Revenue soared 181.8%! Aixin Yuanzhi (00600) launched an edge AI breakout battle with self-developed NPU full computing power matrix

During the period of intensive disclosure of results, Aixin Yuanzhi (00600) handed over an outstanding questionnaire.

According to the 2026 interim results announced by Aixin Yuanzhi, in the first half of the year, the company recorded revenue of 402 million yuan (RMB, same below), a year-on-year increase of 181.8%, gross margin was significantly optimized to 29%, and the performance indicators were improving across the board; the three major business lines of edge AI reasoning, smart cars, and terminal computing went hand in hand. Among them, the edge AI business achieved a year-on-year growth of 251.9%, which has become an important engine of performance.

Established less than ten years ago, Aixin Yuanzhi has become one of the fastest growing manufacturers in the domestic edge AI chip field. According to Insight Consulting data, Aixin Yuanzhi ranked among the top three in China's edge AI inference chip industry with a 12.2% share, and ranked first in the world with a market share of 24.1% in the middle and high-end visual edge chip segment.

Behind the boom in performance, Aixin Yuanzhi chose a completely different path from most manufacturers: “tailor-made” NPU processors for AI, breaking NPU's existing computing power bottlenecks, and building a gradient computing power product matrix of up to thousands of TOPS; the same architecture can be reused and migrated to multiple downstream fields such as physical intelligence, smart driving, and vision, covering a wide range of application scenarios.

Exclusive self-developed AI processor technology to create a “hard core” technology base

In the AI computing power system, the choice of hardware architecture is not a simple line opposition, but rather stems from the precise division of labor between computational scenarios and physical constraints.

With its strong computing power flexibility and omnipotent parallel computing capabilities, general-purpose GPUs are still the main force for large-scale model training and ultra-complex computation in the cloud; while on the edge side and end side close to the data source, they are limited by power consumption, heat dissipation, BOM costs, and real-time response requirements at the millisecond level. NPUs (neural network processing units) specially customized for neural network computation have shown irreplaceable energy efficiency ratio and economy.

Currently, AI computing architectures are evolving at an accelerated pace from “centralized cloud” to “distributed collaboration on the edge of the cloud”. Along with the implementation and popularity of large models, pure cloud inference faces network latency (it is difficult to meet millisecond interaction between smart driving and robots), data privacy compliance (data does not leave the domain), and high continuous token call costs.

As a result, it has become an industry consensus to offload high-frequency, low-latency, and privacy-focused inference tasks to edge nodes and terminal devices, and form a collaborative system of “cloud training/complex decision-making + efficient local edge inference” with cloud models.

Under this trend, the advantages of NPU's high efficiency and low power consumption are becoming irreplaceable. From end-side AI chips to inference acceleration cards, from edge computing power boxes to inference servers, NPU architectures have become the mainstream choice for low power consumption scenarios.

Aixin Yuanzhi's self-developed “Aixin Tongyuan NPU” was created to break the power consumption and performance bottlenecks of edge reasoning. It abandons the redundant design of traditional chips, completely breaks through the “traffic jam” bottleneck of data handling through natively compatible Transformers models and mixed accuracy calculations. The measured energy efficiency ratio can reach up to 10 times that of traditional architectures (in the case of AX8850, it can process nearly 200 frames of images in real time for every 1 watt of power consumption).

More importantly, it breaks the barrier of traditional special purpose chips being “difficult to reuse” — based on the same technical base, it can not only be used as a small terminal with low power consumption, but also expanded upward to support advanced smart driving (M97), a robotic brain, and even a thousand T-level edge clusters (Yuanxi series). This “one architecture covers all scenarios” platform-based reuse capability greatly reduces repetitive R&D and hardware manufacturing costs, and supports the company's extremely efficient and rapid implementation in all physical AI scenarios.

Up to now, the company's product matrix, which covers all computing power, has been implemented in thousands of industries related to edge AI, smart mobility, and smart terminals, so that AI can truly transform into productivity and effectively benefit people's livelihood.

