Revenue surged by 181.8%! AXERA (00600) launches a breakthrough in edge AI with its self-developed NPU full-computing matrix.

During the intensive performance disclosure period, AXERA (00600) delivered an impressive report card. According to AXERA's mid-year performance report for 2026, the company recorded revenue of 402 million yuan (RMB, the same below) in the first half of the year, a year-on-year increase of 181.8%, with a significantly improved gross profit margin of 29%, and overall performance indicators showed positive trends; the three major business lines of edge AI inference, smart cars, and terminal computing advanced in tandem, with the edge AI business achieving a year-on-year growth of 251.9%, becoming a key driver of performance. Founded less than ten years ago, AXERA has become one of the fastest-growing companies in the domestic edge AI chip sector. According to data from Zhenzhong Consulting, AXERA ranks among the top three in China's edge AI inference chip industry with a market share of 12.2%, and holds the number one position globally in the mid-to-high-end visual edge chip sub-sector with a market share of 24.1%. Behind the soaring performance, AXERA has chosen a path distinctly different from most manufacturers: creating NPU processors tailored for AI, breaking through the existing computing power bottleneck of NPUs, and building a hierarchical product matrix with computing power up to thousands of TOPS; the same architecture can be repurposed for various downstream fields such as embodied intelligence, intelligent driving, and vision, broadly covering diverse application scenarios. Exclusive self-developed AI processor technology creates a "hardcore" technological foundation. In the AI computing architecture, the choice of hardware architecture is not merely a route opposition but originates from the precise division of computing scenarios and physical constraints. General-purpose GPUs, with their strong computing elasticity and versatile parallel computing capabilities, remain the mainstay for training large models and conducting ultra-complex calculations in the cloud; however, on the edge and terminal side, constrained by power consumption, heat dissipation, BOM costs, and millisecond-level real-time response requirements, NPUs (Neural Processing Units), specially designed for neural network calculations, exhibit an irreplaceable energy efficiency ratio and economy. Currently, the AI computing architecture is accelerating its evolution from "centralized cloud" to "cloud-edge-terminal distributed collaboration." As large models are being adopted more widely, the purely cloud-based inference faces challenges such as network latency (which cannot meet the millisecond-level interactions required by intelligent driving and Siasun Robot & Automation), data privacy compliance (the requirement that data does not leave the domain), and high ongoing costs for token calls. Therefore, offloading high-frequency, low-latency, and privacy-focused inference tasks to edge nodes and terminal devices, creating a collaborative system of "cloud training/complex decision-making + efficient local inference at the edge," has become an industry consensus. In this context, the advantages of NPUs in terms of efficiency and low power consumption are becoming irreplaceable. From edge AI chips to inference acceleration cards, from edge computing boxes to inference servers, NPU architectures have become the mainstream choice for low-power scenarios. AXERAs self-developed "Aixin Tongyuan NPU" was born to address the power consumption and performance bottlenecks of edge inference. It abandons redundant designs of traditional chips and, by being natively compatible with Transformer large models and mixed precision calculations, has completely resolved the "traffic jam" bottleneck in data handling, with an energy efficiency ratio recorded at up to 10 times that of traditional architectures (for example, the AX8850 can process nearly 200 frames in real-time with just 1 watt of power). More importantly, it breaks down the barriers of traditional dedicated chips that are "hard to repurpose"based on the same technological foundation, it can serve as a low-power small terminal but can also scale up to support advanced intelligent driving (M97), embodied Siasun Robot & Automation brains, and even thousands of T-level edge clusters (Yuanxi series) ultra-high power chips. This capability of "one architecture covering all scenarios" significantly reduces redundant research and development and hardware manufacturing costs, supporting the company in rapidly deploying physical AI across various scenarios. As of now, the company's comprehensive product matrix has already penetrated thousands of industries related to edge AI, smart mobility, and smart terminals, enabling AI to genuinely transform into productivity, effectively benefiting people's livelihoods. Complete power coverage product matrix: from consumer level to the "thousand-T era." Edge AI is deployed on servers, gateways, or base stations that are close to data sources to perform real-time local inference, achieving a balance between high performance in the cloud and low latency and data security on the terminal side. From an industry trend perspective, driven by three factorsstricter privacy regulations, sensitivity to network latency, and high bandwidth costsAI deployment is shifting from centralized cloud inference to a new phase of distributed intelligence. The privacy compliance risks from uploading raw data to the cloud continuously rise, while edge inference inherently realizes "data not leaving the domain," providing technological safeguards for compliance; scenarios such as autonomous driving require millisecond-level responses, and local inference eliminates round-trip network delays, ensuring high real-time performance and reliability; the rapidly increasing token costs of cloud inference, compared to significantly lower marginal inference costs, make the business model for AI services more sustainable. Based on keen insights into the industry, edge AI has become AXERA's highly focused strategic priority. Since 2025, the company has proactively positioned itself at three critical landing nodes for edge private clouds, edge servers, and intelligent terminal endpoints, actively launching secure and controllable, high-cost-performance AI inference computing power products on the edge to meet the explosive growth in computing power demand. In July 2026, AXERA announced the establishment of its wholly-owned subsidiary "Aixin Computing," further deepening the companys strategic deployment in edge AI and injecting acceleration into the commercialization of related products. It is understood that Aixin Computing will deeply cooperate with its parent company to transform the underlying chip capabilities into standardized solutions such as computing power cards and core boards, reducing customers' development thresholds through system-level product delivery, achieving rapid landing for applications. At the latest WAIC, the company unveiled the Yuanxi series high-performance computing power products for the first timethis series of AI inference cards boasts over 1000 TOPS of computing power, extremely large memory, and ultra-high bandwidth, specifically designed for high concurrency scenarios such as data center clusters, multi-channel video analysis, and