NIO, XPeng, and Li Auto should have a separate table for their chips.
While having a meal and scrolling through my phone, I saw a piece of news.
At Xiaomi's Investor Day, Lei Jun said that the shipment of Xuanjie O1 chips had exceeded one million units. Right after that, he added that Xiaomi's future cars would also use self-developed chips.
My first reaction at that time was: Oh, it's news from the end of April. Then I scrolled past it.
But algorithms are such that the more you look at something, the more they push similar content to you. Soon after, I saw news about the launch of NIO's LeDao L90, which is equipped with a self-developed chip called "Shenji". Further down the feed, I learned that XPeng's Turing chips have a shipment target of one million units this year, and all its models will switch to using them.
Seeing this, I couldn't help but laugh, and a question suddenly popped into my mind: Why are all these car companies getting into the chip business?
I did a quick search and was quite surprised. Li Auto's Mach 100 chips are in mass production this year and have already been installed in cars. BYD is also promoting its own intelligent driving chips. It seems that all these car companies are involved in the chip business.
Out of habit, I opened the semiconductor fund I bought and checked its holdings. The top ones were North Huachuang, SMIC, Cambricon, Hygon Information, and Changchuan Technology.
Staring at these names for a while, I felt something was off: Why aren't any of the car companies I just saw making chips, like NIO, XPeng, BYD, and Xiaomi, in the list?
I checked further and found that it's not just the fund I bought. In the major semiconductor ETFs in the market, the heavyweight stocks are basically "purebred" semiconductor companies such as equipment manufacturers, foundries, and AI chip companies.
The chip business of car companies is not among the core holdings of any semiconductor fund, creating a misaligned picture.
In the first half of 2026, most of the biggest news in the semiconductor field, except for Huawei's recent splash, came from NIO, XPeng, Li Auto, BYD, and Xiaomi.
They are raising funds, mass-producing, shipping products, and spinning off independent companies. NIO's chip subsidiary, Shenji, raised 2.257 billion yuan in February this year, and its valuation is approaching 10 billion yuan. XPeng's Turing chips have already been installed in Volkswagen's cars.
These things are truly reshaping the semiconductor industry in terms of "who matters".
But what about the fund in my hand? It remains the same. It's labeled as a "semiconductor" fund, but it still holds traditional semiconductor companies as defined in the past.
To be honest, this is quite misaligned.
I think the root of this misalignment lies in the fact that the entire industry classification system may not have caught up with the actual changes. The fact that car companies are making chips is already a reality in the industrial sector, but it has not been re - classified in the capital market.
I guess when you buy a fund, you just buy it when you see the word "semiconductor". I did the same. The problem is that what these three words represent has changed a lot compared to a year ago.
……
What's the difference? First, we need to figure out who car companies used to buy these chips from.
The answer is not complicated. In the field of intelligent driving chips, NVIDIA has dominated in the past few years. The third - party supplier with the largest sales volume in China is Horizon Robotics.
Horizon Robotics went public on the Hong Kong Stock Exchange in October 2024, claiming to be "the first Chinese stock in the intelligent driving chip field".
There was quite a stir when it went public. Yu Kai talked about the company's vision and the goal of making the Journey series of chips among the top three in the world. The capital market also bought into it, and its market value once soared to over 100 billion Hong Kong dollars.
I looked through its annual report for 2025. On the surface, the numbers looked quite good.
Its annual revenue was 3.76 billion yuan, a nearly 60% increase. It shipped over 4 million sets of chips, and the cumulative shipment exceeded 10 million units. It had partnerships with over 40 car companies, and its list of customers was quite long.
But on the next page, it was a different story.
It ended up with a loss of 2.81 billion yuan, nearly 70% more than the previous year. It spent 5.15 billion yuan on R & D, 1.37 times its revenue. What does that mean? For every 1 yuan it earned, 1.3 yuan was spent on R & D.
Another eye - catching figure is the gross profit margin. It was 77.3% in 2024, but dropped to 64.5% in just one year, a nearly 13 - point decrease.
When the annual report was released in March, the stock price dropped by 3% on the same day. Compared with the high point in the past six months, it has dropped by over 20%. In April, news spread that Chen Peng, the core R & D leader of the chips, had left the company, and Vice Chairman Chen Liming was also gradually fading out.