Full Computing Power Product Matrix: From Consumer to the “1000T Era”

Edge AI is deployed on a server, gateway, or base station close to the data source to perform real-time local inference to balance high performance in the cloud with low latency and data security on the end side.

Looking at industry trends, AI deployment is moving from centralized inference in the cloud to a new stage of distributed intelligence, driven by the triple drive of stricter privacy regulations, sensitive network delays, and high bandwidth costs. Privacy compliance risks brought about by uploading raw data to the cloud continue to rise, and edge inference naturally achieves “no data leaving the domain” to provide technical guarantees for compliance; scenarios such as autonomous driving require millisecond response, local inference eliminates network round trips to ensure high real-time and reliability; the token fee for cloud inference is growing rapidly, while marginal inference costs are significantly lower, making the business model of AI services more sustainable.

Precisely based on a keen insight into the industry, edge AI has become the strategic focus of Aixin Yuanzhi. Since 2025, the company has implemented three major models of card edge private clouds, edge servers, and smart terminals in advance, and actively launched secure, controllable, and cost-effective AI inference computing power products on the edge side to meet the exploding demand in the computing power market. In July 2026, Aixin Yuanzhi announced the establishment of a wholly-owned subsidiary “Aixin Computing”. This move further deepened the company's strategic deployment of edge AI and injected accelerators into the commercialization of related products.

According to information, Aixin Computing will collaborate deeply with the parent company to transform the company's underlying chip capabilities into standardized solutions for computing power cards, core boards, etc., lower the customer's development threshold through system-level product delivery, and achieve rapid application implementation.

At the latest WAIC, the company unveiled the Yuanxi series of high-computing power products for the first time — this series of AI inference cards has over 1000 TOPS computing power, ultra-large video memory and ultra-high bandwidth. It is specially designed for high-concurrency scenarios such as computer room clusters, multi-channel video analysis, and privatized knowledge bases. It efficiently supports high token consumption businesses such as long text inference and high-definition image analysis, and greatly reduces the industry's dependence on cloud computing power.

Facing low power consumption, lightweight, and standardized scenarios, the AX8850 computing power card supports “plug and play” to help micro, small and medium-sized enterprises significantly reduce the cost of daily AI applications. It has been implemented in batches in vertical scenarios such as smart education and smart industry, and shipments increased dramatically during the reporting period.

From consumer-grade lightweight to edge computing power, Aixin Yuanzhi has built a complete spectrum of computing power products. Compared with the GPU route, the NPU route has significant economic advantages in edge-side reasoning, which will also become a competitive barrier for Aixin Yuanzhi in all scenarios.

At the same time, Aixin Yuanzhi continues to increase investment in R&D to ensure the company's continued technological leadership and accelerate the mass production process of next-generation products. According to financial reports, in the first half of 2026, Aixin Yuanzhi invested 516 million yuan in research and development. During this period, iXin Yuanzhi successfully completed various advanced process SoC streaming films. According to information, Aixin's next-generation edge AI chip will significantly improve computing power specifications, can fully support the core nodes of large-scale model implementation, and can meet many industry pain points such as long context processing, efficient token generation, and real-time robot sensing and decision-making.

New products with big computing power accelerate the process of edge AI strategy

Through a unified set of operator instructions across the end side, edge side and data center, Aixin Yuanzhi's NPU architecture enables seamless migration of models and optimization experiences between the cloud, edge, and end. This highly reusable underlying technology platform has brought significant R&D efficiency and commercial expansion advantages to the company. Since 2026, the company has used this architecture to quickly enter multiple high-growth tracks, intensively launch new products with high computing power in emerging fields such as physical intelligence, advanced smart driving, and intelligent visual perception, and accelerate the transformation of the underlying technology moat into scene implementation results.

