2026 can be described as a "breakout year" for Chinese automakers' self-developed autonomous driving chips.
From BYD's 4nm "Xuanji A3" kicking off scaled mass production, to Li Auto's stunning debut of the 5nm "Mach M100"; from NIO's "Shenji" delivery volume surging past 250,000 units, to XPeng's "Turing" securing a production nod from Volkswagen:
The core supply chain of China's intelligent vehicles is entering a brand-new era, shifting from "purchasing external brains" to "independent definition."
Undeniably, with domestic autonomous driving chips being densely integrated into vehicles, chips have been elevated from supporting roles in the supply chain to strategic commanding heights. Meanwhile, the reality of a 180% surge in memory chip prices has made the supply chain security crisis increasingly imminent.
Collective Entry: The Underlying Logic of Automakers Making Chips
The wave of automakers making chips did not just start this year, but the concentrated explosion in 2026 marks the transition of this trend from "testing the waters" to the "harvest period."
Looking at the timeline, NIO's layout is the earliest and the deepest. In 2025, NIO launched its self-developed Shenji NX9031, defining it as the "world's first automotive-grade 5nm autonomous driving chip" with computing power exceeding 1,000 TOPS.
Subsequently, NIO spun off its chip business into an independent subsidiary, Shenji Technology, and formed a joint venture with AXera and OmniVision to launch the M97 chip (with over 700 TOPS of computing power) targeting a broader market, actively reaching out to automakers like Leapmotor and Geely to seek external supply.
According to information from NIO's Q1 2026 earnings call, the Shenji NX9031 has delivered over 250,000 units. Ma Lin, Vice President of NIO, stated frankly at the Chongqing Forum in June that during peak years, NIO's annual procurement of NVIDIA chips reached $300 million, while self-developed chips "have already saved the company a lot of money."
XPeng has also carved out a unique path. The Turing AI chip released in 2024, touted as the "world's first multi-terminal general-purpose chip," is not only equipped across XPeng's entire vehicle lineup but also bridges three major terminals: intelligent vehicles, flying cars, and humanoid robots.
More notably, the Turing chip has secured a mass production nod from Volkswagen. Meanwhile, CARIAD, the software company under Volkswagen, has also established a joint venture, CARIZON, with Horizon Robotics, investing $200 million to self-develop the high-computing-power autonomous driving SoC (System on Chip) chip C7H. XPeng Group has entered over 60 countries and regions globally, and the "external output" of the Turing chip is opening up a second growth curve.
Li Auto, on the other hand, completed the leap from "skepticism" to "debut" at Livis Day on June 15. CTO Xie Yan defined the Mach M100 as the "world's most powerful AI (Artificial Intelligence) chip," featuring a 5nm process, 1,280 TOPS of computing power per chip, paired with the self-developed Mach VLA intelligent driving assistance system, officially announcing a reaction speed 40% faster than humans.
Holding the chip at the launch event, Li Xiang said, "Take a picture of me, otherwise the internet will only have pictures of me lifting tables."
Behind this joke lies Li Auto's strategic intention to treat chips as the core carrier of the brand's technological image.
BYD's entry is even more symbolic. As the world's best-selling NEV (New Energy Vehicle) automaker, BYD released the Xuanji A3 autonomous driving chip in May. With a 4nm process, a single chip delivers over 700 TOPS of computing power, and three chips working in synergy exceed 2,100 TOPS, supporting L3 and L4 autonomous driving. BYD's scale means that once self-developed chips are deployed at scale in vehicles, the marginal effects of cost amortization and supply chain security will be highly significant.
So, why are automakers determined to make their own chips?
The most direct driving force is cost. Taking NIO as an example, the annual expenditure of $300 million on procuring NVIDIA chips will only rise as sales grow rapidly. Although self-developed chips require huge upfront investment—the R&D of a single high-computing-power autonomous driving chip often costs over a billion RMB—once mass-produced and deployed in vehicles, the per-vehicle chip cost can be significantly reduced. More importantly, self-developed chips can customize architectures based on their own algorithm requirements, avoiding the waste of computing power caused by purchasing general-purpose chips, and achieving deep collaborative optimization of "hardware-software integration."
Secondly, supply chain security is a consideration that cannot be ignored. Geopolitical uncertainties have turned "chip supply disruptions" from a theoretical risk into a realistic anxiety. When a single intelligent vehicle is equipped with up to thousands of chips, a supply interruption of any single one could lead to a production halt. Automakers developing their own chips are building a security barrier for themselves at core links.
The deeper logic lies in the shift of the competition paradigm. The competition in intelligent driving has moved from the stage of "having it or not" to "how good it is," and the underlying support for "how good it is" is computing power and chips.
