Capital Is Pulling Back from World Models

Yeng-yu was born in the temple of Zhengfei.
QbitAI is public.

It's a very exciting time.Start-up event in TaikoyuIt's probably just that the annual windshield world model is turning upside down.

In the heated debate of opinion, VC capital has become an ongoing topic of reflection outside the industry.

But underwater, the world model racetrack on which Koyu was betting, is flowing from the hot money of the first half of the year, to the second half of the year, a T.S. difficult to raise capital is indeed raising the threshold of the world model investment intention to the book, the hot money is less, and the world model entrepreneurs who get the money are also getting less.

Because it started this summer,It's not a world model.

This change has been particularly evident in the recent interactions with a number of leading investors, including the heads of investment financing at World Model Companies.The world model project that the head of the fund is now willing to undertake can be counted on the fingers of one hand.

We asked if there was a better way to fund a company.

Yes, families have them.

One of the most promising investors this year has told us this.The names of four or five founders.

Thinking about it, he added to a family of smart startups that were part of the world model because he was the younger brother of the factory and was only a brain.

Is there anything else?

You know, you've almost forgotten these things.

Surprisingly, the names given by other respondents did not exceed this short list.

Some of the world's leading model firms, which are responsible for the financing of humanity, feel even more strangely the same.

It's getting harder and harder to get money.

There is no denying the technological prospects of the world model, and some investors are still actively looking for the next company worth investing in.

But the situation in the first half of the year has definitely changed significantly.The four-word generic valuation premium of the world's silver model is shrinking rapidly, shrinking extremely rapidly.

The window of conceptual and imaginative spatial financing is narrowing, and the teams that can extract technical evidence, accumulate research, and even real-world application progress are still rushing in.

As for the others...

The patience of the market began to become limited. Everyone could not escape a bit of anxiety.

Extreme cooling

In fact.From the beginning of the hot summer, the wind direction changes.

Until the WAIC in mid-to-late July of this year, the world model was undoubtedly the hottest, most sucking red-fried chicken.

There's a half-year-old trick waiting for a collective porcupine screen on WAIC.

A large number of startups are releasing new models to showcase technological progress and demonstrate demos, and some systems allow users to explore freely in AI-created environments, like playing games.

For the general public, of course, it is fresh enough.

ButFor those in the industry who have seen dozens of projects, the freshness is fading rapidly.。

The more you see, the more the problem (or even the common problem) becomes apparent. So after the end of the WAIC and the WRC, the previously hot-rolled projects started to cool down.

However, to tell the truth, this cooling should not be a complete surprise that the world model company and the world model concept are linked.

In the first half of this year, the capital itself has been a bit overwhelming.

Many investment firms were worried about missing the next star company, and the layout was completed at a very early stage.

In order to bet on AI, a leading RMF has more start-ups in the first half of the year than any similar project in the decades since its inception.

In the first half of the year, one of the top dollar funds has spent a ton of money on AI investments for the entire year of 2026.

In a short period of time, continuous investments, old shareholders' combined bets, and the merger of three rounds and one round at a higher price, further boosted expectations for the entire direction of the market.

In addition,Some US funds enter a new round of fundraising(e.g. Red Cedar, High Cedar, Boreham, Enlightenment, Monolith are raising money, Starlink, Linear, Morningstar are announcing the first round of closing of a new US dollar fund in the second half of the year.) This will inevitably reassess the pace of capital reserves and investment.

It's about this time.The secondary market also taught us a lesson.

On 19 August, Yuri Tech launched a research project.

On its first day on the market, the share price of Utrecht once soared to 1,100 yuan, with a market value of about 444.9 billion yuan.

Although there was a sharp rebound in late October, this cross-country car is a wake-up call for investors in the smart field.

Utrecht is not the world's model company.

But you can't deny thatFor institutions that are pricing projects related to physical intelligence and AI, U-tree provides an unavoidable open market reference.

Even the leading robotics companies with a product and a commercial base could face such drastic revaluation adjustments, and how much would it cost to model companies that have not yet demonstrated scalable application value?

This account, investors began to count more carefully.

Some may wonder, opening up to the financing news in the second half of the year, that the world model seems to be still hot, that money is still coming in from the source.

There's actually a time-lapse in it.

