Infiforce has raised nearly 1 billion yuan, or roughly $150 million, across Series A and A+ financing rounds to develop embodied AI models, expand data infrastructure, and deploy more robots in industrial settings. The funding is a bet on a difficult proposition: robots will become more useful not only by receiving better commands, but by accumulating more first-person experience of the physical world.
The company’s official announcement says the financing will support three priorities: research on its AtomBrain embodied-intelligence system and causal world models, continued upgrades to its DataGrid AI infrastructure, and scaled delivery and validation of multiple robot forms in real industrial settings. AI Insider reported the financing as nearly $150 million and said Infiforce is using the capital to increase robot deployments in industrial and commercial environments.
The headline amount is substantial, but the company’s technical thesis is the more important part of the story. Infiforce argues that robots need data collected from a first-person perspective, which it calls Ego-native data, rather than relying primarily on third-person demonstrations or data gathered through robot teleoperation. The objective is to give an embodied model a closer record of what an actor sees, how it moves, and what happens in the surrounding environment.
Ego Data Is Infiforce’s Answer to Physical-World Training
Infiforce describes Ego data as first-person records of real interactions with the physical world. The company says these records include visual observations, actions, spatial relationships, and environmental feedback. Its premise is that a robot needs more than a video of someone else completing a task. It needs data that better reflects the perspective and consequences of acting in a real setting.
The distinction is important for embodied AI. A language model can learn from large bodies of text gathered from the internet. A robot must learn how actions affect objects, surfaces, people, and spaces. That kind of experience is harder to collect because it happens in the physical world, and it can vary greatly across locations and tasks.
Infiforce says it has been building this approach since 2025. Its planned embodied model will be centered on high-quality Ego data, according to the company announcement. The strategy is intended to create a larger and potentially lower-cost source of training experience than collecting every example directly through a robot body.
The company’s argument is not that first-person data automatically solves embodied intelligence. Collecting human-perspective data still requires tools, processing systems, consent practices, and a way to translate observations into useful robot actions. But the funding gives Infiforce resources to build the data collection and training loop it believes is necessary.
That places Infiforce alongside a wider group of Chinese companies trying to make data infrastructure a competitive advantage in robotics. EastFrontier’s report on SCALEFORCE’s embodied AI data infrastructure showed how investors are funding systems that collect, process, and organize physical-world data. Infiforce is pursuing the same broad objective through its own DataGrid and Atom model stack.
DataGrid Links Collection Systems to AtomBrain Models
Infiforce says DataGrid is an end-to-end training system that combines data collection, data processing, and hardware. The company lists handheld grippers, first-person collection devices, and robot teleoperation among its collection methods. The stated purpose is to turn real tasks into physical-interaction data that can be learned from, trained on, and executed by a robot.
The data infrastructure sits alongside Infiforce’s Atom series of models and AtomBrain, its proposed unified embodied brain. The company says its robot portfolio includes the AstroDroid wheeled humanoid, UltraDroid general-purpose robot, Little Atom bipedal robot, and specialized FORCE systems. The ambition is to develop intelligence that can be reused across different robot bodies rather than being locked into a single machine.
That is a major technical and commercial challenge. A model trained on a wheeled service robot may not transfer cleanly to a bipedal robot or a specialized industrial unit. Each form factor has different sensors, movement constraints, safety requirements, and available tasks. Infiforce is betting that a common data and model layer can make that transfer more efficient, but the company has not yet demonstrated a universal cross-robot system through independently reviewed commercial results.
Infiforce’s announcement includes several research results on Libero, RoboTwin 2.0, and CALVIN benchmarks. Those performance claims are reported by the company and concern research evaluation rather than customer deployments. They should be read as evidence of the company’s technical direction, not as proof that its robots have achieved comparable results in factories, warehouses, or public environments.
The more relevant commercial question is whether the company can establish a productive feedback loop. A robot performs a task, its activity creates data, DataGrid processes that data, and a later model iteration improves the robot’s performance. If that loop works at scale, each deployment can become both a product and a source of training material.
The Funding Targets Real Deployments and Industry Scale
Infiforce said the funds will support multiple robot forms in real industrial settings. The official announcement says the company is carrying out validation or commercialization in more than 30 Chinese cities and more than 100 real-world scenarios, including commercial services, warehousing, logistics, and industrial manufacturing. Those are company-reported deployment figures and have not been independently audited in the sources reviewed.
The company also said it signed a strategic agreement with CRRC High-Tech in July covering embodied intelligence for infrastructure applications. It described its involvement in a proposed national technical specification for crowdsourced embodied-intelligence data collection and management. These developments show that Infiforce is trying to position itself not only as a robot developer but also as a supplier of data and training infrastructure for the broader physical-AI sector.
Its financing round was led by Dunhong Asset Management and leading state-owned investment platforms, according to the company. Zhejiang University Science and Technology Innovation Group, Yandu State-owned Assets Management, Lishui State-owned Assets Management, and other investors also participated, while existing shareholder Genesis Partners Venture Capital increased its investment, AI Insider reported.
The investor list underlines the industrial focus of the round. Embodied AI requires long development cycles, hardware manufacturing, field deployment, and large-scale data work. That is a different capital profile from a software-only AI startup. Infiforce needs to fund research, collect data, build systems, and place robots in environments where they can generate new experience.
China’s current wave of world-model investment shows why that approach is attracting money. EastFrontier found that world-model startups had raised $5.6 billion in robotics deals as investors looked for the software layer that could make physical machines more adaptable. Infiforce’s nearly 1 billion yuan round adds a company with a particular view of how that software should be trained.
The next test is execution. Infiforce has described an Ego-data model, a DataGrid system, AtomBrain, and deployments across more than 30 cities. The funding will matter if those pieces produce reliable robots in commercial settings and if the resulting data improves performance across tasks and hardware forms. The company has articulated a detailed training thesis. It now needs to demonstrate that its first-person data approach can turn real-world experience into durable embodied intelligence.