MoXin AI completed nearly 1 billion yuan in Series C financing, and the next-generation chip SparsePrime® will be launched within this year.
MoXin AI (hereinafter referred to as "MoXin") recently officially completed its Series C financing, with an amount of nearly one billion RMB. This round of financing has attracted heavyweight industrial capital and market - oriented institutions such as Shenzhen Capital Group, Yanshan Technology, Greater Bay Area Common Home, Liding Capital, and Yunsheng Capital. Many old shareholders, including Triumph Venture Capital, Chuangxiang Investment, and Shengjing Jiacheng, also participated. This diversified shareholder structure of "industrial giants + state - owned endorsement + financial capital" not only ensures the depth of technical synergy but also provides solid resource support and industrial backing for MoXin in the national computing power network layout. It marks that sparse computing is accelerating from the technical verification stage to a new stage of large - scale industrial explosion.
When the financing was announced, the company's core product: the new generation of computing card SparsePrime® (hereinafter referred to as "SparsePrime®") will be officially launched within this year. The SparsePrime® computing card is a high - performance AI general - purpose inference computing card for intelligent computing centers and data centers. Based on the self - developed Antoum2.0 chip architecture, it is specially designed for optimization in large - model and complex inference scenarios. This product adopts a top - down overall design concept, is widely applicable to mainstream Transformer models, strengthens general adaptability, and is equipped with a complete toolchain to enable customers to obtain sparse acceleration quickly with zero acceptance cost. Developers' existing model codes based on PyTorch and TensorFlow, as well as efficient inference frameworks such as vLLM, can be migrated and directly deployed and run with almost no code modification. At the same time, it supports developers to use the Triton language for custom operator development, minimizing the usage threshold. SparsePrime® will achieve new breakthroughs in sparse computing efficiency based on the real - load data accumulated during the deployment of multi - thousand - card clusters in multiple computing power centers, further consolidating MoXin's differentiated competitiveness in the field of AI inference computing power and initially realizing the technical path possibility of doubling computing power without loss of accuracy.
The confidence of SparsePrime® comes from MoXin's continuous accumulation of technical strength in the field of sparse computing. Before this, MoXin's computing cards such as S30 and S40 have won the championship for three consecutive sessions in the international authoritative AI benchmark test MLPerf™ Inference, demonstrating leading energy efficiency ratios and unit computing power inference throughput in mainstream model tasks such as vision, natural language processing, and large models. They achieve better inference performance with significantly lower power consumption than industry flagship products, fully verifying the engineering feasibility and commercial value of sparse computing under real data center loads.
Full - speed commercialization: Multi - thousand - card clusters are deployed in the east, west, south, and north, and large - scale implementation in multiple industry scenarios
The technical value is also echoed in industrial penetration. MoXin has entered the stage of "national multi - regional multi - thousand - card cluster deployment" from single - point project verification. The inference cluster based on self - developed sparse computing technology is becoming the core computing power base of intelligent computing centers in multiple key regions, further realizing a differentiated technical route of zero accuracy loss and upgraded computing power.
Currently, the 14th Five - Year Plan emphasizes that the added value of the core industries of the digital economy accounts for 12.5%. The "East - West Computing" project requires that the proportion of green electricity in newly built data centers at hub nodes exceed 80%. The Two Sessions this year established "computing - power and electricity coordination" as the key direction of new infrastructure. In terms of regional layout, MoXin has strategically expanded in the four major regions of the northwest, southwest, east, and north China, achieving large - scale applications in multiple industry scenarios and fields, highly resonating with the national macro - strategy and closely following the "East - West Computing" and "computing - power and electricity coordination".
In the northwest region, a multi - thousand - card - level inference cluster is deployed to support the intelligent transformation of traditional industries. Multiple factory security projects are implemented in scenarios such as electronic manufacturing and consumer goods production, realizing efficient and real - time AI analysis at the edge. In the southwest region, the abundant local green electricity resources are fully utilized to build a low - power - consumption green computing power pool. In the east region, a computing power cluster for high - end service industries such as bioinformatics analysis and medical health is deployed, which can significantly accelerate the gene sequencing data analysis process. It has cooperated with leading enterprises in the industry to provide high - performance AI computing power support for computationally intensive tasks such as high - throughput sequencing and protein structure prediction. In the north China region, it empowers urban governance and community intelligent upgrading, implementing visual multi - modal applications such as face recognition and pose recognition, and realizing real - time intelligent monitoring and early warning of abnormal behaviors.