For a company with nearly a 60% revenue increase, its stock price dropped by 20%, and its core team was loosening. What is the market worried about?
I did some calculations and it all made sense.
The top five customers of Horizon Robotics contributed more than half of its total revenue. Who are these customers? NIO, XPeng, Li Auto, and Leapmotor, a group of leading new - energy vehicle startups.
Now, let's see what these customers are doing.
NIO spent over 300 million US dollars on NVIDIA's Orin chips in 2024, but it doesn't buy them anymore. Its self - developed Shenji NX9031 chips are now installed in all its models, and the cumulative shipment has exceeded 250,000 units.
Even more impressively, Li Bin revealed in the earnings conference call in March this year that the second chip of Shenji has successfully completed the tape - out and is in mass production. It uses a 5 - nanometer process, has a performance equivalent to three Orin X chips, costs much less than the first one, and is positioned in the mid - end market.
This chip is targeting the main battlefield of Horizon Robotics' Journey 6 series.
As for XPeng, over 200,000 Turing chips have been shipped. Starting from the second quarter of this year, all its models will switch to using them, with an annual target of one million units.
He Xiaopeng himself has said that this chip can be sold to other companies in the future. Volkswagen's cooperative models are also waiting for mass production this year.
For Li Auto, the Mach 100 chips have been installed in cars this year. The combined computing power of two chips is 2560 TOPS. As Li Xiang put it, it's equivalent to three times the effective computing power of NVIDIA's Thor - U. It sounds ambitious, but they have actually achieved it.
BYD has a different approach. It has already established its own system in power chips, and its self - developed intelligent driving chips are also in the works and are planned to be installed in cars in the second half of the year.
Putting all these pieces together, the picture becomes clear.
Horizon Robotics' former best customers are gradually becoming its competitors one by one.
This is not the same as the general situation of "more intense competition". When an external competitor enters the market, at least you know who they are and where they come from. But when your customers turn against you, the impact is much more severe.
Why? Because they hold what you want the most, and they know what the market demand looks like.
NIO knows better than anyone what an autonomous driving chip should be like because it is constantly running data and adjusting algorithms. The chips it designs are naturally more in line with its own needs than those bought from outside.
Horizon Robotics is not sitting idle. I checked and found that on April 22nd this year, the day before the Beijing Auto Show, they launched a new chip called "Xingkong".
It uses a 5 - nanometer process, has a computing power of 650 TOPS, and its core selling point is cockpit - driving integration, which combines intelligent driving and intelligent cockpit functions on a single chip. The first - batch mass - production customers include BYD, BAIC, Chery, and Changan, and it is planned to be installed in cars in the third quarter of this year.
Yu Kai should be aware that the chips self - developed by car companies are currently mainly focused on pure intelligent driving. Horizon Robotics is moving into a category that car companies have not fully covered, trying to gain an edge.
It is also expanding overseas. J.P. Morgan predicts that the overseas shipment volume of Horizon Robotics' HSD solution will reach 460,000 units in 2026, more than 17 times that of 2025.
So, Horizon Robotics won't "die", but it is going through a forced transformation of its customer structure. Since its top customers are all self - developing chips, it has to look for mid - tier brands and overseas markets.
Its business model has also changed, from "serving the best customers" to "serving customers who don't have the ability to self - develop chips".
This incident itself is the first bomb thrown by car companies' chip - making efforts.
Your semiconductor fund may not hold Horizon Robotics' stocks as it is listed on the Hong Kong Stock Exchange, but this bomb is not just affecting Horizon Robotics. Any third - party automotive intelligent driving chip company is facing the same problem: your customers are becoming your competitors.
This situation has actually happened before in the mobile phone industry fifteen years ago.
……
In 2010, Apple put its self - designed A4 processor into the iPhone 4. That was Apple's first self - developed SoC. Before that, iPhones used Samsung's chips.
At that time, few people paid attention to this decision. A mobile phone company making its own chips? It sounded like a sideline business.
