The Battle for the Entrance to the Physical World: How Far Is "Jarvis" Away From Us?

In the 2026 tech industry, NVIDIA and OpenAI remain firmly in the top tier, while SK Hynix and Samsung have risen to prominence with strong momentum. Among the multitude of old and new giants vying for the spotlight, the truly thought-provoking player is Jeff Bezos.

Since stepping down as CEO of Amazon in 2021, Jeff Bezos has not been idle at all: he traveled to space using Blue Origin's space technology, invested in Altos Labs to explore longevity technologies, and participated in follow-on funding for Figure AI to bet on humanoid robots.

After a round of engaging in his "billionaire ventures", Bezos likely has only one feeling: the R&D cycle for complex physical products is far too long.

The space industry, biomedical sector, and embodied intelligence are all top-tier fields in advanced manufacturing. These tracks may seem distant from one another, but they collectively point to a long-standing challenge: when the composition of a physical product becomes so complex that it involves tens of thousands of components, the difficulty of the product's journey from R&D and validation to mass production will rise exponentially. Taking the Boeing 777X airliner as an example, the total number of parts in its jet engine alone reaches an astonishing 300,000.

Increasing the thrust of a jet engine by 10% based on existing levels could take a decade using traditional R&D methods, due to the inherent complexity of engineering. Bezos hopes to leverage AI to accelerate this cycle from concept to manufacturing by ten times or more [1].

In November 2025, Jeff Bezos returned to entrepreneurship, founding Prometheus. This marks the first time since he stepped down as Amazon's CEO that he has officially taken on the role of full-time CEO at another company.

Just seven months later, Prometheus completed its latest Series B financing, raising an impressive $12 billion. This is the largest Series B funding round for an existing AI startup, instantly valuing the company at $41 billion.

With such massive capital consumption, yet investors are so willing to place their bets, what makes Prometheus so exceptional?

What Jeff Bezos has presented to the market is far more than just a product — it is an entry ticket to reshape the global industrial system.

Prometheus describes what it is doing as building a "future CAD (Computer-Aided Design) system". CAD is software that engineers use to model any physical object before manufacturing, but in Bezos' view, this description still falls short of capturing the full picture [2].

It is more akin to an engineering foundation, the closest real-world implementation of "JARVIS" from Iron Man — creating an AI tool system that can assist engineers in designing and manufacturing physical products [2]. In other words, using AI to reengineer engineering and manufacturing.

Prometheus's technology is rooted in the physical world, and behind this lies a larger industrial theme: Physical AI.

The World is 3D

At last June's GTC event in Paris, Jensen Huang calculated the massive economic potential of Physical AI: Behind factories, logistics, and humanoid robots lies a $50 trillion industrial playing field [3].

What does $50 trillion actually mean?

According to Bloomberg's latest forecast, by 2030, the market size of broad generative AI — encompassing computing power, software, and related services — will only reach $2 trillion [4]. Jensen Huang's estimate for Physical AI clearly represents a reshaping of the real economy, covering trillion-dollar markets such as manufacturing, logistics, and energy.

At present, there is no strict definition for Physical AI. In simple terms, it refers to AI that can understand physical laws and act upon the physical world.

Autonomous driving and humanoid robots are the two most intuitive applications so far. The former enables machines to perceive, judge, and act on roads, while the latter attempts to give AI a physical body that can interact with the real environment.

However, regardless of whether the final carrier is a car, a robot, or other industrial equipment, for AI to truly exert influence from the silicon-based world on the carbon-based world, one fundamental fact cannot be avoided: the real world is 3D.

NVIDIA Robotics Research: Exploring the 3D World

With enormous market potential and unprecedented track challenges, tech giants, cutting-edge scholars, and industrial capital are all seeking their own paths to entry as AI expands into the physical world.

Jensen Huang was the first to bring Physical AI to the public's attention. He has repeatedly proclaimed at tech conferences like GTC and CES that "the next generation of AI is Physical AI". Furthermore, Jensen Huang placed his daughter, Madison Huang, in NVIDIA's Omniverse division (focused on 3D simulation), and his son, Spencer Huang, in charge of the robotics business. Undoubtedly, both divisions are key strategic departments tied to Physical AI for NVIDIA's next decade.

"AI Godmother" Li Fei-Fei founded World Labs to bet on spatial intelligence, aiming to enable AI to understand the 3D world. Yann LeCun established AMI Labs after leaving Meta, with a core mission of enabling AI to predict changes in the real world. And Jeff Bezos's entry into the arena with Prometheus directly targets manufacturing scenarios to build the underlying engine.

