The first Physical AI stock is about to be born. Can Wuyi Vision make AI understand the physical world better?

As large models become entangled in the repeated game of parameter scale, inference costs, and price wars, the narrative focus of the AI industry is undergoing a critical shift.

Recently, according to disclosures from the Hong Kong Stock Exchange, Beijing 51WORLD Digital Twin Technology Co., Ltd. (51 WORLD, hereinafter referred to as "51WORLD") has passed its listing hearing and plans to use the Chapter 18C framework to become the "first stock of Physical AI."

Over the past three years, generative AI has almost monopolized all attention. Models can write copy, code, and make PPTs, but they still fail to solve a fundamental problem: Does the content generated by AI truly conform to the physical laws of the real world?

In the face of frequent "misjudgments" in autonomous driving, robots struggling to perform complex actions stably, and industrial simulations heavily reliant on empirical parameters, this issue is evolving from an academic debate into an industrial bottleneck.

This is where Physical AI comes to the forefront. It does not aim to create models that are better at "talking" but instead seeks to enable AI to understand gravity, friction, space, causality, and constraints, allowing algorithms to complete training and trial-and-error in a virtual yet realistic physical world.

It is against this backdrop that 51WORLD's listing has, for the first time, placed the question of "whether the physical world can be computed at scale" on the table of the capital market.

As Large Models Begin to "Float," 51WORLD Chooses to Start Over from the Ground Up

Over the past decade, the evolution of AI has primarily accelerated within the information world. But today, more and more industries are beginning to feel the accompanying "weightlessness."

The growth in the capabilities of large models has not linearly translated into reliability in the real world. Autonomous driving still requires extensive real-world road testing, robots frequently fail in unfamiliar environments, and industrial systems remain highly dependent on manual parameter tuning. The core contradiction behind this lies in the fact that information intelligence does not inherently understand the physical world.

Physical AI seeks to address this very disconnect. It requires that AI's learning process be embedded in physical rules, spatial structures, and real-world limitations, which, in turn, places extremely high demands on underlying infrastructure—not only requiring massive data but also "controllable, reproducible, and scalable" real-world mapping capabilities.

This precisely explains why 51WORLD's technical path differs so markedly from that of mainstream AI companies. Founded in 2015, 51WORLD did not start with algorithms or models but rather with a seemingly "clumsy" goal: cloning 510 million square kilometers of Earth.

Its core is not "Earth" itself but rather the construction of a computable, simulatable, and repeatedly callable foundation for the physical world.

Around this goal, the company has gradually developed three interlocking technical pillars. The first is the 51Aes digital twin platform, used for high-precision modeling of spatial structures and physical rules.

The second is the 51Sim synthetic data and simulation system, which replaces high-cost, non-repeatable real-world data collection with virtual environments. The third is the 51Earth digital Earth platform, which extends these capabilities to macro-scale applications such as cities, transportation, and energy.

The combination of these three forms a complete Physical AI loop, from synthetic data generation to spatial model construction and then to simulation training and validation.

Globally, few companies truly possess this complete chain. Even products like NVIDIA's Omniverse and Meta's Habitat remain largely at the tool or platform level, rather than being systematic projects centered on the computability of the real world.

This is also why 51WORLD's value in scenarios like autonomous driving, robotics, and industrial simulation is not reflected in any single hit application but rather in its long-term digital abstraction of the real world.

To date, its products have served over 1,000 medium and large enterprises across 19 countries and regions, with deep involvement in the formulation of 41 international and domestic standards.

This standard and infrastructure attribute positions it closer to a foundational supplier in the era of Physical AI, rather than a traditional solutions provider.

As such, 51WORLD's value lies not in short-term commercial returns or the success of a single application but in its construction of a sustainable, reproducible foundation for the physical world and the complete Physical AI loop formed around it.

The Long-Term Gamble Under Chapter 18C: 51WORLD's Controversy Is Not About Technology

Although its technical path determines 51WORLD's uniqueness, the real question troubling the capital market remains an old one: When will it turn a profit?

From a financial perspective, 51WORLD does not fit the traditional narrative of a growth stock. From 2022 to 2024, the company's revenue grew from RMB 170 million to RMB 287 million, with a year-on-year increase of 62% in the first half of 2025, showing steady overall growth.

However, the phased expansion of losses reflects the upfront nature of R&D and market investments. From 2022 to 2024, net losses were RMB 190 million, 87 million, 79 million, and 94 million, respectively.

But the problem is that evaluating 51WORLD using the logic of application software companies is inherently misplaced. The commercialization rhythm of Physical AI is naturally different from that of consumer internet or SaaS. It is closer to the early development trajectory of cloud computing, operating systems, or even semiconductors—capital-intensive upfront with delayed returns, but once standards and ecosystems are formed, marginal value rapidly amplifies.

The existence of Chapter 18C itself is a regulatory space reserved for such companies. The Hong Kong Stock Exchange allows unprofitable but high-tech and strategically significant companies to list, signaling the capital market's attempt to understand a new growth logic: measuring technological depth not by short-term profits but by betting on long-term industrial positioning with technological depth.

51WORLD's investor structure also corroborates this judgment to some extent. Its backers include well-known funds like Lightspeed and Cloud9, as well as industrial capital from companies like SenseTime and Moore Thread, the latter focusing more on technological synergy and long-term ecosystem value rather than single-point financial returns.

As of the latest funding round, the company's valuation reached RMB 4.4 billion, ranking first in fundraising scale in China's digital twin solutions market.

Thus, the real controversy lies in whether the capital market is willing to price "Physical AI infrastructure" ahead of time. In an era where large model companies are frequently revalued and computing costs become a core variable, the value of Physical AI has yet to be fully validated but cannot be easily replaced. This uncertainty is precisely the biggest highlight of 51WORLD's IPO.

From a longer-term perspective, if AI is to truly penetrate core systems of the physical world—such as transportation, manufacturing, cities, and energy—then synthetic data, spatial modeling, and simulation training will no longer be optional but essential. Once the inflection point arrives, the valuation framework for Physical AI may completely break away from traditional software logic.

Conclusion

51WORLD's listing may not signify that Physical AI has matured, but it could very well mark the first time the capital market systematically prices the question of "how AI understands the real world."

While large models determine what AI "can say," Physical AI determines whether AI can truly "do something." And this time, capital is being forced to take sides early.