Financing official announcements are often later than the actual transaction.It's a common practice in the industry to put pressure on the outside world for three months.

As a result, the pace of capital markets has slowed, but transactions completed in the first half of the year are continuing to become the financial news of the second half.

While it's true that there are companies that are making money at a disadvantage.Nuclear power is infinite.

In August, the furniture body brain company announced the completion of nearly 1 billion yuan A round and A+ round financing. At the end of October, it also disclosed the acquisition of nearly 1 billion yuan A++ round investment by Zhejiang Provincial startup investment group.

Quantibit learns that the companyIt's a unicorn.In the meantime, a new round of financing has been launched.

However, it is difficult to represent the temperature of the market as a whole.

According to FSA's investment data, according to incomplete statistics, a total of five first-tier market financing events related to physical AI and world models in October have significantly decreased.

Even with the introduction of physical AI companies, there has been a marked scarcity of market financing.

In this way, the changes in the global model investment financing market in the second half of the year are becoming increasingly clear.

Some companies are still making money, and one is big money. But more projects are not getting the treatment that investors were chasing in the first half of the year.

But you will say that the world model has cooled down, lest there be some injustice done to the people.

Highly Differentiated

At a time when many projects are saddened by the next round of financing, few companies are still getting the money fairly well.

And that's hundreds of millions of dollars at a time.

What companies are treated like this?

Looking back at the list given by the investor who opened it, you can find some doors.

Let's see the money.

In July this year, Natural Will, founded by Professor Dingning, an assistant professor at Tsinghua University, and after completing the funding in April, introduced investment groups such as Red Cedar China, Liu Group, Ningde Era and others.

In September, LiberAI, which was founded by Liu Xiaoping, announced the completion of a multi-million-dollar Pre-A++ round of financing.

In addition, the company, founded by Phnom Penh, has completed several rounds of financing in just a few months and is valued at hundreds of millions of dollars.

In June, the inverse matrix, founded by Northeastern University graduate Chen Bo-yong and Jingjia-ming, officially announced over a billion dollars in seed+ rounds of financing. Qubit recently learned that a new round of inverse matrix financing has already been delivered, but has not yet been disclosed.

The pace of the financing seems to be a bit fractured by the cautious attitudes of the investors before, but China, North Korea, MIT... can probably get a taste of it.

Chen Boyi founded the reverse matrix at the age of 22, and has previously served as the head of the Center for World Model Innovation in Behavior at the Institute of Intelligent Studies; Dingning is an assistant professor and a doctoral student instructor at Tsinghua University; he received scholarships from Tsinghua University during his undergraduate studies, and later went to MIT to do post-doctoral research; Yang Matsunming is also a Tsinghua Prize winner and a post-00 entrepreneur.

It's a great way to get to the top of the university, young people, and research.

In the first half of the year, when the world model was heating up, this type of configuration was the hottest. What happened to the second half of the year when investors started selecting carefully?

The answer is that a few outliers can get the money.

In addition to the history of the founders, each of these companies has achieved something else.

For example,The reverse matrixThe Physis series of models has been released, attempting to allow AI to learn the state of the physical world and further predict and deploy.

Materials and technologyGeoPT, previously published by its founder Liu Bei, explores how to learn the laws of motion of objects from static geometric information, with no small impact on the circle.

Both teams are studying how AI understands the physical world, just starting off on different technological paths.

The other two intersections are not exactly the same.

The natural willNatura-1, released in October, attempts to use over a million hours of human first-person data to teach robots how to manipulate humans.

LiberAIThis week's release of the world's action model, Liber-0, puts the same stakes on the massive human experience, hoping robots will be able to migrate the learned operational capabilities to more tasks.

There's a special one on the list.

What's the matter with you?

It doesn't have a story of Qingdao's small-scale entrepreneurship, but the family is a big factory boy. It has introduced several models of the LingBot series, as well as computing, engineering and application scenario resources that are different from startups starting from scratch.

The overall investment financing logic is somewhat blurred.

The conceptual dividend of the world model is declining, but there is still a chance to make money for those who have already achieved results in relevant research directions, who have accumulated special technology, or who are able to mobilize large amounts of R&D resources.

Much of this money is still betting that the technological routes for the future world model have not yet been crossed, and that products and business models are still being explored.

