ZhenFund Early-Stage Portfolio Company Moffett AI Closes Nearly One Billion Yuan Series C Round

Moffett AI has officially completed its Series C funding round, raising nearly one billion RMB. The proceeds will be directed primarily toward mass production and commercialization of its next-generation compute card, SparsePrime, and further expansion of its national compute network footprint.

ZhenFund participated in Moffett AI's Pre-A round in 2021 and has backed the company ever since.

Alongside the funding announcement, the company revealed that its core product — the next-generation SparsePrime compute card — will officially launch later this year. SparsePrime is a high-performance, general-purpose AI inference accelerator designed for intelligent computing centers and data centers. Built on Moffett AI's self-developed Antoum2.0 chip architecture, it is purpose-optimized for large language models and complex inference workloads. The product adopts a top-down design philosophy, broadly supports mainstream Transformer models with enhanced general compatibility, and comes with a comprehensive toolchain that enables zero-friction adoption and rapid access to sparse acceleration benefits.

Developers can migrate their existing PyTorch and TensorFlow model code, as well as efficient inference frameworks like vLLM, with virtually zero code changes for direct deployment. The platform also supports custom operator development using Triton, minimizing the barrier to entry. Drawing on real workload data accumulated from kilo-card cluster deployments across multiple compute centers, SparsePrime aims to achieve new breakthroughs in sparse compute efficiency, further solidifying Moffett AI's differentiated competitiveness in AI inference and demonstrating a viable technical path to doubling effective compute without precision loss.

SparsePrime's confidence stems from Moffett AI's sustained technical accumulation in sparse computing. Previously, the company's S30 and S40 compute cards won three consecutive championships at MLPerf™ Inference, the authoritative international AI benchmark, demonstrating leading energy efficiency and per-unit inference throughput across mainstream tasks in computer vision, natural language processing, and large models. Achieving superior inference performance at power consumption significantly below industry flagship products, these results have fully validated the engineering feasibility and commercial value of sparse computing under real data center workloads.

Commercialization in Full Swing

Technical value is finding resonance in industry adoption as well. Moffett AI has progressed from single-project validation to "multi-regional kilo-card cluster deployments" nationwide. Its inference clusters, built on self-developed sparse computing technology, are becoming the core compute foundation for intelligent computing centers in multiple key regions, further realizing a differentiated technical roadmap of precision-preserving compute upgrades.

Currently, the 15th Five-Year Plan emphasizes that the digital economy's core industries should account for 12.5% of GDP. The "East Data, West Computing" initiative requires that new data centers at hub nodes source over 80% of their power from green electricity, and this year's Two Sessions elevated "compute-power synergy" as a critical direction for new infrastructure. In regional deployment, Moffett AI has strategically expanded across four major zones — Northwest, Southwest, East China, and North China — achieving scaled applications across multiple industries and scenarios, resonating closely with national strategic priorities around "East Data, West Computing" and "compute-power synergy."

In the Northwest, kilo-card inference clusters support intelligent transformation of traditional industries, with multiple factory security projects deployed in electronics manufacturing and consumer goods production, enabling efficient real-time AI analytics at the edge. The Southwest leverages abundant local green power resources to build low-power, sustainable compute pools. East China hosts clusters targeting bioinformatics analysis, healthcare, and other high-end services, dramatically accelerating gene sequencing data analysis workflows. The company has partnered with industry leaders to provide high-performance AI compute for throughput sequencing, protein structure prediction, and other compute-intensive tasks. North China empowers urban governance and community intelligence upgrades, deploying facial recognition, pose estimation, and other visual multimodal applications for real-time intelligent monitoring and anomaly detection.

This nationwide compute network also serves the foundational large model training and inference needs of internet CSPs. Currently, beyond self-built infrastructure, CSPs are seeing growing demand for third-party, high-quality inference compute. Moffett AI's kilo-card clusters precisely address this market with low-TCO, energy-efficient compute supply options, opening new opportunities for scaled application of the upcoming SparsePrime card.

