GMI Cloud Raises $668 Million as Contracted ARR Grows More Than Ninefold Since Year-End

GMI Cloud, a Mountain View-based provider of GPU infrastructure and inference services, said it has raised \$668 million in new financing - a hybrid structure that pairs \$223 million in equity with a \$445 million credit facility led by Taiwan's CTBC. The credit line is roughly twice the size of the equity check, an unusual ratio that says as much about the state of AI infrastructure lending as it does about the company.

The Series B was led by ARCHIV, a new San Francisco investment firm specializing in AI and robotics, with NVIDIA participating. The round drew heavy participation from Asia-Pacific investors, including DSC Investment, Trend Micro, KB Investment, Kyobo Life Insurance and KT Corporation. The Korean contingent is notable: KB Investment, DSC, Kyobo Life and KT are buying into a US-based cloud as a way to secure a position in the GPU layer of the AI value chain.

The growth figures are striking, and they are the company's own. GMI Cloud says contracted annual recurring revenue has grown more than ninefold from its level at the end of 2025, while live ARR from services actually in production has grown more than 4.5 times over the same period. Its inference platform now processes approximately 4 trillion tokens per week. Named customers include Fireworks, Higgsfield, Nous Research, OpenRouter, Reflection, Cartesia, Trend Micro and Utopai Studios. None of these figures has been independently audited.

The money goes toward GPU capacity in the United States, Taiwan and the rest of Asia-Pacific, building on the company's Taiwan AI Factory - announced in 2025 as its first Asian facility - and a Japan sovereign AI initiative unveiled earlier this year. The rest is earmarked for inference services and hiring.

GMI Cloud's pitch is that compute demand is no longer regional. US AI companies and hyperscalers need capacity in Asia to serve Asian users quickly; Asian enterprises need to run production AI domestically, under local data and compliance rules. Most AI clouds are built for one side of that equation. The company also leans on supply chain: close relationships with Taiwanese manufacturers, who build most of the world's AI servers, give it a more predictable path from order to deployment.

"In AI infrastructure, a delivery date is a promise," said Alex Yeh, founder and CEO of GMI Cloud. "Customers plan launches, hiring and revenue around it." Chenyu Zhao, co-founder of Fireworks, said GMI Cloud had been one of his company's strongest and most reliable providers across NVIDIA GB200 and GB300 NVL72 systems - a useful endorsement, given that supply allocation is the scarce commodity in this market.

The structure of the round is the tell. A \$445 million debt facility means lenders are now willing to underwrite revenue backed by compute, treating contracted GPU capacity as a bankable asset rather than a speculative bet. NVIDIA's presence as an investor reinforces the same link from the other direction, tying GPU supply to cloud capacity.

What it means: the capital is moving to whoever can promise delivery dates, and the revenue is coming from inference rather than training. That is a different business from the one that funded the first wave of AI clouds - and a far more credit-friendly one.