An automotive executive once stated frankly: "The next competitive focus is chips." This statement is being recognized by more and more peers—whoever masters chips masters the initiative in defining products.
However, automakers making chips is not without controversy.
On June 13, at an industry forum, the head of an automaker's chip business publicly stated: "Developing a chip, especially a high-computing-power chip, requires an investment of over a billion RMB. Automakers must anchor their own positions, considering how much their installed base is and what their intelligent penetration rate is."
Meanwhile, the head of another automaker was even more direct, explicitly stating, "We are not making chips anymore."
These rational voices remind the industry: making chips is not a "mandatory question" but a "multiple-choice question," and the answers vary from company to company.
Computing Power Race: The Ongoing Domestic Substitution of Autonomous Driving Chips
Behind the wave of automakers making chips lies an industrial competition landscape that is being profoundly reshaped.
For a long time, the in-vehicle autonomous driving chip market has been monopolized by overseas giants such as NVIDIA and Mobileye. NVIDIA's Orin-X has almost become the "standard configuration" for high-level autonomous driving. The cost of a single chip remains high, and automakers are constrained by the supplier's product pace in terms of technical routes and functional definitions. This passive situation of "chips being in others' hands" is precisely the fundamental driver for Chinese automakers' collective entry into chipmaking.
In terms of performance parameters, domestic autonomous driving chips already have the confidence to compete head-on with overseas products.
NIO's Shenji NX9031 boasts over 1,000 TOPS of computing power; XPeng's Turing chip reaches up to 3,000 TOPS in its whole-vehicle solution; Li Auto's Mach M100 delivers 1,280 TOPS per single chip and 2,560 TOPS for dual chips; and BYD's Xuanji A3 exceeds 2,100 TOPS with three chips working in synergy.
Compared to the 254 TOPS computing power of a single NVIDIA Orin-X chip, domestic chips have already achieved an order-of-magnitude leap in computing power parameters.
But computing power figures are just the tip of the iceberg. The real competition lies in "effective computing power": the computing efficiency the chip can deliver in real driving scenarios.
XPeng claims that the effective computing power of a single Turing chip is roughly equivalent to 10 Orin-X chips. Although this claim remains to be verified in actual scenarios, it at least sends a signal: what domestic chips pursue is not the piling up of paper parameters, but the optimization of actual application efficiency.
A more critical change lies in the expansion of business models. NIO has spun off Shenji into an independent subsidiary and is seeking external supply; XPeng's Turing chip has secured a nod from Volkswagen; and Li Auto's Mach chip is also expected to be exported to other brands.
This means that automakers' chipmaking is moving from "self-use" to "external supply," shifting from a cost center to a profit center.
A Qualcomm executive once commented at the 2026 Automotive Summit: "No matter how well automakers with self-developed chips perform, they only serve a limited number of enterprises. Each generation of Qualcomm chips is aggregated from the demands of dozens of automakers."
However, the logic of Chinese automakers is exactly the opposite—precisely because the upfront investment for self-developed chips is huge, it is even more necessary to dilute costs and expand scale through external supply, ultimately forming a flywheel effect of "cost reduction through self-use + profit generation through external supply."
From the perspective of industrial security, the dense integration of domestic autonomous driving chips into vehicles also marks the acceleration of China's localization process in the automotive semiconductor field.
Although there is still a gap with the top international levels in advanced processes and underlying architecture design, the first step of "domestic substitution"—completing the breakthrough from 0 to 1—has been achieved. The successful mass production of NIO's 5nm and BYD's 4nm automotive-grade chips is itself a landmark achievement of the coordinated development of China's semiconductor and automotive industries.
Meanwhile, integrated cockpit and driving chips are emerging as a new technological direction.
The "cross-domain" characteristics of AI Agents are forcing intelligent cockpits and autonomous driving to merge from two independent systems. A single chip handling both cockpit interaction and driving decisions simultaneously has significant advantages in terms of cost and computing power utilization efficiency. Brands such as Li Auto and GWM Wey have begun to lay out in this direction, which will also become the next growth point for domestic chips.
However, the computing power race also has its hidden worries.
Excessive pursuit of TOPS figures may lead to "computing power overcapacity"—the vast majority of daily driving scenarios do not need computing power of over a thousand TOPS at all, and the power consumption, heat dissipation, and cost issues brought by high-computing-power chips may instead become a burden on the product.
How to find the optimal balance point between computing power and cost is a question that automakers must answer in chipmaking.