This national - wide computing power network can also serve the basic large - model training and inference needs of Internet CSP manufacturers. Currently, in addition to building their own computing power, CSP manufacturers' demand for renting third - party high - quality inference computing power is continuously increasing. MoXin's multi - thousand - card clusters happen to provide a low - TCO, high - energy - efficiency computing power supply option for this market, which also opens up new opportunities for large - scale applications of the upcoming SparsePrime® computing card.
Meanwhile, MoXin has established a cooperative relationship with leading telecom operators and included the sparse computing inference solution in the operators' computing power service system. In addition, MoXin has cooperated with leading business travel hotel groups to explore application scenarios of sparse computing in hotel intelligent management. In the field of intelligent transportation, MoXin is exploring joint solutions with leading automobile manufacturers to jointly explore a new paradigm of vehicle - road coordination.
Shang Yong, the vice - president of commercialization of MoXin AI, said: "Our deployment of multi - thousand - card clusters is not just simple computing power construction. By deploying high - performance, low - TCO inference computing power nodes close to industrial clusters, we truly inject the technical advantages of sparse computing into the actual application scenarios of all industries - whether it is gene sequencing acceleration in the field of bioinformatics analysis, real - time video analysis in urban governance, or visual inspection on the intelligent manufacturing production line. Each cluster placement is to support large - scale inference needs in industry scenarios nearby, efficiently, and at low cost, making AI computing power as accessible as water and electricity."
Deeply engage in industry - academia - research: Build a moat for next - generation technologies
Beyond the rapid progress of commercialization, MoXin continuously roots technological innovation at the source. In terms of international academic cooperation, MoXin has carried out cooperation with relevant research teams at Carnegie Mellon University around key technologies such as inference acceleration, long - context services, and sparse training. It has achieved phased results in the direction of LLM sparse training and will continue to promote the transformation of large - model acceleration technology from cutting - edge research to industrial implementation. In domestic industry - academia - research cooperation, MoXin has carried out a horizontal project cooperation with the Institute of Trustworthy Embodied Intelligence at Fudan University in the direction of "semi - structured sparsity", aiming to significantly increase the model sparsity rate and improve hardware friendliness through intelligent sparse pattern search, opening up new space for cost reduction in next - generation large - model inference. Meanwhile, MoXin is promoting cooperation with Tsinghua University's CCNI Lab and SparseMind in the frontier topics of sparse computing, jointly exploring more possibilities of sparse computing theory in professional application fields. It has also established a joint laboratory for sparse computing with Hangzhou Dianzi University to explore innovative inference computing power solutions for "cloud - edge - end" coordination.
Relying on the new generation of SparsePrime® computing card, MoXin plans to conduct in - depth research on reducing inference costs and increasing efficiency in cooperation with universities in the future, accelerating the transformation closed - loop of sparse computing from academic frontiers to industrial practice. MoXin positions this as a "two - way pursuit between industrial needs and academic accumulation", hoping to break through the technical closed - loop from algorithm innovation to chip architecture and build an industry - academia - research - integrated talent ecosystem for sparse computing.
Resonance between capital and industry: Investing in the future of computing power infrastructure
This nearly one - billion - RMB Series C financing is not only a capital event but also a condensation of industrial consensus around the sparse computing technology route. The company has been supported by many well - known institutions such as Ant Group, Shenzhen Angel Mother Fund, Triumph Venture Capital, Jiangmen Venture Capital, Jinpu Investment, ZhenFund, and Cornerstone Capital. The participation of industrial capital and state - owned forces such as Shenzhen Capital Group, Liding Capital, Yunsheng Capital, Yanshan Technology, and Greater Bay Area Common Home in this round has injected full - cycle capital impetus into MoXin from early - stage innovation to large - scale fission.
It is reported that the financing funds will be mainly invested in the mass production and commercialization of the new generation of computing card SparsePrime® and the further expansion of the national computing power network map. Wang Shuaiyu, the secretary of the board of directors of MoXin and the general manager of the corporate development and capital market department, said: "Inference cost is the key bottleneck for the popularization of AI, and sparse computing is providing a fundamental answer. From an investment perspective, when evaluating the value of an AI chip company, one should not only look at the theoretical computing power of a single card but also at the effective computing power and energy efficiency ratio when completing the same AI task in a real cluster environment. MoXin's multi - location deployment and continuous customer expansion are hard - core verifications of its product strength and commercial value. We hope to become an indispensable green computing power base in the AI infrastructure layer through the combination of self - developed chips and computing power networks."
From the self - developed breakthrough in chip architecture, to the rapid development of computing power centers in the four major regions, to the large - scale penetration in multiple industry scenarios and the accelerated commercialization of next - generation products, MoXin is committed to continuously breaking through the ultimate inference cost and empowering industrial - belt computing power solutions for all industries in the context of the AI 3.0 era, pushing forward the role of the leader in sparse computing to new chapters.