TensorWorld Secures Over $100 Million in Funding Within a Month of Launch
KuCoinFlashShare
TensorWorld, a physical AI startup, has secured over $100 million in project funding just one month after its launch in July 2026. The company is developing a physics-driven AI model to enhance real-world machine interaction, with applications in robotics, autonomous driving, and industrial simulation. Investors include Kaiwei Fund, Sequoia China, Ant Group, and Meituan Strategic Investment. The rapid funding underscores strong investor interest in the project, with a potential token launch expected in the coming months.
This is one of the most intense battlegrounds in the AI industry today. As large language models drive rapid advancements in intelligence within the digital world, AI’s capabilities are extending into the physical world—embodied intelligence, autonomous driving, industrial simulation, and more. But the physical world doesn’t play by tokens; it operates by its own immutable laws: gravity, friction, collision, fluid dynamics... the complexity increases exponentially compared to language.
On this more challenging, more fundamental track, TensorWorld, a startup founded just four months ago, has secured over $100 million in funding within a month. Behind the capital's bet lies the rarity of its capabilities: academic background, technical approach, and practical implementation.
What problems does the industry need to solve?
Currently, in the public’s general understanding, AI is already very intelligent—it can write code, find vulnerabilities, and even solve mathematical problems that have puzzled humanity for centuries. However, getting robots to reliably predict and interact with the trajectories of real-world objects remains extremely difficult. The reason is that AI has not yet grasped the laws of physics; bridging the gap from the “software interface” to the “physical world” is an absolute necessity.
The industry faces extremely stringent conditions in bridging this gap: high-fidelity physical annotation data is prohibitively expensive, with a single industrial-grade aerodynamic simulation sample often requiring tens of thousands of CPU hours; data collected across 100 scenarios struggles to generalize to 10,000 unknown scenarios; simulators find it difficult to fully replicate real-world friction and deformation, with the Stanford HAI report showing that robot task success rates in software simulations can reach 89.4%, but only about 12.4% in real home environments.
How does WuYuan Technology address this issue? The academic evolution shows a clear path.
Currently, world models in the industry can generally be categorized into two paradigms: visually driven and physically driven. The former is the mainstream approach, exemplified by video generation, aiming to “look realistic” by optimizing pixel consistency between frames. Its advantages include relatively easy data acquisition and highly realistic visuals, making it widely used in video generation, data factories, scene previews, and VLA front-end perception. However, for robots to reliably perform tasks in the real physical world, language and visual understanding alone are far from sufficient.
Therefore, another paradigm is increasingly recognized by the industry, focusing on modeling with physical laws with the goal of “deriving greater realism.” Its advantages include verifiability, measurability, and support for action planning. Traditional industrial simulation software (such as MuJoCo and Isaac Sim) represents the “handcrafted” approach along this path—physical laws are heavily dependent on engineers manually writing each rule, resulting in precise predictions but at high cost, with enormous engineering effort required for irregular shapes, soft bodies, and fluids.
WuYuan Technology follows the path of physical foundation models, proposing "building physical reality in the tensor world," with the goal of enabling models to learn the evolving laws of the physical world from data and achieve human-like "learning one thing and applying it to many."
For example, the native Transformer lacks the ability to decompose time series—how can the model correctly understand “time” in the physical world? Autoformer significantly improves the accuracy of long-term sequence prediction, while TimesNet further makes temporal data more intuitive and visual. Similarly, how can the model understand “space”? Uni3D extends 3D representation learning to a billion-parameter scale, enabling the model to form concepts of objects, geometry, and spatial relationships.
On this foundation, GeoPT answers the fundamental question of this domain: Can "physical intuition" be pre-trained into models? The answer is yes—it creatively enables models to acquire transferable physical representations through pre-training that integrates geometry and synthetic dynamics, reducing labeled data requirements by 20% to 60% in industrial physics simulation tasks and accelerating convergence by up to 2 times. Transolver further enables models to understand forces, flows, and deformations in irregular spaces, comprehending more complex physical dynamics. EfficientZero allows agents to simulate the future based on learned world models and plan accordingly.
The foundation for these cutting-edge achievements lies in the long-standing expertise of the Wuyuan founding team, spanning prestigious international academic institutions such as Tsinghua University and MIT. Following the progressive framework of “temporal evolution—spatial representation—physical laws—action planning,” they have not merely compiled isolated papers, but consistently built upon physical laws to achieve true “high-level physical intelligence,” ultimately shaping the complete outline of a physics-based foundation model. In addition, the Wuyuan team previously led products serving hundreds of millions of users and managed computing infrastructure with hundreds of thousands of accelerators, bringing extensive real-world industry experience. While academic research determines how far a model’s capabilities can extend, engineering prowess determines whether the model can be applied in real-world business scenarios—these two aspects form the two inseparable sides of a physics-based foundation model.
Why is Wuyuan Technology favored? The physics-driven paradigm is rare.
Observing the technical roadmap, it directly addresses the three major pain points of physical AI: on the data side, it replaces physical labels with synthetic dynamics, significantly reducing dependence on expensive real-world data; on the generalization side, it deeply integrates spatial structure, temporal evolution, and physical dynamics, enabling the model to spontaneously discover universal physical laws; on the deployment side, underlying architectures such as Autoformer and Transolver have been widely validated and adopted by industry (e.g., NVIDIA PhysicsNeMo), alleviating concerns about AI deployment in real-world scenarios.
The rarity of the intersection—“true understanding of physical systems + world-model-driven planning capabilities + engineering prowess to scale cutting-edge research”—quickly attracted capital. Since its founding in July, Wuyuan Technology has raised over $100 million in consecutive seed and seed+ rounds within a month. The seed round was co-led by Kaihui Capital and Sequoia Capital China, with participation from IDG Capital,鼎晖VGC, Xinglian Capital, and Ant Group. The seed+ round was co-led by Ant Group, Yuchai Capital, and Meituan Strategic Investment, with continued participation from Dacheng Capital, NIO Capital, Jinghe Capital, and several existing investors.
This funding list also conveys significant insights: Meituan’s corporate investment arm, with its vast C-end scenarios; Ant Group, with its serious industrial applications; and NIO Capital, with its autonomous driving focus, all validate that Wuyuan Technology’s approach possesses strong generalization capabilities and can meet broader industry demands.
The world model赛道 has already become increasingly "crowded," but challenges such as error accumulation in long-term prediction, the final gap between simulation and reality, and the lack of unified evaluation standards remain unresolved. If Wuyuan Technology’s approach succeeds, it would mean the model truly gains the ability to understand how the world operates and anticipate the consequences of actions—its value extending far beyond single applications like robotics or autonomous driving, and instead potentially becoming the foundational standard and base for the Physical AI ecosystem, paving the way for AI’s transition from the digital world to the physical world.
(Source: Ifnar)
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information.
Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.
