Particle.news
Download on the App Store

GMI Cloud Unveils $500 Million Nvidia-Powered AI Factory in Taiwan Targeting March 2026 Launch

The project signals a push for sovereign AI capacity in Taiwan within a wider wave of Nvidia-driven infrastructure.

Overview

  • The Taiwan facility will run on Nvidia Blackwell GB300 GPUs, deploying about 7,000 accelerators across 96 high-density racks with an estimated 16 megawatts of power draw.
  • GMI targets service availability by March 2026 with capacity near 2 million tokens per second for large-scale inference, fine-tuning and multimodal workloads.
  • The architecture features Nvidia NVLink, Quantum InfiniBand, Spectrum‑X Ethernet and BlueField DPUs to enable high-throughput, enterprise-grade computing.
  • Initial customers and partners include Nvidia, Trend Micro, Wistron, Chunghwa System Integration, VAST Data and TECO, with early use cases in cybersecurity and smart manufacturing.
  • GMI projects about $1 billion in total contract value at full ramp and plans a separate 50‑megawatt U.S. data center alongside an IPO timeline of two to three years.