Unveiling the Behind-the-Scenes Story of NIO's Chip Spin-off, Actively Approaching Multiple Leading Automakers

When it comes to stock trading, look at the Golden Kylin Analyst Reports—authoritative, professional, timely, comprehensive—helping you discover potential thematic opportunities!

Source: 36Kr Auto

Text | Fan Shuqing

On March 10, NIO Auto announced its financial report, with profits exceeding 1.2 billion yuan in Q4 2025. This is NIO’s first quarterly profit after several consecutive quarters of losses totaling tens of billions of yuan.

Behind this profitable financial report, Li Bin has carried out a series of reforms to strictly control costs, while also splitting off some highly invested businesses to improve the financial structure.

The chip business is undoubtedly one of NIO’s most “money-burning” operations. Li Bin once estimated at the 2025 China Electric Vehicle Hundred People Forum that the R&D cost of the NX9031 chip—developing a single NX9031 Shenji chip—would be enough to build 1,000 battery swap stations.

36Kr also reported that before being independently spun off, NIO’s chip team was unusually comprehensive, with over 600 people involved in front-end design, back-end design, and testing—close to the configuration of a standard chip company.

But now, NIO expects to see investment returns from this business. In mid-2024, Li Bin mobilized various subsidiaries to seek funding, and external supply of chips seems to be the most promising progress so far.

36Kr learned from multiple industry insiders that after the successful tape-out of the Shenji chip developed in cooperation with Aixin YuanZhi, Shenji and Aixin YuanZhi are actively engaging with automakers like Leapmotor and Geely.

Aixin YuanZhi previously stated in its Hong Kong IPO prospectus that it plans to release this chip in Q3 of this year. The M97 chip has a computing power of over 700 TOPS, comparable to Horizon’s Journey J6P. Horizon publicly states that J6P’s computing power is 560 TOPS.

Last year, after NIO spun off its chip business into an independent subsidiary—Shenji Technology—Shenji Technology also established a joint venture with Aixin YuanZhi and OmniVision. The M97 chip will be shipped from this joint venture to other automakers.

In addition, 36Kr learned that NIO Shenji and Aixin YuanZhi are also planning low-end chips, codenamed 9031e. This chip’s performance targets Horizon’s Journey J6M, which Horizon officially states has a computing power of 128 TOPS.

‘The Formation and Development of NIO Shenji and Aixin YuanZhi Joint Venture’

According to public information, in June 2025, NIO spun off its chip business and established Anhui Shenji Technology, responsible for integrating chip R&D, mass production, and licensing.

On November 14, last year, this subsidiary Shenji, together with Aixin YuanZhi and its parent company OmniVision, jointly established a joint venture—Chongqing Chuangyuan Zhihang Technology Co., Ltd., with a registered capital of 100 million yuan. Aixin YuanZhi holds a 36.4% stake.

The establishment of this joint venture marked NIO’s first external supply of assisted driving chips.

An industry insider close to Aixin YuanZhi told 36Kr that the cooperation method between Aixin YuanZhi and NIO Shenji is “a bit complex, not just buying IP or directly purchasing original chips. It can be understood as both sides working together to develop a new chip, then re-tape-out.”

“This new chip uses the core modules and architecture of NIO Shenji’s chips.”

It is understood that the cooperation between NIO Shenji and Aixin YuanZhi “was finalized in mid-2024.” In September last year, the support team for Shenji chips (responsible for software and hardware adaptation), including some product team members, had already joined Aixin YuanZhi’s DingTalk system. “The design team is still at NIO Shenji,” an insider told 36Kr. NIO gained several hundred million yuan in technology licensing fees through this cooperation.

An engineer from an automaker’s assisted driving team told 36Kr that, to balance independence and efficiency, typical cooperation involves either licensing chip IP for customization or purchasing mature chips. IP licensing offers more flexibility and higher technical barriers—essentially buying the chip’s “design drawings” and then integrating manufacturing and debugging oneself.

However, the cooperation between Aixin YuanZhi and NIO Shenji did not directly adopt these two common methods. This is why an industry insider close to Aixin YuanZhi described the partnership as “complex.”

In the cooperation between Aixin YuanZhi and NIO Shenji, car companies like Leapmotor played a catalytic role. An engineer from a car company told 36Kr that to achieve more precise cost control, companies like Leapmotor are considering using domestically produced high-computing-power assisted driving chips to replace products from Nvidia, Qualcomm, etc. Although Horizon’s J6P chip has high computing power, until the end of last year, it had to be sold bundled with Horizon’s algorithms, which could limit the autonomy of companies like Leapmotor in assisted driving technology. As a result, some automakers hesitated when facing J6P.

Aixin YuanZhi quickly captured the customer’s intention to reduce costs using domestically produced high-computing-power chips.

Moreover, Aixin YuanZhi’s product lineup lacks high-end chips for urban NOA (Navigate on Autopilot). Horizon launched the Journey 6 series chips in April 2024 and began pre-installation mass production the same year. Therefore, to meet customer needs and for strategic defense, Aixin YuanZhi must quickly develop a high-end assisted driving chip suitable for urban NOA.

