Do robot companies with more amazing demos die faster?

Recently, there have been reports that a company specializing in humanoid and quadruped robots has seen its technological capabilities stagnate for a long time, and its core R & D personnel have successively left. Inside the organizational structure, it is common for salespeople to manage R & D.

Those who understand know that once a company shows signs of laypeople managing experts and the departure of core personnel, it is usually not a good sign. This often means that the team is using business rhythm, financing rhythm, and external narrative rhythm to suppress the technological rhythm.

But what if this is the common situation of many robot companies today?

01

Many companies have to make survival their top priority. Under this survival logic, the value of demos seems to be infinitely magnified.

No one can deny the historical value of demos in the early stage of the industry. In fact, it can be said that without demos, 90% of robot companies wouldn't even be able to get their first round of financing.

In the past two or three years, demos have been the most valuable thing in the robot industry. After all, robots are highly physical entities. Without raising funds in the early stage, many things simply can't move forward. Hardware costs money, computing power costs money, data costs money, testing costs money, trial - and - error costs money, the supply chain costs money, and mass production costs even more money.

Especially humanoid and quadruped robots are naturally suitable for visual communication. Once they can run, jump, grab, and perform simple actions, the communication effect will be very direct. People in the industry will think, "Wow, this thing can really move."

Boston Dynamics is the pioneer who has mastered the demo logic most thoroughly. In the early years, it demonstrated the capabilities of multiple robots such as BigDog and Atlas in traversing complex terrains and maintaining dynamic balance through videos that spread across the Internet. It not only gained cooperation opportunities in the scientific research and military fields but also became the core bargaining chip for Google to acquire it for $515 million.

In contrast, domestic companies like Unitree maintain high exposure today. Essentially, it is because demos still have communication power.

This communication logic is especially effective for companies that want to seize the overseas market.

Platforms like YouTube and Reddit naturally prefer this kind of content: strong actions, high stability, and relatively low cost. The videos are easy to understand at a glance. There is no need to explain the algorithm, the architecture, or how many models, layers of control, and steps of planning are used.

Overseas users may not care about the underlying technology. They are more likely to repeatedly watch robots do somersaults, punch, kick, or even kick their own boss. It first establishes a sense of presence.

The logic of AI agents is actually a bit similar. Many times, when facing a problem, people first create a demo to see, first establish visibility, and first make the market react to their existence.

It's just that agents are lighter, the trial - and - error cost is lower, and demos can be iterated more quickly. They can even be repackaged using skills.

However, it's obvious that robots can't rely on demos to survive for long. This may be why the robot industry is more prone to bubbling and more likely to expose the truth than the current AI.

02

Now, the path of many embodied intelligence teams is actually quite fixed -

Most of them come from a hardware background. First, they make their self - developed hardware work; then they build a pipeline for data collection, training, and deployment; then they use open - source models to target scenarios, data, and tasks; create a presentable demo; take on some simple scientific research orders or demonstration - based deliveries; and finally gradually shift the focus to financing.

This path itself is not strange. In fact, it can be said that it is almost an inevitable path in the early stage.

The problem is that when this path finally becomes a dead - end cycle of "telling stories - creating demos - financing - telling more stories", it no longer seems like a product path but more like a narrative path that is extremely dependent on external factors.

At this time, the core capabilities of the company will quietly change. R & D personnel become those who work for the financing goal, and technology is no longer the root of the company's growth but gradually becomes a prop for the company's valuation.

If sales start to dominate the R & D direction, it usually also indicates one thing: the company will give priority to answering "what is easier to sell, what is easier to finance, and what is easier to talk about".

Today, many embodied intelligence companies, after running through a scenario, will quickly package it as their technological barrier. This action is understandable, but the rhythm often seems a bit hasty.

Because many people in the industry actually know that a large number of so - called embodied projects on the market are essentially hardware plus pipelines, combined with open - source models, and finally create a locally over - fitted demo.

Many so - called mass - production routes essentially involve over - fitting a small end - side model to a specific scenario first; if it's not enough, add another layer of real - machine RL for further over - fitting.

Is it wrong? After all, it can run, be delivered, and make customers find it a bit interesting.

But it doesn't equal general capabilities and is still far from true generalization.

Only when you reach the top of the mountain can you see if there is another mountain ahead. In the history of the robot field, such mountains have always existed. Too many teams mistake running through a laboratory demo for reaching the technological peak, without realizing that this is just the first threshold to the real world.

Some people may start to ask, why are investors still willing to continue investing? Many people laugh at investors for investing so much money in these companies that only know how to create demos.

Many times, what investors value is not whether general intelligence will be realized soon but whether this track can continue to tell stories, continue to increase in price, and allow them to exit at an early stage.

But this doesn't mean that investors are unaware of the risks.

On the contrary, they often know very well that the profit model of early - stage projects doesn't rely on the long - term success of the company. What really matters is whether the story can be taken over at a higher valuation in the next round, whether the track can maintain its popularity, and whether the company can survive until a more valuable stage.

So, we can see a very magical phenomenon: everyone knows that general intelligence is still far away, everyone knows that most of the current demos are for show, but everyone tacitly continues to push up the valuation.

Because as long as the track remains popular and there are still new investors willing to take over, this game can continue.

03

But the premise of this game is to have a continuous stream of new funds.

However, when there are projects on the street that dare to seek financing just by fine - tuning open - source models, once large companies like Huawei, Xiaomi, and BYD, which have their own supply chains and scenarios, integrate and optimize open - source solutions and implement them in batches, the companies that rely on hardware assembly and open - source fine - tuning won't even be able to afford a price war with the real money from investors.

Even Boston Dynamics hasn't escaped this curse. To this day, it still hasn't found a path for large - scale commercialization. Its senior management has left en masse, and the number of truly mass - produced robots can be counted on one hand.

This is actually the cruelest and most real law in the primary market: there is no eternal trend, and all tracks have their own shelf - life of popularity.

For concept - based tracks like the metaverse and Web3, the real golden window period is usually only 1 - 2 years; for heavy - asset physical industries like semiconductors, photovoltaics, and new - energy vehicles, the market cycle they can maintain is only 3 - 4 years.

An investor predicts that if he remembers correctly, embodied intelligence began to explode in the market in the second half of 2023, and it has been almost 3 years now.

According to the general popularity clock of this industry, it has reached a critical node for clearing bubbles and returning to value.

Looking at the robot industry today, I prefer to understand it as a track still in the screening period.

Demos are still important, communication is still important, and the financing rhythm is still important. But what really determines who can go far is ultimately technological accumulation, organizational ability, and long - term delivery ability.

Demos will never die. They will always be the first step in technological innovation and the best way to verify ideas, educate the market, and attract talents. In the future, truly excellent technology companies will still create amazing demos.

But the business model driven purely by demos has come to an end.

Although the current wave of financing enthusiasm has not subsided, with the arrival of the listing wave of leading enterprises, the industry differentiation is likely to accelerate in the second half of the year.

Those that won't become cannon fodder are either companies with a stable core R & D team that is not just a porter of open - source models;

or those that focus on a specific vertical scenario, such as industrial inspection, logistics handling, and agricultural picking, and solve a problem to the extreme;

or those that can cross the mass - production chasm and control the yield and cost;

or the rarer self - sustaining players that can generate revenue and don't rely on financing.

Soon, we will witness the first real cleansing of the robot industry.

This article is from the WeChat official account "Zhijian autoweek", author: Xiao W. Republished by 36Kr with authorization.