In the last six months, companies that have no models, no products, no concepts, and no slices, all of which we are familiar with, have been getting money.

Unbelief in another controversial young entrepreneurWhat are you doing?。

……

Looking back, in the first half of the year, companies were busy proving that they were the world model, even companies that didn't start with the world model were trying to prove that they could be the world model.

In the second half of the year, investors began to select carefully.

We've been challenging and challenging top universities, star researchers, and young academics.

Writing this really made me laugh a little bit, and it sounds like the story finally turned into a crazed little genius.

But isn't that in line with the investment logic?

This is not to say that more substantial geniuses will be buried.

The last thing I want to emphasize is this list.It's just a slice of the market that Qubit got in this interview.This is not to say that these companies have won, nor is it to say that they have no authority.

But I believe that the mainstream attitude of the investment market can be seen from it.

Concept of watering with flags on the shore

In retrospect, this year's world model could be top-of-the-line, and for good reason, of course.

Language models with physical intelligence are evolving, but AI's ability to understand, predict, and interact with the physical world remains a challenge.

If the model can build an internal understanding of the laws of the world and predict the results of different actions, the ability of AI to enter the physical world could take a new stage.

This is why the world model is so promising.

Moreover, researchers such as Li Feifei, Li Li Quan, and tech giants such as Google and InVida are also working on the technology.

The surge of capital with physical intelligence has provided a huge space for application of the world model.

Research breakthroughs, field imagery, star entrepreneurs, giant investments, robot industry demand, several forces come together and the market quickly warms up.

To put it bluntly, this is the kind of technology that has gotten a lot of companies into big money, and the industry has yet to come up with a unified definition of it.

Producing realistic videos, simulating three-dimensional physical states, and predicting robotic movements, although they can be part of the study of world models, differ widely in their solutions and evaluation criteria.

These differences can easily be overshadowed by a grand technological concept at a time when the world model is hot to the eye-catching turn.

At the same time, although the boundaries of concepts are not yet clear, new definitions can be extended.

The model of the physical world, the model of the spatial world, the model of the causal world, the world model of action...

One word is not enough, add one.

Instead of one adjective not highlighting the peculiarity and the avant-garde, add one. This is not done to separate the racetrack, but simply to emphasize the peculiarity and the avant-garde.

It was then that the whole world began to be modeled after the world.

In June, Li and his startup company, World Labs, published an article on the functional classification of the world model in an attempt to create a clearer classification system for this widely used concept, dividing it into three categories: renderer, simulator, planner, and simulator.

But it doesn't seem to work.

However, this is not the end of the vacuum-seeking world model.

On 11 October, the Wall Street Journal reported that the newspaper was in a state of disrepair.Amazon is in talks to acquire Decart, the world's leading model startup, for about $7 billion.

Decart has previously developed the real-time video editing model Lucy and Oasis, which can generate simulated environments.

In Amazon's ledger, the company's appeal is not only to generate the model itself, but also to its technical capabilities for AI infrastructure and cross-chip operation.

We were very impressed by the reaction of one practitioner:

It's a real feeling to see the sunset on the beach.

The sun will not set... a little magnificent, a little sad, and occasionally a little warm.

One More Thing

At the end of September, AMD announced plans to acquire World Labs, the world-model company founded by Li Feifei, for $8.2 billion.

For a startup that's been around for more than two years, $8.2 billion is a pretty impressive acquisition bid.

But some people in the world of modeling aren't happy.

Lifeifei has a world-leading research reputation, World Labs has a model R&D team, technology accumulation and financing capabilities, and is home to a chip giant AI infrastructure layout.

For AMD, the value of the transaction also includes access to cutting-edge model research capabilities to help the company understand future AI workloads and optimize chips, software and systems.

Even Li Feifei, one of the two big players, and her team, who finally chose to join AMD for $8.2 billion, was it a little too big to imagine the value of the world model before that?

Some people have even questioned Lifeifei's start-up and the willingness to choose to be acquired...

In any case, the event became a nod to the world model.

It could be the start of a round of raffles, or it could be the turning point of the era.

The world model has been around for less than three years, but it obviously needs to be more ground-breaking, and it needs a well-thought-out story from proven value.

Who could have imagined that the championship would be so fierce that autumn would come almost overnight?