Skip to main content

Iluvatar CoreX

Category: AI Chips / Semiconductors

A leading Chinese semiconductor company specializing in high-performance General-Purpose Graphics Processing Units (GPGPU) designed for AI cloud training and inference. Iluvatar CoreX was founded in 2015. The company is led by Gai Lujiang. Based in Shanghai, China. Team size: 800+. Total funding raised: $1.4B+. Latest round: Public Listing (HKEX: 9903.HK). Key investors include Centurium Capital, Princeville Capital, HOPU Investment, China Unicom Capital, CedarLake Capital, VMS Group, Gatelanes.

Founded
2015
Headquarters
Shanghai, China
Team size
800+
Total funding
$1.4B+

Value proposition

Provides high-performance, domestically-developed GPGPU alternatives to international incumbents, offering high power efficiency and full-stack software compatibility for AI and HPC workloads.

Products and solutions

Tiangai (Big Island / BI) Series: China's first 7nm cloud training GPGPU chip with up to 147 TFLOPS FP16 performance., Zhikai (Deep Flux / MR) Series: High-performance cloud inference chips for diverse AI deployment scenarios., SkyRiver (Software Stack): A proprietary software platform including SDKs and compilers compatible with mainstream AI frameworks., DeepSpark: An open-source AI and HPC application development and benchmarking platform., SkyAXE / SkyACE: Integrated AI computing and edge computing appliances.

Unique value

The first Chinese company to achieve dual mass production of both 7nm training and inference GPGPUs, breaking international monopolies in high-end AI compute hardware.

Target customer

Cloud service providers, AI data centers, internet giants, government research institutions, and large-scale enterprises in finance and healthcare.

Industries served

Cloud Computing, Finance & Fintech, Healthcare & Medical Imaging, Smart Cities & Security, Energy & Smart Manufacturing, Education & Scientific Research

Technology advantage

Proprietary GPGPU architecture designed for high performance-per-watt; deep software-hardware synergy that ensures compatibility with mainstream ecosystems like PaddlePaddle and DeepSeek without requiring extensive code migration.

How they differentiate

First Chinese company to achieve dual mass production of both 7nm cloud training and inference GPGPUs. Differentiates through its proprietary 'Big Island' (Tiangai) architecture and 'SkyRiver' software stack, which offers high compatibility with mainstream AI frameworks like PyTorch and TensorFlow without significant code migration.

Main competitors

Biren Technology, Moore Threads, Enflame Technology (Suiyuan), NVIDIA

Key partnerships

Baidu (PaddlePaddle framework adaptation), New H3C (Strategic server and infrastructure cooperation), DeepSeek (LLM model hardware adaptation), LongXi Community / OpenAnolis (OS and ecosystem integration), Gitee AI (Cloud-based model service partnership)

Notable customers

Baidu, New H3C, Inspur, DeepSeek, China Unicom

Major milestones

Mass production of China's first 7nm GPGPU training chip 'Tiangai 100' in 2021, Achieved dual mass production of training and inference chips in 2022, Listed on the Hong Kong Stock Exchange (HKEX) on January 8, 2026

Growth metrics

Achieved a revenue CAGR of nearly 70% from 2022 to 2024; total compute cards shipped exceeded 50,000 units by late 2025.

Market positioning

Tier-1 domestic GPGPU leader providing 'controllable' high-performance compute alternatives for the Chinese cloud and AI infrastructure market.

Geographic focus

Mainland China (primary market for domestic replacement), with strategic expansion via the Hong Kong capital market.

Patents and IP

Holds hundreds of self-developed patents covering GPGPU architecture, IP cores, and full-stack software algorithms (proprietary IP cores for training and inference).

About Gai Lujiang

Gai Lujiang is the current Chairman and CEO of Iluvatar CoreX, having succeeded Diao Shijing to lead the company's capital market initiatives and Hong Kong IPO. Under his leadership, the company has solidified its position as China's premier GPGPU developer.

Official website: