Walk into a workshop in Beijing's Yizhuang district and you will not hear the usual single-product factory rhythm. One assembly bay quietly swaps between completely different humanoid (人形机器人) bodies every quarter of an hour. The same conveyor that finished a tall, athletic machine this morning is bolting together a shorter service-model unit by afternoon.
This is not a robot brand showing off its own hero product. It is a contract manufacturer — the kind of company that usually stays invisible — suddenly building the machines that other robotics companies put their own names on.
Why a parts maker became the story
For two years the humanoid (人形机器人) conversation in China was dominated by the brands: the companies that unveil a prototype, post a video of it walking, and promise volume. The harder problem was always behind the curtain — turning a convincing demo into a product that ships by the thousand with consistent quality.
Lingyi Precision (领益智造, Shenzhen-listed 002600), best known for precision metal and assembly work in phones and consumer electronics, opened what it calls its Beijing embodied-intelligence (具身智能) super-factory on April 17, 2026. The first batch rolled off the line at the Xiaomi Smart Port in Yizhuang, and the models were not Lingyi's own — they were the TianGong Ultra and TianGong 3.0 machines developed by the Beijing Humanoid Robot Innovation Center. In other words, the factory's first celebrity products belonged to a partner.
The ODM model arrives for humanoids
The significance is the business model, not the brand on the box. Lingyi is operating as an ODM/OEM base: it takes a partner's design — or co-develops the hardware — and handles core components, joint modules, final assembly, testing and intelligent warehousing under one roof. The company says it has signed deep cooperation with more than 20 head-tier players, including the Beijing Humanoid Robotics Innovation Center, Agibot (智元机器人), Magic Atom (魔法原子) and BrainCo (强脑科技), and has taken batch ODM orders from several North American AI and robotics firms.
A state-media report in late October described the line in concrete terms: four assembly zones — lower limbs, torso, head and arms — with a finished humanoid (人形机器人) coming off the line roughly every 15 minutes. From a February 9 start to the first units in April, the build-out took about two months plus a 13-day sprint. That speed is the point: the factory is built to absorb demand swings from many clients instead of betting everything on one hero product.
What the capacity targets say
Lingyi's published roadmap is aggressive:
- 2026: planned output of about 10,000 embodied-intelligence (具身智能) robot units (sets)
- 2027: planned capacity of about 20,000 units (sets)
- 2030: planned annual capacity of about 500,000 units (sets)
The Beijing site is described as the first benchmark among a "five bases + global delivery" network, with supporting plants in Chengdu, Zhengzhou, the Greater Bay Area and overseas. A 15-minute changeover between models is the technical enabler that lets one line serve many brands.
According to Li An, chief scientist at BrainNet (脑机网), the real inflection point for humanoids is not any single flashy demo but the arrival of contract manufacturers who can absorb demand swings across many brands instead of staking everything on one hero product.
Where the real bottlenecks sit
A super-factory does not by itself solve the parts that still constrain the whole industry. Dexterous hands, force sensors and high-performance joints remain the hard, high-margin components where consistency at scale is difficult. A 500,000-unit capacity target is meaningless if the downstream buyers — factories, warehouses, hotels — are not ready to deploy, maintain and trust the machines.
The ODM shift also changes who carries the risk. Brands can outsource the brutal manufacturing learning curve to specialists, but they still own the harder software problem: making a humanoid (人形机器人) do useful, reliable work in messy real environments.
Honest limitations
Every figure above except the April 17 rollout date is a company planning target, not audited output. "Planned capacity" of 500,000 units by 2030 describes what the company intends to be able to build, not what the market will buy. The 15-minute changeover and 15-minute-per-unit figures come from company and state-media descriptions of a newly opened line and have not been independently verified at scale. Deployment of humanoids in real workplaces is still early, and component maturity for dexterous manipulation remains uncertain. Treat this as evidence of manufacturing readiness, not proof of mass adoption.
What readers can do now
- If you run a site that could use robots: frame any humanoid (人形机器人) pilot as a shared-platform question, not a single-vendor bet — ask which ODM backs your supplier's volume claims.
- If you track the industry: watch ODM capacity (Lingyi and peers) as a leading indicator of real shipments versus demo hype; announced capacity and booked orders are very different numbers.
- If you invest: separate planned capacity from confirmed orders and from the softer software problem of making the machines actually useful on a factory floor.