Over the past few years, the silicon-based capabilities of AI have continuously evolved along the dimensions through which humans perceive the world. Text solved language expression, images addressed visual representation, and video added the dimension of time. With each new dimension unlocked, capital and talent pour in, rapidly turning this new technological continent into a bustling battlefield.

Returning to the 3D physical world, everyone will eventually enter. But this time, the form of the battlefield will be completely different from that of the generative AI world.

The Physical World Does Not Accept "Roughly Right"

In the exploration of how to complete 3D mapping of the physical world, there have long been two fields with years of deep cultivation — gaming and industry.

Gaming is a natural use case, and the reason is simple. The gaming industry requires massive amounts of 3D assets, such as characters, scenes, props, etc. At the same time, character movements and environmental interactions must conform to physical laws. Real-time 3D engines like Unity and Unreal are built precisely for this purpose.

Industrial 3D follows a completely different logic and represents a scenario that is far more closely tied to the physical world.

In the probabilistic world of generative AI, it is surprisingly satisfying when images and videos are "roughly right". But in industrial-grade scenarios, products with precision below 99% may be considered waste. If a screw hole is misaligned by half a millimeter, the entire assembly line will come to a halt. If the continuity of a curved surface is not handled properly, both processing and simulation will encounter problems.

Gong Minyan, Founder of Ziqian Technology, has experience in both these two 3D worlds: one is the industrial software system represented by Dassault Systèmes, the French CAD giant; the other is the gaming and real-time 3D world represented by Unity.

This experience gives him a more direct understanding of the differences between the two paths. He summarizes it as the distinction between "drawing the skin" and "drawing the bones": 3D in the virtual world first serves perception and presentation, while 3D in the manufacturing world must serve verifiable engineering results [5].

In the "virtual-to-virtual" world, AI's most critical capabilities are generation speed and expressive power. It solves the problem of content production efficiency, and its commercialization path is closer to game assets, film and television production, marketing displays, and virtual spaces.

In the "virtual-to-manufacturing" world, generation is only the first step. A 3D model must also be editable, measurable, simulatable, processable, and preserve engineering intent. At the same time, connected to the physical world, AI needs to specifically understand geometry, materials, constraints, forces, processes, tolerances, and manufacturing feedback, ultimately bringing a digital object into the real world that can withstand practical testing.

The upper limit of the gaming 3D market is determined by content spending, while industrial 3D faces the R&D, design, and production budgets of the entire manufacturing industry.

Jeff Bezos re-entering the arena to build Prometheus, and Jensen Huang continuously advocating for Physical AI, point to the same trend: The main battlefield of AI is shifting from generating content to generating products, and from transforming the software world to transforming the hardware world.

In the movie Iron Man, Tony Stark only needs to put forward a vague idea, and JARVIS can mobilize various devices to turn that sentence step by step into a wearable piece of equipment. The real-world engineering system is far less seamless. Human creativity must first be translated into a geometric model, then go through calculation, validation, processing, and assembly. If any link loses information, the final product may deviate from the original vision.

Over the past few decades, this industrial-grade high-precision access has been primarily realized through CAD (Computer-Aided Design) software.

Detailed CAD Diagram of a Jet Engine

CAD may look like a drawing software, but in reality, it is the physical foundation of modern civilization. Industries such as architecture, machinery, aerospace, shipbuilding, medical equipment, textile and apparel are all closely linked to it. It can even be said that CAD has accumulated the most industry know-how of humanity in the physical world.

As AI begins to enter the 3D space, the closer it gets to the physical world, the more precision and constraints it requires. The closer it gets to manufacturing systems, the more prominent CAD's gateway value will become. But here lies the difficulty of connecting AI to reality: how to re-implement general capabilities within the constraints of engineering problems, so that AI can transform the world faster and more reliably?

AI CAD: A New Gateway to Connect the Physical World

Traditionally, CAD software was essentially sold for engineers to use. But with AI capabilities evolving to their current level, CAD has gained a new type of user: machines.

This means that CAD's business logic is no longer just about enabling humans to draw more efficiently; it has also become a tool that allows AI to perform engineering operations more reliably. This is a transformative impact of AI on CAD, which not only changes the front-end user experience of the gateway but also reshapes value distribution.

When machines become new users, CAD transitions from a human-computer software that relies on interfaces and commands to an infrastructure that can be directly invoked by AI.

The underlying software needs to modularize its tools and provide interfaces, enabling AI to understand intent, break down tasks, and complete closed-loop operations. Furthermore, the business model will shift from per-seat licensing fees to billing based on tasks, invocation volume, and workflows.

