If you're building artificial intelligence infrastructure in Asia, you've got some well-funded competition. Nvidia (NVDA) Cloud Partner GMI Cloud just announced plans for a $500 million AI factory in Taiwan, and the specs are pretty impressive.
The facility will function as critical infrastructure for the region, giving enterprises the ability to train and deploy AI models at serious scale. Once it's up and running, the site is expected to handle nearly 2 million tokens every second. That's the kind of processing power that makes AI researchers smile.
The announcement puts GMI Cloud in the same competitive arena as Google, Amazon (AMZN), and Microsoft (MSFT) in the race to dominate Asia's cloud computing landscape, according to Bloomberg. Not bad company for a startup to keep.
Inside the Factory: 7,000 Nvidia Blackwell Ultra GPUs Across 96 Racks
The Mountain View, California-based startup's 16-megawatt AI factory will run on 7,000 Nvidia Blackwell Ultra GPUs distributed across 96 high-density GB300 NVL72 racks, the company said in its Nov. 17 announcement.
The infrastructure also incorporates Nvidia's NVLink, Quantum InfiniBand, and Spectrum-X Ethernet networking, along with BlueField data processing units. GMI Cloud designed the facility specifically for large-scale inference, fine-tuning, and multi-modal workloads. In other words, this thing is built to handle the heavy lifting that modern AI applications demand.
"This data center is intended to become the blueprint for the heart of Asia's AI future," GMI Cloud CEO Alex Yeh said in the company's statement. "With thousands of next-generation NVIDIA GPUs running in synchrony, our AI infrastructure helps turn the world's AI visions into reality."
Trend Micro, Wistron, and TECO Among First Customers
GMI Cloud already has several prominent companies committed to the first wave of work at the new Taiwan site, which is a good sign for a facility that hasn't even opened yet.
Trend Micro plans to run cyber simulations on the system, modeling threats in a controlled digital environment. Wistron, which builds hardware for global clients and holds an early stake in GMI Cloud, intends to create new factory tools using computer vision, predictive upkeep, and digital replicas of production lines.
TECO Electric & Machinery will support power needs and broader system setup while exploring energy service products that depend on the computing power inside the facility.
"AI factories are where intelligence is produced — turning data into insight and innovation for the future," Nvidia Senior Vice President of Asia Pacific Raymond Teh said in GMI Cloud's statement. "GMI Cloud's AI factory infrastructure will help continue the region's leadership in AI infrastructure and innovation."
Partnership With $8 Billion Startup Reflection AI
In a separate announcement on Nov. 20, GMI Cloud said it entered a new collaboration with Reflection AI, a research group focused on large-scale autonomous systems. Reflection AI reached an $8 billion valuation after a $2 billion capital infusion, according to media reports.
Reflection AI plans to run advanced model training on GMI Cloud's GPU clusters in the U.S., with both groups studying ways to expand their work across large AI factory projects and sovereign-level AI programs. Sovereign computing has become a hot topic as countries seek to maintain control over their AI infrastructure and data.
"Contributing our support and expertise to the future of cutting-edge innovation like Reflection AI is what makes us the global leader in computing infrastructure," Yeh said in the company's statement. "Our mission is simple: ensure every AI and ML company that partners with us wins their market."
The Taiwan facility represents a significant bet on Asia's AI future, and GMI Cloud is positioning itself as a serious alternative to the hyperscale cloud providers that have dominated the market. Whether enterprises will choose a specialized GPU-as-a-Service provider over established giants remains to be seen, but with this kind of infrastructure investment, GMI Cloud is making sure it has the hardware to compete.