Fifteen years later? Apple's A - series chips have evolved to the A19, and its M - series chips have driven Intel out of Macs. In February this year, Apple also installed its self - developed baseband chip C1 in the iPhone 16e.
What does this mean? Apple is also starting to break away from Qualcomm in the 5G baseband field.
Is Qualcomm dead?
No. I checked, and Qualcomm's revenue in the third quarter of 2025 was 10.3 billion US dollars. It's doing quite well, and Apple still contributes about 20% of Qualcomm's revenue.
Hearing this, you might think: The same goes for car companies making chips. Horizon Robotics won't die, and NVIDIA won't die either. The market is big enough for everyone to survive.
That's not the case. Qualcomm is still alive, but it's not the same Qualcomm as it was ten years ago. After Apple self - developed its SoC, Qualcomm lost its biggest customer in the mobile phone chip market. How to fill this gap? Where to fill it?
I looked through Qualcomm's financial reports in recent quarters, and the answer is clear: Automotive. In the fourth quarter of 2024, Qualcomm's automotive business revenue increased by 61% year - on - year, far exceeding the 13% growth rate of its mobile phone business.
Apple forced Qualcomm to find a new way out, and that way out happened to be automotive chips.
Looking at Horizon Robotics again, since leading car companies are self - developing chips, it is forced to target mid - tier brands and overseas markets. Yu Kai's promotion of the cockpit - driving integration chip is essentially a search for a new category that car companies have not covered.
It's a case of misaligned competition, which is quite normal. The logic is exactly the same. Customers' self - development won't kill the suppliers, but it will force the suppliers to transform into different companies.
So, the question is: Will car companies become chip companies? I dare not even think about it.
So, I did a search and was quite shocked. NIO's Shenji is no longer just for its own use. This year, it has officially started to supply chips and supporting solutions as a Tier 1 supplier and is in talks with Leapmotor and Geely for cooperation.
What's the scale of Leapmotor?
It delivered nearly 600,000 vehicles in 2025, and its target this year is 1.05 million. If the cooperation is successful, the scale of Shenji chips installed in cars will be larger than NIO's own sales volume.
He Xiaopeng once said: In the future, all top AI companies will self - develop chips. He also did some calculations and said that in the past ten years, automotive hardware accounted for over 90% of the total vehicle value, but in the next ten years, it will shrink to 50%, while software will increase from 10% to 50%.
You see, these people no longer consider themselves just car companies.
NIO is supplying chips to other car companies, but it still calls itself an automotive company on its business card, yet what it's doing has crossed that line.
Think about it. Apple has never publicly said "I want to kill Qualcomm", and NIO has never said "I want to replace Horizon Robotics". They are just solving their own problems, aiming to regain control over the product rhythm.
But what's the result? Qualcomm is forced to enter the automotive chip market, Horizon Robotics is forced to look for mid - tier customers and overseas markets, and NIO itself has become a chip supplier. No one planned to become a different type of company, but everyone is changing.
Okay, let's look at the fund issue. I think it can be divided into two levels:
No matter who designs the chips, they all need to be manufactured by foundries, and they all need to buy EDA tools and use semiconductor equipment.
Companies like North Huachuang and SMIC in the fund are essentially selling shovels. The more people are "gold - digging" (making chips), the better the business of selling shovels. At least for now, the entry of car companies into the chip - making field is a positive factor for this level.
What's being squeezed is another level. Third - party intelligent driving chip companies like Horizon Robotics are experiencing a transformation of their customer structure, and their valuation logic is also changing. From supplying chips to the best car companies, they are now "supplying chips to car companies that don't have the ability to self - develop". These two scenarios are worth different amounts of money in the capital market.
When we buy semiconductor funds, we just see the word "semiconductor".
These three words were quite clear a year ago. Chip - making companies were doing their own thing, and car - making companies were doing theirs, with no interference between them.
Now, the line is blurred. Car - making companies are making chips, chip - making companies are losing customers, and the "shovel - selling" companies welcome everyone. Under one label, there are companies in completely different situations.
I'm not saying there's a problem with the fund. The "shovel - selling" level may even have an easier time. What I'm trying to say is that the word "semiconductor" can no longer summarize everything as it used to.
I just found these news items I saw today quite interesting, so I wrote this down.