The AI gold rush is moving into a new, more tangible stage. People have been talking about how well large language models (LLMs) like ChatGPT can hold conversations for the last few years. But the biggest companies in the business are already looking for the next big thing.
Alibaba Cloud just led a huge 2 billion yuan ($290 million) investment in ShengShu, a three-year-old Chinese business. This is a huge change in the tech world. But this isn't just another chatbot investment. It is a strategic gamble on "World Models," a kind of AI that can understand and mimic the laws of the real world instead of just guessing what the next word in a sentence will be.
As we get closer to 2026, it's becoming evident that text-based AI has its limits. The race to construct the "brain" for the next generation of robots and self-driving cars has officially begun.
The $290 Million Power Move: Breaking Down the Deal
Alibaba was not the only company that led the Series B fundraising round for ShengShu. TAL Education and Baidu Ventures also gave a lot of money. This comes after a 600 million yuan fundraising just a few months ago, which shows how quickly investors are moving toward physical-world simulation.
Vidu is a high-end AI video production tool that has continuously been in the top 10 in the world. ShengShu is the engine powering it. But Zhu Jun, the founder of ShengShu, sees video generation as just the beginning. The end objective is a "General World Model," which is a system that connects the digital pixels of a game to the real gears of a humanoid robot.
What is a World Model? (And Why LLMs Aren't Enough)
To understand why this investment is important, we need to know the main difference between the AI we use now and the AI that Alibaba is paying for.
The Problems with Big Language Models (LLMs)
LLMs are like master statisticians for language. They learn to guess what word should come next by reading trillions of words. They are great at writing poetry, coding, and reasoning, but they don't have "physical intuition." An LLM can read about glass breaking and know that it does, but it doesn't know what gravity, speed, or the glass's strength are.
The Growth of World Models
A world model is put together in a different way. It learns from input from other senses, like sight, sound, and even touch. It doesn't guess words; it guesses paths and physical results instead.
As ShengShu stated following the investment, "A general world model... more naturally captures how the physical world works than large language models.” By training AI on video and 3D data, developers are teaching the system the "common sense" of physics: how objects bounce, how light reflects, and how humans move through space.
The Holy Grail: Embodied AI and Robotics
What makes Alibaba so interested in physics? The answer is embodied AI.
For years, a "software ceiling" has made it impossible to have a robot in every home or a car that can drive itself. A robot that only works with an LLM might be able to advise you how to clean a kitchen, but it will have a hard time getting around the mess, picking up a fragile egg, or reacting to a pet racing across the floor.
Three big industries need world models to complete the puzzle:
Robotics: To move around in 3D space, humanoid robots need to be able to "see" and "plan." They can use world models to quickly test out a thousand possible movements and pick the safest one.
Self-driving cars need to be able to forecast not only where a pedestrian is, but also where they will be in three seconds depending on how fast they are moving.
Industrial Automation: AI needs to know how heavy machinery works in smart factories so that it can work safely with people.
Alibaba's investment in ShengShu is a direct way to deliver the cloud infrastructure and the basic models that will run these machines. Alibaba has released a specific model for powering robots in February 2026. This additional investment makes their ecosystem even stronger.
China vs. The World: The Video Generation Arms Race
This investment's timing is not a coincidence. There is a lot of rivalry throughout the world to be the best in AI video. While OpenAI’s Sora made waves with its realistic (though initially restricted) outputs, Chinese companies have been moving at an incredible pace.
ShengShu (Vidu): Vidu was released worldwide months before Sora became publicly available. It has concentrated on consistency and realism based on physics.
The people who made TikTok and Kwai, Kuaishou and ByteDance, have released their AI video tools. They trained their models using their huge internal datasets of short-form videos.
Alibaba has also put millions of dollars into Tripo AI and PixVerse, which are platforms that specialise on making 3D models and "directable" AI video.
By diversifying its investments across ShengShu, Tripo AI ($50M), and PixVerse ($60M), Alibaba is constructing a comprehensive physical AI suite. They are not solely relying on a single approach; instead, they are constructing a comprehensive suite of solutions.
The Kevin Kelly Perspective: The Three Pillars of Intelligence
Wired co-founder Kevin Kelly recently argued that for AI to truly replicate human intelligence, it requires a triad of capabilities:
- Reasoning: Provided by LLMs (the "Knowledge" element).
- Physical Understanding: Provided by World Models (The "Perception" element).
- Continuous Learning: The final frontier that is still being developed.
As Kelly notes, we have already conquered the "Knowledge" part. The current industry-wide pivot toward world models represents the quest for the second pillar. Without an understanding of the physical world, AI remains a "brain in a jar” brilliant, but unable to interact with reality.
What Happens Next? (2026 and Beyond)
The $290 million investment in ShengShu will probably speed up a few important trends that we should see by the end of this year:
1. Video that goes from "static" to "interactive"
Usually, AI video is passive. You give it a prompt, and it provides you a clip. With the next generation of world models, we'll see "interactive" film where people may change the physics of a scene in real time. This will make it hard to tell the difference between a movie and a video game.
2. The Commercialization of Humanoid Robots
Expect to see more "embodied" systems in commercial settings. Alibaba’s strategic partnerships with robotics firms mean we will likely see Vidu-trained "brains" appearing in industrial robots that can handle complex, unscripted tasks in warehouses.
3. The "China Shedding" Challenge
As noted in the original reporting, there is a complex geopolitical dance occurring. While Chinese VCs and founders are navigating "China shedding" (investors pulling back due to tensions), domestic giants like Alibaba and Baidu are stepping in to ensure that China remains a leader in the next phase of the AI revolution.
Final Thoughts: The New Era of Simulation
Alibaba’s $290 million investment isn't just about making cooler videos for social media. It is about simulating reality. We are moving away from an era where AI merely "talks" to us and toward an era where AI "understands" our world. Whether it's a robot helping in a hospital, a car navigating a snowy street, or a digital twin of a city, the foundations are being laid right now through the development of world models.
The limit of the large language model wasn't a failure; it was a signpost. It told us that language is only a fraction of human experience. By backing ShengShu and Vidu, Alibaba is betting that the future of intelligence belongs to those who can master the physical world, one pixel and one physics calculation at a time.