Today, Moffett Artificial Intelligence (hereinafter referred to as "Moffett AI") announced the completion of a C+ round funding of nearly CNY 1 billion. This round was jointly participated in by industrial investors such as Renai Capital, Haitang Fund, and Dongheng Petroleum, as well as prominent state-owned capital and financial investment institutions including Qiu Capital, Paramount Capital, and Zhongcai Capital.
It is reported that the funds raised will be invested in the full-stack upgrade of sparse-specific inference chips, with a focus on pre-research for next-generation chips. Additionally, Moffett AI will advance the construction of an industry-academia-research alliance for sparse computing to further build long-term competitive barriers.
Inference Demand RisingSparse Computing Moves to Industrial Foreground
In recent years, as large model applications have entered the scaling phase, the focus of computing power has been shifting from training to inference. Continuously rising inference costs have become a core bottleneck constraining the widespread adoption of large models.
Industry players are seeking better paths to reduce computing costs, and sparse computing is one of the most promising technical routes among them. Its core concept is: calculate only valid data, skip invalid operations generated by zero or near-zero values, thereby completing inference tasks with less computational load.
Moffett AI is precisely a company deeply rooted in this path. Founded in 2018, as a leader in the field of sparse computing, it focuses on R&D of sparse computing architectures, AI inference chips, and software-hardware co-computation platforms.
Theoretically, sparse computing can save computing power, but whether it can truly achieve this depends on whether the chip, compiler, and software ecosystem can coordinate to skip zero-value operations. This not only requires enterprises to possess chip design capabilities but also cross-layer capabilities encompassing algorithm optimization and compilation toolchains.
Moffett AI's founding team comprises chip R&D talents from companies such as Intel, Qualcomm, Marvell, and Oracle. According to public company information, core members have over 20 years of experience in the semiconductor industry and have participated in the development of dozens of chip products, with cumulative mass production exceeding 5 billion units.
It is this team with full-stack background that supports Moffett AI's choice of the "Algorithm-Software-Hardware" co-design route: performing model sparsification on the algorithm side, building compilers and deployment toolchains on the software side, and developing dedicated inference chips supporting sparse instruction sets on the hardware side, gradually connecting the entire chain of sparse computing.
Years of sustained full-stack efforts have accumulated into systematic intellectual property and translated into reproducible performance proofs. To date, Moffett AI has obtained over 100 self-developed patent authorizations, covering key modules such as sparse instruction set design, hardware scheduling, compilation optimization, and model sparsification, forming a relatively complete patent protection wall. In the internationally authoritative AI benchmark evaluation MLPerf Inference, Moffett AI won championships in the open division for three consecutive sessions with its SparsePrime® S30 and S40 computing cards. This result demonstrates that Moffett AI's sparse computing architecture can deliver high computing performance under open division evaluation conditions, providing technical support for subsequent cluster deployment and promoting industry model migration.
Commercialization Fully AcceleratedBenchmark Cases Landing Faster
Technical validation laid the foundation for scaled delivery. In recent years, as sparse computing architectures have been gradually validated in real business scenarios, Moffett AI's commercial implementation has continued to accelerate.
Currently, Moffett AI's self-developed inference clusters have been deployed across four major regions: Northwest, Southwest, East, and North China. They provide efficient, low-cost inference computing power support for industries planned under the "15th Five-Year Plan," such as biomedicine and intelligent manufacturing, as well as emerging applications like embodied intelligence and world models. Meanwhile, cooperation has been established with leading telecom operators. Leveraging the nationwide computing power network covered by operators, its technologies and products are expected to extend to more business scenarios.
Regarding benchmark projects, in assisting new-type computing infrastructure construction, Moffett AI participated in China's first offshore computing Token Factory created by a Shanghai state-owned capital platform company. Addressing the issue where general GPU inference costs might offset green electricity benefits even after reducing PUE and electricity costs through offshore wind power and seawater cooling, Moffett AI leveraged its sparse architecture advantages combined with heterogeneous computing power modes to reduce single-token inference costs from the chip side, achieving a superposition of "green electricity cost reduction" and "computing architecture cost reduction" benefits.
Facing the frontier fields of spatiotemporal AI and spatiotemporal big data, Moffett AI effectively alleviates the computing bottleneck of high-concurrency AI inference with sparse computing power. Recently, Moffett AI collaborated with the Institute of Earth and Space Technology at the Peking University Shanghai Lingang International Science Center. Based on its self-developed sparsification technology, they completed the adaptation and testing of two international mainstream remote sensing image data intelligent analysis base models on Moffett AI's computing cards. Data shows that under conditions of almost no loss in business accuracy and without retraining downstream tasks, inference throughput increased by up to 2.88 times, and model compression reached up to 7.64 times, significantly reducing computing consumption and processing time for large-scale remote sensing image interpretation. This achievement validates the engineering feasibility of sparse computing power in geospatial base model inference scenarios, providing a cost-effective computing foundation for applications such as spatiotemporal big data intelligent computing.
Dual Support from Capital and PolicyIndustrial Layout Continues to Advance
Impressive commercial implementation results have enabled Moffett AI to continuously receive additional capital support. Just five months ago, Moffett AI had just completed its previous round of nearly CNY 1 billion in financing, bringing the total for both rounds to nearly CNY 2 billion.
Beyond recognition from the capital side, the sparse computing technology Moffett AI specializes in has also received clear policy support. In September 2026, the Ministry of Industry and Information Technology (MIIT) and the National Development and Reform Commission (NDRC) issued the "'15th Five-Year Plan' for the Development of Electronic Information Manufacturing Industry" (MIIT Joint Regulation [2026] No. 216), proposing to "strengthen breakthroughs and innovations in dynamic sparse computing technology" and "layout new routes such as software-hardware coupled inference chips."
At the same time, Moffett AI's own product iteration pace is accelerating. It is reported that its next-generation board SP200 is scheduled to launch before the end of the year, and the upgraded software stack SECA (Sparse-native Efficient Computing Architecture) will soon be released. This will strengthen support for large model inference deployment and cluster operation and maintenance, further lowering the threshold for model migration and deployment.
Furthermore, to strengthen its own technical barriers, Moffett AI initiated the establishment of China's first industry-academia-research alliance for sparse computing in August this year. Focusing on frontier research, key technology breakthroughs, and achievement transformation, it aims to accelerate the industrialization process of sparse technology. The alliance brings together universities such as Fudan University, Tsinghua University, Xi'an Jiaotong University, Nankai University, and Shanghai Jiao Tong University for joint research topics, with relevant research outcomes to be released successively.
The true value of a foundational technology lies in whether it can grow into reliable infrastructure within the industrial soil. The journey of sparse computing from academic concept to engineering reality reflects a shift in China's AI computing supply approach: moving from pursuing scale expansion of general computing power to deep mining of computing efficiency. And Moffett AI is precisely an important force driving the realization of this shift.
Text | Nayun Editor | Yida