Among them, in the field of embodied intelligence, which is regarded as the ultimate form of physical AI, Aixin Yuanzhi's newly launched physical brain controller has 1500 TOPS of extreme computing power and ultra-flagship bandwidth about twice the industry-leading level. With an AEC-Q100 level full-dimensional functional safety design, it ensures reliable operation of the equipment. Its open development system supports custom operator development, is natively compatible with cutting-edge algorithms such as world models and VLA, and creates a high-performance, high-security, and highly flexible intelligent computing center for general-purpose robots. Combined with the AX8910 special vision chip for the sensing end, which has already been implemented on a large scale, the company's chip solution has covered the full link requirements of embodied intelligence, from decision making to perception.

In addition to physical robots, intelligent driving tracks, which also have extreme requirements for cutting-edge algorithms such as high bandwidth and VLA/world models, are another core main position for the commercialization of the company's large computing power chips. In the field of smart driving, Aixin Yuanzhi's high-end smart driving chip M97 chip was successfully released in February 2026. Project samples were delivered in the first half of the year, and many leading car companies have already begun program evaluation and model selection. The chip targeted the long-standing “insufficient bandwidth” shortcoming of mainstream domestic chips, increased the bandwidth to double that of the industry's flagship smart driving chip, and unleashed effective computing power to the greatest extent. Its single-chip computing power exceeds 700 TOPS, supports intelligent driving functions from L2+ to L3/L4 levels, and natively supports algorithm architectures such as VLA and world models.

While the M97 is targeting the forward-looking market for high-end smart driving, the company's current high growth in the automotive business is strongly led by its main product, the M57. Up to now, the M57 chip has been designated as standard models by many leading domestic OEMs, and targeted projects by overseas car companies are progressing steadily. In the first half of the year, the company's smart car business revenue increased 252.1% year on year. SoC shipments reached 420,000 units, added 24 targeted mass-produced models, and cooperated with 25 domestic and foreign OEM brands. It has entered the fast track of volume release and globalization.

While edge AI and smart driving are growing rapidly, the terminal computing business, which is the company's revenue base, has made steady breakthroughs under the upgrading trend of “perception+reasoning”, firmly entrenched the position of “one brother” in the industry, and provided stable cash support for emerging businesses.

In terms of terminal business, as the core of next-generation visual terminal AI inference chips shifts from a single “perception” to “perception+local reasoning” in parallel, the company's related high-end products began to rapidly occupy the market. According to information, in the first half of the year, the company's terminal computing business revenue increased 169.5% year on year, and product sales increased by more than 100% year on year. Among them, the sales volume of the “Black Light” series products grew by more than 200%, and the market share increased significantly.

During the reporting period, sales of the company's latest generation “black light” high-computing power visual perception product SoC AX615 series grew rapidly. The product not only inherits excellent “black light” video effects, but also integrates high-computing power NPU, which can better meet the needs of various end-side AI applications, and has gradually been implemented in emerging fields such as physical robots and industrial vision.

Summarize

The sinking of AI computing power from the cloud to the physical world is an irreversible industry trend.

In the first half of 2026, Aixin Yuanzhi relied on the structural upgrade of the terminal's visual infrastructure (AX615 “black light” outbreak) and strong price transmission capabilities to drive the overall gross margin to 29.0%. At the same time, both edge AI and smart car businesses have surged by more than 200% year-on-year, verifying the platform-based multiplexing advantages of self-developed NPU “one architecture covering the full computing power spectrum”.

With the establishment of the wholly-owned subsidiary “Aixin Computing”, the company has further deepened its edge AI strategy, and will use standardized AI inference solutions to further lower the customer's development threshold and accelerate the commercialization of AI applications. Meanwhile, R&D investment and forward-looking preparation brought about by concentrated streaming in the first half of the year are also locking in production capacity and intergenerational technology advantages for the company during the large-scale explosion of edge AI.

As a “shovel seller” in the AI wave, Aixin Yuanzhi is particularly worthy of being looked forward to by the market due to its deep technical barriers and cutting-edge business layout.