GMI Cloud to build $500M Nvidia-powered AI factory in Taiwan

GMI Cloud has announced plans today to build a $500 million AI Factory in Taiwan. The U.S.-based GPU-as-a-Service provider and Nvidia cloud partner noted that the project will be powered by 7,000 of Nvidia’s Blackwell Ultra GPUs supporting large-scale AI inference, training, and multimodal workloads amid a growing demand for compute across Asia. 

The AI Factory will have a capacity of 16 megawatts, with an expected deployment of 96 high-density GB300 NLV72 racks, delivering a processing capacity of up to 2 million tokens per second. GMI Cloud noted that the site will be operational by March 2026. 

TrendMicro to deploy digital-twin cybersecurity capabilities on the AI platform

According to a press release by GMI Cloud, the $500 million AI factory project aims to strengthen the trans-Pacific AI infrastructure model by combining U.S. semiconductor innovation with Taiwanese manufacturing capabilities.

GMI Cloud CEO Alex Yeh revealed that the data center project is a strategic asset for Taiwan and a roadmap towards regional AI development. He noted that GMI Cloud infrastructure helps turn the world’s AI visions into reality, with GPU utilization across the existing facilities almost complete. So far, TSMC and Foxconn are the emerging critical locations for advanced AI infrastructure. 

The GPU-as-a-Service provider stated that the AI factory will integrate Nvidia NVLink, Quantum InfiniBand, Spectrum-X Ethernet networking, and BlueField DPUs to provide high-performance and scalable compute capacity for enterprises across Asia. 

Alongside the announcement, GMI Cloud has also revealed plans to collaborate with U.S.-based Reflection AI to develop an AI platform for Taiwan. Other early customers and collaborators, including Nvidia, Trend Micro, Wistron, Chunghwa System Integration, VAST Data, and TECO, have also shown interest in the project. 

TrendMicro is deploying digital-twin cybersecurity capabilities on the platform using Nvidia AI Enterprise software. BlueField technology and GMI’s compute stack will enable the simulation of real-world threats without altering production systems. According to GMI’s press release, the collaboration will allow for continuous stress testing, policy adaptation, and real-time risk modeling. 

Winstron, on the other hand, will use the AI factory to support automated production systems powered by accelerated computing, predictive maintenance, and computer vision applications. Winstron focuses on electronic manufacturing and is also a key server assembler for Nvidia. According to GMI Cloud, the integration will enable companies to train and deploy models directly on active factory lines. 

VAST Data will also offer an exabyte-scale data infrastructure layer for the AI factory, ensuring higher throughput across the GPUs. Based on GMI Cloud, the unified data architecture provided by VAST Data is designed to support continuous AI workloads, including large-scale model training and real-time inference. 

Yeh says the project is expected to generate $1 billion once operational

TECO, which is a global industrial solutions firm, will offer energy optimization and modular data center systems. The firm’s goal is to provide an Energy-as-a-Service model that integrates engineering expertise with AI-driven digital infrastructure. 

According to Raymond Yeh, Nvidia’s Senior Vice President, AI factories represent the next stage of intelligence production, turning data into actionable insights for businesses and governments. Nvidia has also supported similar large-scale data center clusters in Saudi Arabia and South Korea as part of the global AI expansion initiative. 

According to Yeh, GMI Cloud is planning to seek an initial public offering in two to three years as it prepares to build another 50-megawatt data center in the U.S. The GPU-as-a-Service firm’s planned project in Taiwan will extend its footprint across the United States, Singapore, Thailand, and Japan.

GMI Cloud noted that the new facility in Taiwan will meet the demand and allow companies to deploy AI products at scale. The firm expects the project to generate approximately $1 billion in total contract value once fully operational, as it aims to secure $400 million from Taiwanese banks and an additional $200 million by the end of 2025. 

The smartest crypto minds already read our newsletter. Want in? Join them.