AINEWS
A reducer turns motor spin into slow, precise joint motion. Taoshi's 1.4 inch version closed a ¥140M round, roughly $20M, at a 1B+ RMB valuation, plus a 100,000 unit deal to feed robot hands.
A 1.4-inch gear-reducer built in Shenzhen now ships into Foxconn's assembly lines and into the joints of humanoid robot hands, under a 100,000-unit supply agreement signed in 2025. The structural choice that lets the same part do both jobs, an orthogonal 90° design that integrates reduction and turning into a single piece, is becoming a chokepoint in the humanoid robot hardware stack.
A reducer is the unglamorous gear that turns a motor's fast spin into slow, precise joint motion. Most of the ones in industrial robots today are harmonic, RV, or planetary designs, all well-understood and all roughly the size of a small tin of mints. Taoshi Intelligent Technology, founded in Shenzhen in 2016, bet on a different geometry: a ring-envelope worm-gear reducer with the input and output shafts at right angles. The company says the resulting part is about 40% smaller than conventional reducers, hits ±0.5 arc-minutes of positioning precision, and survives 10,000 hours of operation before its tolerances drift.
The bet, eight years in, is now a receipt. The company closed a ¥140M round (about $20M at recent exchange rates) at a valuation above ¥1B (about $140M), with checks from Qianhai Financial Holdings, Guangzhou State-owned Capital, Haichuan Juyi, and Hangzhou Zhongshen, and Detai Capital as financial advisor. The factory is already running on Apple's contract-manufacturing chain: Foxconn, Luxshare, Lens Technology, Lead Intelligent, and Lingxun Optics buy Taoshi reducers for industrial automation, and the company says it now works with close to 100 humanoid robot customers. A 2024 order from an aerospace customer opened a second lane. Then in 2025 the company signed a 100,000-unit supply agreement with dexterous-hand makers Lingxin Chaoshou, Chaowei Dongli, Shengji Miaosuan, and Zhongke Guiji, putting Taoshi parts inside the joints of humanoid robot fingers.
The same Shenzhen production line that feeds Foxconn is now feeding robot hands, and the bottleneck is not the AI model, the battery, or the chip. Industry data, cited in the company's 36Kr first-release coverage, puts the combined share of actuators (motors, reducers, lead screws) and sensors in a humanoid robot's bill of materials above 60%, roughly 45% for the actuator system and 15% for sensors. A reducer that fits in a fingertip and can hold sub-arc-minute precision for thousands of hours is the part that decides whether a robot hand can pick up a wine glass without crushing it.
The mechanism is straightforward: a 90° orthogonal worm-gear design, machined on an in-house seven-axis five-linkage grinder. Founder Tao Yulong told 36Kr that the company's moat is not the gear design but the full design-to-verification production system, including dedicated heat-treatment and lubrication lines. The first lifetime tests on the company's reducer beat a traditional worm-gear baseline by only 50% before the team had to rebuild those processes from scratch to get to the 10,000-hour figure now in the spec sheet.
The round is small by humanoid-robot standards but large for a single reducer maker. The current Taoshi base, about 20,000 square meters in Shenzhen, produces roughly ¥50-70M a year in output and 500,000-700,000 key joint modules. A new plant under construction will, when running, push that to 1-1.5 million modules a year. The 2025 R&D plan, Tao said, is the first time the company has tilted engineering away from industrial automation and toward robot joint modules, including a new highly integrated dexterous-hand joint module that the company says will be compatible with direct-drive, linkage-drive, tendon-drive, and hybrid-drive architectures.
The bet is not risk-free. The 100,000-unit figure is a signed supply agreement, not a delivered order, and the company has not published pricing, a delivery schedule, or how the units will be split across the four named customers. The performance claims, ±0.5 arc-minutes of precision, 10,000-hour life, 40% volume reduction, and 1300 MPa tensile strength, are the company's own, and have not been independently benchmarked against the harmonic, RV, or planetary reducers that dominate today's industrial robots. Orthogonal worm-gear designs have been tried at scale before and are not a default choice for high-precision robotics. The founder himself describes 2025 humanoid demand as "commercial exploration" rather than a guaranteed market, and the round's investors are regional state-affiliated and family-office vehicles, not pure strategic capital.
The watch item is whether Taoshi's production line can hold tolerance across a 3x capacity expansion while keeping the actuator stack share of humanoid BOM cost at the 45% the company cites. The next data point will be the first deliveries under the 100,000-unit agreement.