Gong Minyan, Founder of Ziqian Technology, has a more direct summary of this: The previous generation of CAD helped humans project their intentions into the digital world. What Ziqian aims to do is the opposite — enable electrons to truly manipulate the world of atoms, forming a closed loop between digital design, physical validation, and real-world manufacturing.

Overseas giants are also approaching this goal along different paths. NVIDIA is attempting to build a training ground for Physical AI using Omniverse, simulation platforms, and its computing power ecosystem. Traditional industrial software giants like Dassault and Siemens have chosen to expand the boundaries of engineering by focusing on industrial software and digital twins.

Traditional industrial software giants have deep product portfolios, but their historical architectures are massive, making it difficult for these "giant ships" to turn easily. An AI-native transformation is bound to affect the entire product stack. Generative 3D companies excel at visual generation but lack engineering constraints and manufacturability. General large model companies have stronger intelligence, but their strong generality means they lack the necessary core tools and data assets within vertical industries.

As a team that has been deeply engaged in the 3D world for many years, Ziqian Technology believes that CAD lies at the intersection of the virtual world and the real world. It is not only a tool for humans to design products, describe spaces, and express engineering intent, but also an infrastructure that connects simulation, manufacturing, and physical execution.

What CAD models carry is not just three-dimensional shapes, but also physical principles and manufacturing processes. Compared with images and videos, it provides a digital representation that is closer to the essence of the physical world.

Driven by this philosophy, Ziqian Technology has developed a technical path called the Dream Loop, which spans generating the physical world, validating the physical world, and closing the loop on the physical world.

The first stage is "Agentic CAD" — where the system precisely acts according to user instructions.

With the reasoning capabilities brought by large language models, built on Ziqian Technology's cloud-native 3D CAD foundational tools, long-term accumulation of industrial data, and iterative evolution of its harness system, Agentic CAD can understand the engineering intent behind natural language, break down user goals into a series of executable modeling actions, and invoke reliable geometric modeling tools to complete the operations.

This process is similar to AI coding in the 3D world. It changes not only modeling efficiency but also the fundamental way humans interact with industrial software — shifting from humans learning the machine's operating language to machines understanding human creative intent.

The second stage is to move from physical structures to verifiable physical products.

As the core principle for realizing Physical AI lies in understanding physical laws and acting upon the physical world, at this stage, products will be validated using Ziqian's physical world simulation platform, which verifies the accuracy of a series of physical rules such as mechanics, kinematics, dynamics, and motion control.

This process can automatically perform closed-loop iterations with the product generation in the first stage, until the most reasonable and high-quality solution is produced.

The third stage is to advance toward manufacturing, fulfilling the ultimate goal of acting on the physical world.

The seamless operation of these three stages will truly reduce the time from the emergence of a human creative idea to its physical realization to one-tenth of the original, or even shorter.

For Ziqian, this is not just a simple cloud-based CAD software, nor is it merely a parallel efficiency improvement like Prometheus. It represents a precise alignment of AI with the physical world.

When more judgment and imagination can be left to human engineers, future CAD will become a brand-new gateway for human creativity.

Epilogue

The narrative of emerging industries often begins with a grand concept, eventually growing into a comprehensive industrial ecosystem.

Looking back at several waves of technological progress, changes in access points often determine the power structure of the industry. In the PC era, operating systems became the gateway for software development and hardware adaptation. In the mobile internet era, iOS and Android set the rules for developers. In the AI training era, CUDA transformed GPUs from a piece of hardware into an unavoidable computing platform for developers.

The connection between AI and the physical world will follow the same path.

On this path to reshaping the global industrial system, Jeff Bezos is not the only one who sees the opportunity. In China, there is also a group of cloud-native industrial software companies that started embedding AI capabilities into the real workflows of engineering and manufacturing even earlier. Today, the goal that all parties are chasing has become increasingly clear:

Whoever becomes the first stop for AI to enter the physical world will truly hold the right to define "JARVIS".

References

[1] Prometheus, Jeff Bezos' AI startup, is now worth $41 billion，Axios

[2] Jeff Bezos' AI Startup Prometheus Completes $12 Billion Financing, Valued at $41 Billion，Wall Street CN

[3] Nvidia believes physical AI systems are a $50 trillion market opportunity，GamesBeat

[4] Generative AI Market Poised to Reach $2.3 Trillion by 2032 as Agentic Systems Proliferate and Infrastructure Demand Surges, According to Bloomberg Intelligence，Bloomberg

[5] Digital Twin: From "Drawing the Skin" to "Drawing the Bones", Leading to Industrial Internet and Smart Cities，Yicai