Meanwhile, Moffett AI has established partnerships with leading telecom operators, integrating its sparse computing inference solutions into operator compute service offerings. The company is also collaborating with a major business travel hotel group to explore sparse computing applications in intelligent hotel management. In smart mobility, Moffett AI is jointly developing solutions with a leading automaker to explore new paradigms in vehicle-road coordination.

Shang Yong, Moffett AI's VP of Commercialization, stated: "Our kilo-card cluster deployment isn't simply about building compute capacity. By positioning high-performance, low-TCO inference nodes close to industrial clusters, we're injecting sparse computing's technical advantages directly into real-world applications across industries. Whether accelerating gene sequencing in bioinformatics, real-time video analytics in urban governance, or visual inspection on smart manufacturing lines — every cluster placement is designed to support large-scale inference needs locally, efficiently, and cost-effectively, making AI compute as accessible as water and electricity."

Deepening Industry-Academia-Research: Fortifying the Next-Generation Technical Moat

Beyond its commercial momentum, Moffett AI continues to root its technological innovation at the source. In international academic collaboration, the company is working with research teams at Carnegie Mellon University on key technologies including inference acceleration, long-context services, and sparse training. Its LLM sparse training initiative has achieved preliminary results, with continued efforts to advance large model acceleration from frontier research toward industrial deployment.

In domestic industry-academia partnerships, Moffett AI has launched a horizontal research program with Fudan University's Trustworthy Embodied Intelligence Institute on "semi-structured sparsity," aiming to dramatically increase model sparsity rates and hardware friendliness through intelligent sparse pattern search — opening new space for next-generation large model inference cost reduction. Simultaneously, the company is advancing collaboration with Tsinghua University's CCNI Lab and SparseMind on frontier sparse computing research, jointly exploring theoretical possibilities for sparse computing in specialized application domains. A joint sparse computing laboratory has also been established with Hangzhou Dianzi University to explore innovative "cloud-edge-device" collaborative inference solutions.

Building on the next-generation SparsePrime compute card, Moffett AI plans to conduct in-depth research with universities on inference cost reduction and efficiency improvement, accelerating the closed-loop transformation of sparse computing from academic frontier to industrial practice.

Moffett AI frames this as "a two-way journey between industry demand and academic accumulation," aiming to close the technical loop from algorithmic innovation to chip architecture and build an integrated industry-academia-research talent ecosystem for sparse computing.

Capital and Industry in Resonance: Investing in the Future of Compute Infrastructure

This nearly one billion RMB Series C round is not merely a capital event, but a crystallization of industry consensus around the sparse computing technical path.

Shuaiyu Wang, Moffett AI's Board Secretary and General Manager of Corporate Development & Capital Markets, stated: "Inference cost is the critical bottleneck for AI普及. Sparse computing is providing a fundamental answer. From an investment perspective, valuing an AI chip company shouldn't rely solely on theoretical single-card compute — what matters is effective compute and energy efficiency in real cluster environments for equivalent AI tasks. Moffett AI's multi-site deployments and customers' continued capacity expansion are hard validation of product strength and commercial value. We aim to become an indispensable green compute foundation in the AI infrastructure layer through the combination of self-developed chips and compute networks."

From breakthroughs in self-developed chip architecture, to the spreading footprint of four regional compute centers, to scaled penetration across multiple industries and accelerated commercialization of next-generation products, Moffett AI is committed to pushing the boundaries of extreme inference cost reduction and empowering compute solutions for industrial belts across all sectors in the AI 3.0 era.

Moonshot AI｜Hero Games｜BianSai Technology

RightBrain AI｜Vastech Energy｜UZIS

Helium Starlight｜Yahaha｜SmartMore Perfect Diary｜Jingying Technology｜Hesai Technology

Youibot｜Zero Zero Robotics｜CoEvolution NuMiao｜KnowYourself｜Manifold Tech Mogic AI｜Xiaohongshu｜Black Lake｜YuanGu

Yuguang Technology｜StarDetect｜Qingyue Technology CellX｜Yunzhou-Tech｜GuNa Technology Di-Matrix｜Weiming Shiguang｜Xellar Biosystems