Magic-Spring AI
Category: AI Infrastructure
Chinese AI infrastructure company building the world's leading "Token Superfactory" — producing high-quality AI tokens at scale through hardware-software co-optimization and supernode clusters. Magic-Spring AI was founded in 2024. The company is led by Xu Lingjie (徐凌杰). Based in Shanghai, China. Team size: 11-50. Total funding raised: $50.0M. Key investors include 达泰资本 (Data Capital), 云启资本 (Yunqi Capital), 联新资本 (Lianxin Capital), 香港科技园创投基金 (HKSTP Venture Fund), 上海半导体产投 (Shanghai Semiconductor Investment), 永兴材料 (Yongxing Materials), 邦盛资本 (Bangsheng Capital), 亿合资本 (Yihe Capital).
- Founded
- 2024
- Headquarters
- Shanghai, China
- Team size
- 11-50
- Total funding
- $50.0M
Value proposition
Magik Compute operates as a "Token Superfactory" that vertically integrates energy, chips, infrastructure, and large models to produce and sell AI tokens directly to customers. Unlike traditional GPU cloud providers or pure software optimization companies, they combine cluster-level hardware system design (supernode clusters) with LLM optimization expertise to deliver high-quality, cost-efficient tokens at scale.
Products and solutions
1) Magik Token Factory - Token production service with 100B+ token throughput, end-to-end monitoring, multimodal support, 2) MagikCloud Compute Platform - Cloud computing and API services covering mainstream AI chips with fine-tuning and inference capabilities, 3) Supernode Cluster - Self-developed high-density hardware with HVDC power supply, liquid cooling, and HW-SW co-optimization
Unique value
One of the few companies globally with dual expertise in cluster-level hardware system design AND LLM optimization, enabling them to deliver tokens with superior cost-performance through full-stack hardware-software co-design. They directly produce and sell tokens (not just rent compute), positioning as a token manufacturer rather than a traditional cloud/AI Infra provider.
Target customer
AI application developers, enterprises deploying LLMs at scale, AGI companies needing high-volume token production, and organizations requiring optimized inference infrastructure
Industries served
AI/AGI, Enterprise AI, Cloud Computing, Large Language Model Deployment
Technology advantage
Hardware-software co-design expertise spanning GPU architecture (CEO ex-NVIDIA/AMD), cloud infrastructure (ex-Alibaba Cloud), and LLM optimization (CTO ex-Graphcore). Self-developed supernode cluster technology with high-density integration, HVDC power, and liquid cooling. Team combines chip-level, system-level, and model-level expertise rarely found in a single company.
How they differentiate
Unlike pure software AI Infra companies (e.g., SiliconFlow) or pure hardware/cloud providers, Magik Compute owns the full stack from hardware system design (supernode clusters) to model optimization, producing tokens as a finished product. They are vertically integrated — designing their own server clusters, optimizing the software stack, and selling tokens directly rather than just renting compute or selling optimization services.
Main competitors
SiliconFlow (硅基流动), Qingcheng Jizhi (清程极智), Qujing Technology (趋境科技)
Key partnerships
菲沃泰 (Favored Tech, SHA: 688371) - AI partnership for LLM deployment and GPU computing platform, Hong Kong Science & Technology Parks Corporation (HKSTP) - strategic partnership for international expansion, Sunrise - strategic partnership to build AI inference ecosystem (Jan 2026)
Notable customers
菲沃泰 (Favored Tech) - publicly announced customer for AI model deployment
Major milestones
Founded June 2024 by ex-Biren Technology president Xu Lingjie, Angel round closed Sep 2025 with Yunqi Capital, Lianxin Capital, HKSTP, Pre-A round of several hundred million RMB closed May 2026 led by Data Capital, Launched Token Factory, MagikCloud, and Supernode Cluster products, Achieved 100B+ token throughput in production
Growth metrics
100B+ Token throughput validated in production; completed 3 funding rounds in 2 years; business增速显著 (significant growth rate per investor statements)
Market positioning
Early-stage but fast-growing player in China's AI Infrastructure (AI Infra) layer, positioned as a "Token manufacturer" rather than a traditional compute provider. Competes in the rapidly expanding Chinese AI token economy alongside other Infra startups. Backed by top-tier VCs (Data Capital, Yunqi Capital, Lianxin Capital) and Hong Kong Science Park.
Geographic focus
China (primary), with Hong Kong as a strategic base for international expansion
Patents and IP
CEO Xu Lingjie holds multiple US patents in GPU computing and neural network optimization (US 11410016, US 11223838, US 10061591)
About Xu Lingjie (徐凌杰)
Ex-NVIDIA GPU Architect; Ex-AMD; Ex-Samsung North America Research; Ex-Alibaba Cloud AI Architecture Director; Co-founder & President of Biren Technology (壁仞科技, listed on HKEX ~HKD 124B market cap). Education: Shanghai Jiao Tong University (BSEE), UT Austin (MS Comp Eng), UC Berkeley Haas (MBA).
Official website: https://magikcompute.ai