Deals

DexForce Hits $1.4 Billion Valuation Training Humanoid Robots With Synthetic Data, Toyota and Samsung Already On Board

A Shenzhen startup's simulation engine is replacing physical training floors for tier-1 manufacturers, compressing humanoid robot deployment cycles that once required thousands of hours of human demonstration.

Mainstay Communications Team · October 15, 2026 · 4 min read

DexForce Hits $1.4 Billion Valuation Training Humanoid Robots With Synthetic Data, Toyota and Samsung Already On Board

In brief

DexForce Technology raised 1 billion yuan ($147 million) in a Series B round in June 2026, pushing its valuation above 10 billion yuan. The company's DexVerse simulation engine trains humanoid robots entirely on synthetic data, then transfers that learning to physical machines, a Sim2Real approach now running inside facilities at Toyota, Samsung, Foxconn, and a growing list of industrial manufacturers. CEO Jia Kui is reportedly considering a Hong Kong IPO, according to Bloomberg.

The Problem With Physical Training Data

Training a humanoid robot on factory tasks has always been expensive and slow. Traditional pipelines require human operators to demonstrate each motion 100 or more times, in a single environment, under a controlled set of conditions. Every new task restarts the cycle, and every new facility means starting over.

DexForce, founded in 2021 by Jia Kui, a machine-learning professor at the Chinese University of Hong Kong, Shenzhen, built its thesis around eliminating physical data collection as the bottleneck. Jia stated the problem in stark terms: "Trying to achieve 'physical artificial general intelligence' purely through real-world data collection would require 10 trillion samples."

The company's DexVerse Engine creates a digital replica of a workspace or production process, varies conditions algorithmically across lighting, object geometry, and task angle, trains a robot model inside that environment, then transfers the learned weights to a physical machine. The gap between what works in simulation and what holds up on an actual factory floor, a known vulnerability in Sim2Real systems, is the central engineering problem DexForce has been working against since founding.

What's Running in Production

The approach has moved past proof-of-concept. Toyota, Samsung, Panasonic, Foxconn, Haier, and Hisense are among the manufacturers running DexForce's software inside active operations. The client list spans automotive assembly, consumer electronics manufacturing, and home appliance production, three categories with different task profiles and near-zero tolerance for failure.

DexForce's own hardware, the W1 Pro humanoid, has 40 degrees of freedom, runs a pure vision sensor array with plus-or-minus 1mm precision, and operates real-time joint control at 1000Hz. DexVerse-trained models run on the W1 Pro as the reference deployment target. The company has also placed consumer-facing robots at malls and tourist sites across more than 15 Chinese cities, where they handle coffee service and customer assistance: a lower-stakes environment that generates operational feedback at volume.

The Round

DexForce closed a 1 billion yuan Series B in June 2026. Investors in the round included Shenzhen Capital Group, Lenovo Capital (the investment arm of Lenovo), Lens Technology, ICBC Capital, Hengjian Asset Management, Qianhai Fund of Funds, and Guiyang Digital Economy Fund. The post-money valuation cleared 10 billion yuan, equivalent to roughly $1.4 billion.

The capital structure reflects who wants the category to succeed. State-aligned funds from three separate Chinese cities are on the cap table. Lenovo Capital brings hardware distribution relationships. Lens Technology, a Changsha-based component manufacturer that supplies Apple and other consumer electronics brands, has existing production relationships with some of DexForce's industrial clients.

The roughly 500-employee company is reportedly in preliminary conversations with advisers about a Hong Kong IPO that could raise several hundred million US dollars, according to Bloomberg reporting, which cited people familiar with the matter. No listing has been confirmed.

What the Model Is

The investment case rests on software infrastructure, not on the W1 Pro itself. DexForce's deployable product is the DexVerse Engine running inside a manufacturer's operations, on whatever physical hardware that manufacturer already runs. The W1 Pro matters as a validation platform and as a revenue line, but the repeatable software business is the company training other manufacturers' robots at the simulation layer.

That model is closer to industrial software than to the humanoid robot category it gets grouped with. The manufacturing clients contract a training pipeline that can add new tasks in simulation before anything touches a production floor, running on hardware they already own.

The open question for every Sim2Real company is whether the reality gap stays manageable as task complexity scales. Smartphone assembly, automotive sub-assembly, and packaging each introduce edge cases that a simulation can represent but may not fully capture. DexForce's commercial deployments at Toyota and Samsung are the early evidence one way or the other.

Sources

Companies mentioned

Mainstay Digital · Spatial and 3D visualization for industrial companies

Back to Newsroom