privatized knowledge bases, efficiently supporting long-text inference and high-definition image processing businesses that consume high tokens, significantly reducing the industry's reliance on cloud-based computing power. Targeting low-power, lightweight, and standardized scenarios, the AX8850 computing power card supports "plug and play," significantly reducing daily AI application costs for small and micro enterprises and has already seen mass deployment in vertical scenarios like smart education and smart industry, with substantially increased shipment volume during the reporting period. From consumer-grade lightweight to edge high computing power, AXERA has constructed a complete spectrum of computing power products. Compared to the GPU route, the NPU route offers significant economic advantages in edge-side inference, which also serves as a competitive barrier for AXERA across all scenarios. At the same time, AXERA continues to increase its investment in research and development to ensure sustained technological leadership and accelerate the mass production process of new-generation products. According to financial reports, in the first half of 2026, AXERA invested 516 million yuan in research and development, successfully completing the tape-out of several advanced process SoCs during the period. It is understood that the next-generation edge AI chip from Aixin will significantly enhance computing power specifications to fully support the core nodes for large model implementation, addressing various industry pain points such as long-context processing, efficient token generation, and real-time perception and decision-making by Siasun Robot & Automation. New high-performance products accelerate the edge AI strategy. By unifying the operator instruction set across the terminal, edge, and data center, AXERA's NPU architecture achieves seamless migration of models and optimization experiences between cloud, edge, and terminal. This highly reusable foundational technology platform brings significant research and development efficiency and commercial expansion advantages to the company. Since 2026, the company has rapidly entered multiple high-growth tracks using this architecture, intensively launching high-performance new products in emerging fields such as embodied intelligence, advanced intelligent driving, and smart visual perception, accelerating the transformation of its foundational technological moat into tangible results in various scenarios. In the field of embodied intelligence, considered the ultimate form of physical AI, AXERA's latest launched embodied brain controller features an exceptional computing power of 1500 TOPS and ultra-flagship bandwidth approximately twice that of industry leaders. Its AEC-Q100-grade comprehensive functional safety design ensures reliable equipment operation. The open development system supports custom operator development and is natively compatible with cutting-edge algorithms such as world models and VLA, creating a high-performance, high-security, and highly flexible intelligent computing hub for general Siasun Robot & Automation. Coupled with the already deployed perception-oriented AX8910 dedicated vision chip, the company's chip solutions have covered all the needs for decision-making to perception in embodied intelligence. Besides embodied Siasun Robot & Automation, the intelligent driving sector, which also demands high bandwidth and cutting-edge algorithms such as VLA/world models, stands as another crucial battlefield for commercializing high-performance chip products. In the field of intelligent driving, the AXERA's high-end intelligent driving chip M97 successfully completed its engineering sample delivery in February 2026 and has commenced evaluation and selection processes with multiple leading automotive manufacturers. This chip specifically optimizes the longstanding insufficient bandwidth issue that mainstream domestic chips have faced, doubling the bandwidth compared to the industry's flagship intelligent driving chips to maximize effective computing power. Its single-chip computing power exceeds 700 TOPS and supports intelligent driving functions from L2+ to L3/L4 level, natively accommodating architectures for algorithms such as VLA and world models. While the M97 targets the mid-to-high-end intelligent driving market, the company's current high growth in automotive business is primarily driven by the M57 product. So far, the M57 chip has been designated for standard models by several leading domestic OEMs, and projects with overseas automobile manufacturers are steadily advancing. In the first half of the year, the company's intelligent automotive business revenue increased by 252.1% year-on-year, with 420,000 SoCs shipped, adding 24 new mass production model alignment projects and collaborating with 25 OEM brands both domestically and internationally, entering a rapid expansion and globalization phase. As edge AI and intelligent driving rapidly grow, the terminal computing business, which serves as the company's revenue base, steadily breaks through under the upgrading trend of "perception + inference," firmly maintaining its position as industry leader and providing stable cash support for emerging businesses. On the terminal side, with the shift of the new generation of visual terminal AI inference chips core from single "perception" to parallel "perception + local inference," the company's high-end products have quickly gained market share. In the first half of the year, the companys terminal computing business revenue increased by 169.5% year-on-year, with product sales growing over 100% year-on-year. The "Blacklight" series products saw sales growth exceeding 200%, significantly enhancing market share. During the reporting period, the latest generation of "Blacklight" high-performance visual perception product SoC AX615 series saw rapid growth in sales. This product not only inherits the excellent "Blacklight" video effects but also integrates a high-performance NPU, better meeting various end-side AI application needs and gradually landing in emerging fields such as embodied Siasun Robot & Automation and industrial vision. In summary, The trend of AI computing power sinking from the cloud to the physical world is irreversible. In the first half of 2026, AXERA, driven by the structural upgrade of its terminal visual base (AX615 "Blacklight" breakout) and strong price transmission capability, propelled its overall gross profit margin to 29.0%. Concurrently, both edge AI and intelligent automotive businesses achieved over 200% year-on-year explosive growth, validating the platform-based reuse advantages of the self-developed NPU, which covers the entire spectrum of computing power. With the establishment of its wholly-owned subsidiary "Aixin Computing," the company further deepened its edge AI strategy, aiming to lower customers development thresholds through standardized AI inference solutions, thereby accelerating the productization of AI applications. Moreover, the research and development investments from concentrated tape-outs during the first half are positioning the company to secure capacity and technological advantages in the scaling period of edge AI. As a "shovel seller" in the AI wave, AXERA, with its profound technological barriers and cutting-edge business layout, is particularly worthy of market expectations.