"MetaX Co-Launches Shanghai Cube, a New Generation of Domestic, Self-Contained, Full-Stack High-Density Server Rack | Xinwen NEWS"
Build a benchmark project for fully autonomous, intelligent satellite production lines in China's commercial space sector, with complete supply-chain independence.
On March 21, 2025, MetaX, a portfolio company of Heart Capital, jointly unveiled the Shanghai Cube — a next-generation, domestically developed, full-stack high-density AI server rack — alongside multiple partners.
Shanghai Cube is a full-stack, software-hardware-integrated AI datacenter-grade product. Designed to address China's core strategic needs in AI computing power, it brings together leading technology companies from Shanghai to collaboratively build a benchmark product that reflects the latest trends in the compute sector.
Full-Stack Optimization Sets a New Standard for Intelligent Computing
The first-generation Shanghai Cube features a customized OAM module built on MetaX's C550 series high-performance GPGPU chips, deployed in a high-density, single-rack configuration with 128 liquid-cooled cards. The system integrates switch equipment powered by domestically developed switch chips from CloudSilicon, a customized domestic operating system from DaoCloud, and volume manufacturing from Luxshare Precision Industry Co., Ltd. Infini-AI, Infinigence AI, and other teams participated throughout in the joint optimization of the software-hardware integration, while Fudan University and the Innovation Academy contributed model fine-tuning. Additionally, Shanghai Cube incorporates domestic liquid cooling systems, domestic high-performance parallel storage systems, domestic retimers, domestic motherboard capacitors, and other locally sourced components. The next generation plans to introduce domestic optical interconnect solutions and domestic high-speed NICs.
Industry-Academia-Research Collaboration Validates Technical Capabilities
The first Shanghai Cube prototype — comprising compute nodes, management nodes, storage nodes, and network nodes with full software and hardware stacks — has been installed at Fudan University's Zhangjiang campus in Shanghai. Benchmark data shows significant integration and optimization advantages. The system has already achieved efficient inference with DeepSeek's full-parameter 671B model, as well as training and fine-tuning support for other mainstream large models. Shanghai Cube has completed initial compatibility testing with multiple scientific computing platforms and compute scheduling platforms.
The deep software-hardware integration and specialized architectural design deliver further improvements in compute performance, communication efficiency, storage access, and software environment compared to standard compute nodes, while the customized liquid cooling solution optimizes thermal management and power consumption. These gains in performance and efficiency, combined with reduced production and energy costs, create substantial room for improved cost-performance ratios — savings that can be reinvested into tackling targeted hardware and software challenges, forming a virtuous cycle of product development and iteration.
Industrial Ecosystem Accelerates Domestic Substitution
At NVIDIA's GTC event from March 17–19 this year, the company unveiled its GB300 NVL72x4, a 288-card high-density rack product, with plans to eventually release configurations scaling up to 576 cards per rack. Shanghai Cube represents one of the closest domestic equivalents in terms of high compute-density rack design. It has been described as "the Apple of the compute industry" for a new generation. Going forward, as technologies such as optical interconnects, heterogeneous stacking, multi-card interconnects, and disaggregated inference architectures advance, Shanghai Cube will continue iterating its product forms and solutions to provide the market with optimal full-stack, software-hardware-integrated offerings.
As a domestic SuperPod-class product positioned against NVIDIA's NVL72 series, the name "Shanghai Cube" captures both the collaborative, forward-looking spirit of China's domestic compute supply chain and the aspiration of Shanghai-based compute industry players to deliver world-class products — "Shanghai-made, guaranteed quality." This carries significant strategic weight given external constraints on domestic chip manufacturing processes due to geopolitical factors and bottlenecks in per-card compute density. The industrial ecosystem and partnership model of Shanghai Cube have been widely welcomed and praised by domestic compute hardware and software vendors.
Founded in 2022, Heart Capital is an early-stage Chinese venture capital fund focused on technology and digitalization. The team is led by Yan Han, founding partner of Lightspeed China, alongside core investors, a CFO, and seasoned investors with deep industry experience. Notable past investments include Series A stakes in Xpeng Motors (NYSE: XPEV, 09868.HK), Full Truck Alliance (NYSE: YMM), as well as FinVolution (NYSE: FINV), RoboSense (02498.HK), Baichuan, Manman Lengyun, Fan Deng Reading, World Logistics, Micro-nano Starry Sky, LandSpace, Lanhu, Starfield, and others. Rooted in China with a global outlook, Heart Capital seeks true value in non-consensus opportunities. The firm honors the value of people and champions the potential of the human spirit, aspiring to support more young Chinese entrepreneurs in strengthening China and expanding onto the world stage.