Memory Price Hikes: An Alarm Bell for Supply Chain Security
Just as the competition among automakers for autonomous driving chips is in full swing, another crisis in the chip field is quietly spreading.
According to CCTV Finance, the overall price of automotive-grade memory chips has risen by about 180% in the past three months. In Q1 2026, the average price of DDR5 memory accumulated a 288% increase, with some spot prices nearly tenfold.
The three major original manufacturers—Samsung, SK hynix, and Micron—have diverted 70% to 80% of their advanced process capacity to HBM (High Bandwidth Memory) and high-end server DDR5 memory to meet the explosive demand for AI servers.
In this battle for capacity, the automotive industry accounts for only about 3% of the global DRAM market and is placed at the very end of capacity allocation.
The transmission effect of price hikes has already emerged. BYD officially stated that the core reason for the price adjustment is the continuous rise of global automotive-grade storage hardware costs; the high-end version of Exeed ET5 changed the originally free intelligent driving package worth 28,800 RMB into a paid option; XPeng Motors raised the price of the XNGP full-scenario high-level intelligent driving option package by 20%; and the lidar intelligent driving option packages for multiple models under BYD's Dynasty and Ocean networks were raised from 9,900 RMB to 12,000 RMB, an increase of over 21%.
According to industry estimates, the price hike of memory chips has increased the cost of each NEV by about 15,000 to 20,000 RMB. For automakers whose gross profit margins are already under pressure, this cost must either be absorbed internally, eroding profits, or passed on to consumers—and the latter is almost impossible in the current fiercely competitive price war market.
NIO's Q1 2026 earnings call provided an intuitive sample. NIO founder William Li revealed that the rise in raw material costs has led to an increase in average per-vehicle cost pressure of over 10,000 RMB. NIO Vice President Ma Lin frankly stated, "Losing money by spending $300 million on NVIDIA chips; self-development is the only way to save money."
Placed in the context of rising memory chip prices, this statement carries a more profound meaning. When price fluctuations in the external supply chain can directly eat up per-vehicle profits, self-development is not only a need for competition but also a need for survival.
"In the first half of 2026, structural tightness emerged in in-vehicle memory chips. This is not a traditional cyclical fluctuation, but a cross-domain capacity struggle formed by the explosion of the AI computing power industry and the rapid increase in the penetration rate of intelligent vehicles, causing a deep mismatch in the supply and demand system." Zheng Yali, Deputy Secretary-General of the Society of Automotive Engineers of China and Executive Dean of the China Automotive Strategy Institute, stated that unlike the acute shortage of general-purpose chips in 2021, this round of "chip shortage" belongs to a long-term "chronic disease" of high-spec dedicated memory chips. The fundamental reason is that the traditional passive response and multi-level progressive supply chain management model can no longer adapt to the rapid iteration pace of intelligent vehicle demands.
She pointed out that although this situation brings short-term challenges, it also creates a rare strategic window for domestic memory chips. In the long run, the automotive industry's response strategy should shift from short-term supply guarantee in the form of "emergency relief" to systematic reconstruction of long-term capability systems, emphasizing deep collaboration with chip enterprises in demand definition, collaborative design, and supply chain mechanisms, rather than simply pursuing self-developed chips.
Notably, the chip challenges faced by automakers are not limited to autonomous driving chips and memory chips. Automotive-grade MCUs (Microcontroller Units), power semiconductors, and sensor chips are equally the "lifelines" of intelligent vehicles.
From BYD's self-developed IGBT to automakers' layout of SiC (Silicon Carbide), from NIO making autonomous driving chips to XPeng exporting technology to Volkswagen, Chinese automakers are systematically reconstructing their own chip supply chain landscape.
Forecasts from TrendForce show that the trend of rising memory chip prices is difficult to alleviate in the short term, and automakers will have to endure it for at least another year. This means that 2026 may be the year when automakers face the greatest pressure on profits, and it may also be the turning point that pushes more automakers to make up their minds to self-develop chips.
Returning to the essence of automakers making chips. This is not just a technological upgrade, but a redistribution of industrial power.
From "car makers" to "chip makers," Chinese automakers are breaking through the boundaries of the traditional automotive industry and extending their tentacles into the semiconductor field. The independent operation of NIO's Shenji, the external supply output of XPeng's Turing, and the scaled mass production of BYD's Xuanji—these cases all point to a trend: automakers' chipmaking is moving from strategic defense to strategic offense, and from cost-driven to value creation.
As an industry insider put it, the competition in intelligent vehicles has electrification in the first half and intelligence in the second half, and what determines the winner of the second half is precisely that fingernail-sized chip. Automakers have obviously understood this question—and have already started answering it.