Jabil Says Humanoid Makers Are Heading for Tens of Thousands of Units

Thomas Brown, senior director for engineering and technology at contract manufacturer Jabil, said the humanoid industry is "entering a phase of tens of thousands" of units, up from the low volumes most customers have built so far.

Brown made the remarks in answers to questions put by The Robot Report, published Oct. 8 and written by Eugene Demaitre, the editorial director of its robotics group. Jabil is a design, supply chain and manufacturing services provider based in St. Petersburg, Fla. Brown said it works with major humanoid developers in the U.S. and abroad but cannot name them. The one customer relationship the article names is Apptronik, whose Apollo humanoid Jabil is helping bring into active manufacturing environments.

The constraints Brown named

Brown's list is short and each item is a cost line. Expensive robots limit how many a developer can build to gather training data and to validate customer use cases. Edge compute, the processing done on the robot rather than in a data centre, needs more capacity; he said compute is sufficient for many demonstrations and cited NVIDIA's Jetson series as popular with customers. Price, he said, is "an ongoing pain point for scale."

Constraint	What Brown said
Volume	Industry "entering a phase of tens of thousands"
Compute	Enough for many demos, more edge capacity needed; Jetson popular
Wiring	Automotive-grade robustness in a consumer-electronics form factor
Consumer timeline	"Five-ish years," but "could be 20 years"

Wiring is the least-discussed item on the list. Brown said humanoid harnesses need the robustness of automotive wiring in a form factor closer to high-grade consumer electronics, which is the reason he gave for developers wanting partners that can iterate quickly and swap power supplies.

The photograph shows a Boston Dynamics Atlas, not a robot Jabil builds. Boston Dynamics is not among the companies Brown named.

Where he thinks the first units will work

Brown said humanoids suit many factory tasks, but that most existing automation is built for one task and remains the right tool for some. He cited a machine tender running at 1.5 meters a second (3.3 mph) as already fast, and said wheeled bases are an easier problem than bipedal ones.

For consumer use he was less specific. He described cases such as folding laundry as vague, and said humanoid companies are still collecting clear requirements from end customers. His consumer estimate was "five-ish years," though he said it "could be 20 years." That is a spread of 15 years on one estimate, offered with no confidence interval.

What is not on the record

The article gives no unit price, no order backlog, no Jabil production count and no date for Apptronik's series production. Brown's "tens of thousands" has no denominator: it does not say per year, per company or across the industry, and the interview does not say which.

On safety, Brown said standards for standoff distances and cages exist for robot arms and need to be extended to humanoids, with failure modes, certification in new environments and prediction of model behavior still open.

The article also notes that Tesla, 1X Technologies, XPeng and UBTECH are preparing to ramp humanoid production. It does not give their targets. For one that does carry a figure, Boston Dynamics' Alberto Rodriguez said on Oct. 1 that the team is working toward 100,000 hands a year; our report on AWS's physical A.I. toolchain covers the training software that sits behind such robots. Related coverage of the sensing side is in our report on Helm.ai's $70 million in signed contracts.

The next public checkpoint is RoboBusiness 2026, Oct. 20 and 21 in Santa Clara, Calif., where humanoids are a listed session topic.