Indian Express, shows the robots stacking boxes and sorting toys into bins, with human operators nearby and often guiding actions remotely.
The novelty is less the robot and more the throughput. The facility runs seven days a week, and Apptronik's pitch — per CEO Jeff Cardenas and the Business Insider facility reporting — is that every guided cycle produces a stream of training data that loops back into Apollo's AI models. Each task adds a data point; each data point nudges the next version of the model. Co-founder and CEO Jeff Cardenas has framed the site as a "manufacturing plant whose output is knowledge," calling it a "data factory," a label borrowed from the playbook of self-driving car programs that logged millions of human-driven miles.
The money behind the framing is real. In February 2026, Apptronik closed more than $935 million in Series A funding, including a new $520 million extension that tripled the company's valuation to over $5.5 billion, according to Bloomberg. The raise is on the public record in a Form D filing with the SEC. Capital is no longer the binding constraint. The binding constraint is whether the data loop produces a robot that learns faster than it has to be steered.
The current commercial test is a Mercedes-Benz pilot. Apptronik and Mercedes-Benz entered a commercial agreement to deploy Apollo inside Mercedes-Benz manufacturing facilities, an arrangement described in a Manufacturing Dive write-up. The agreement is a pilot, not a fleet rollout. The number of units, the production lines involved, and any timeline for scaled deployment have not been disclosed. That gap is where most humanoid programs still live: a working demo in a controlled cell, with a customer willing to test it.
The wider industry is converging on the same data-first bet. Tesla chief executive Elon Musk has described an "Optimus Academy," a concept in which Tesla's Optimus humanoid would train continuously from real-world interactions. No such facility is shipping, and the description is a stated goal, not a deployed product. The underlying argument is the same: industrial collections of human-guided data are increasingly treated as the precondition for useful humanoid autonomy. Apptronik's framing puts the company in that camp, with a working site rather than a render.
Strip away the marketing, and what happens at Robot Park is human labor, logged at industrial scale. A teleoperator produces a labeled action that becomes a training example, which is supposed to reduce the need for the next teleoperator. The loop only closes when the model can replace the human at the controller. Until then, every box on the conveyor belt is a paid data point, and Apptronik is betting close to a billion dollars that the model closes the loop before the runway does.