Cooperation is naturally the fastest way to fill this gap. However, industry insiders close to Aixin YuanZhi say, “There are not many suitable partners.”

“Among automakers developing their own assisted driving chips, Xpeng’s Turing chip has production experience, but Xpeng was not very short of money last year, so negotiations on price were probably tough. Ideal’s chips won’t be mass-produced until 2026 on the all-new L9, and they’re not in a hurry to profit from chips. Since NIO’s L90 and all-new ES8 hadn’t launched yet and faced significant performance pressure, NIO’s Shenji chips were a relatively suitable choice.”

To reduce costs, Aixin YuanZhi retained the architecture and modules of NIO NX9031 but made some “cutbacks” in specifications. NIO NX9031’s computing power exceeds 1,000 TOPS, while Aixin YuanZhi reduced it to about 700 TOPS.

“Plus, Aixin YuanZhi offered a fairly sincere price to automakers. Now, companies like Leapmotor and Geely are interested in this chip.” In the future, chips from Aixin YuanZhi and NIO cooperation will be shipped from the joint venture Chongqing Chuangyuan.

‘Automakers’ Self-Developed Chips Fill the Short-Term Gap in High-Performance Computing Market’

In early 2025, Li Bin set a goal for NIO to achieve single-quarter profitability in Q4. This was against the backdrop that NIO had experienced consecutive quarters of losses exceeding 5 billion yuan.

Under enormous financial pressure, spinning off the chip business and bringing in external investors became an inevitable move.

Li Bin publicly stated that the R&D expenditure for NX9031 was equivalent to building 1,000 battery swap stations. With each station costing at least 1 million yuan, the total R&D investment in NX9031 was at least several billion yuan. Therefore, the independence of the chip business is crucial for NIO to achieve profitability in Q4 2025.

An insider told 36Kr that external supply of chips not only brought NIO hundreds of millions of yuan in licensing fees but also, during the joint development of the second and third chips with Shenji and Aixin YuanZhi, specifications and tape-out were mainly led by Aixin YuanZhi, sharing R&D costs.

Moreover, industry sources close to Shenji believe that Shenji has no its own channels and previously did not emphasize supply chain control. Aixin YuanZhi, with its mature business team and channels, is more familiar with market pricing and customer needs, making it easier to secure chip orders and generate higher profits.

For example, Leapmotor, which is close to Aixin YuanZhi, delivered nearly 600,000 vehicles in 2025 and achieved over 500 million yuan in profit. This year, Leapmotor’s COO Xu Jun announced an internal sales target of 1.05 million vehicles. If Shenji’s second chip can be widely adopted by Leapmotor, NIO could earn higher returns from its chip business.

Besides the M97, Shenji is also planning low-end chips with Aixin YuanZhi, codenamed 9031e, performance comparable to Horizon’s J6M. This could further expand Shenji’s external supply opportunities.

Outside NIO, Xpeng is also working to expand the customer base for its Turing assisted driving chips.

A Xpeng technical engineer told 36Kr that although Xpeng’s Turing chip already has Volkswagen as a customer, this cooperation is bundled within the Volkswagen-Xpeng electronics and electrical architecture agreement. “Xpeng also wants to find other automakers and robotics clients.”

The Chinese auto market indeed has significant potential for domestically produced assisted driving chips to replace imported ones.

Currently, many automakers still use Nvidia’s hardware solutions for high-end assisted driving. A single Orin-X chip costs about $300–$500, while Thor chips cost over a thousand dollars. Facing increasingly fierce industry competition, automakers’ cost reduction needs are growing, and Horizon, Momenta, and other self-developed chips are seizing market opportunities.

Furthermore, the scarcity of third-party domestically designed chips with good architecture and high computing power provides a window for companies like NIO to develop and supply their own chips.

An engineer from an automaker’s assisted driving team said, “As long as performance and price are suitable, automakers won’t reject using more cost-effective domestic chips.” However, how long this market window will last remains uncertain.

In January this year, Horizon founder Yu Kai revealed that the Journey 7 series chips will be launched simultaneously with Tesla’s next-generation AI5, with Horizon’s chief architect Su Qing involved in the architecture design and product definition. It’s expected that Journey 7 will be optimized based on Journey 6’s architecture and computing power.

Moreover, under US restrictions, it’s not easy for automakers to use 5nm or more advanced process chips for external supply. Competition from suppliers, US export restrictions on advanced process chips, and automakers’ inherent resistance to parts from competitors all add obstacles to external chip supply.

Even ignoring market opportunities, how to maintain technological leadership while externally supplying chips is a difficult balance for automakers. The original intention of developing in-house chips was partly to reduce costs and partly to build a full-stack capability from chips to systems and algorithms. Chips are fundamental to achieving full-stack leadership.

Especially as automakers’ assisted driving technology shifts toward large models, the architecture and computing power of vehicle chips have become critical factors limiting assisted driving performance. Therefore, although NIO and other automakers’ external chip supply efforts are making progress, they still need to consider two issues—how to seize the short window and secure more orders, and how to maintain their competitive edge in assisted driving while external supply continues.