GMI Cloud rents Nvidia-powered GPU clusters and inference infrastructure to AI developers and enterprises that need to train or run models without building their own computing fleet. The Mountain View company announced $668 million of new financing: a $223 million Series B led by ARCHIV and a $445 million credit facility led by CTBC. Nvidia participated in the equity round.
Customers can select preconfigured environments and launch GPU clusters instead of configuring the underlying infrastructure themselves. GMI Cloud’s Kubernetes-based Cluster Engine automates resource allocation and instance provisioning, while its broader platform covers computing, model inference and infrastructure for AI agents.
The financing structure matters. About two-thirds of the package is credit rather than equity, giving GMI Cloud more capital for a hardware-intensive expansion without an equivalent share sale but adding repayment obligations. The announcement did not disclose the facility’s maturity, pricing, collateral or covenants.
GMI Cloud operates GPU infrastructure across the United States and Asia-Pacific. It pitches that footprint as a way for customers to place workloads near their users, data and regulatory requirements, while its relationships with Taiwan’s server manufacturers provide what the company describes as a more predictable path from hardware orders to deployment.
The company says contracted annual recurring revenue tops $600 million, more than nine times its level at the end of 2025, while annual recurring revenue from services already in production has grown more than 4.5 times. Those are annualized company metrics, not recognized revenue. GMI Cloud also says its inference platform processes approximately 4 trillion tokens a week; named customers include Fireworks, Higgsfield, OpenRouter and Trend Micro. The financing is intended to add capacity in the United States, Taiwan and the rest of Asia-Pacific while supporting inference services and hiring.
Published versions remain available when an article is updated or corrected.
- Revision 1 · Initial publication
Initial publication