GMI Cloud Raises $668M for Global AI Infrastructure

GMI Cloud has secured $668M in new financing to expand AI infrastructure across the United States, Taiwan, and the rest of Asia-Pacific. The October 1, 2026 package combines $223M of Series B equity led by ARCHIV with a $445M credit facility led by CTBC, making it a blend of growth capital and infrastructure financing rather than a $668M equity round.

NVIDIA joined the equity financing alongside DSC Investment, Trend Micro, KB Investment, Kyobo Life, and KT Corporation. The capital is meant to turn contracted demand into live GPU capacity, which matters because an AI cloud customer may build launches, hiring, and revenue plans around a delivery date long before a cluster begins serving traffic.

What GMI Cloud Raised

The official financing announcement puts the exact package at $668M. ARCHIV led the $223M Series B equity portion, NVIDIA participated, and CTBC led the separate $445M credit facility. GMI Cloud did not disclose a valuation, draw schedule, collateral package, or detailed lending terms.

The investor list reaches across the hardware and enterprise ecosystem. NVIDIA supplies the accelerated-computing platform at the center of GMI Cloud's infrastructure, while Trend Micro is both an investor and a named customer. DSC Investment, KB Investment, Kyobo Life, and KT Corporation add South Korean institutional and strategic capital to a company positioning itself between U.S. AI demand and Asia-Pacific infrastructure.

GMI Cloud previously announced an $82M Series A package in October 2024. That financing was led by Headline Asia, with Banpu Next and Wistron participating, and Axios reported that it consisted of $15M in equity and $67M in debt. Across the two announcements, GMI Cloud has disclosed $750M of blended equity and debt or credit capacity since 2024, not $750M of equity funding.

Delivery Has Become Part of the Product

Founder and CEO Alex Yeh framed infrastructure delivery as a promise because customers plan product launches, hiring, and revenue around capacity. That observation gets closer to the commercial product than a list of GPU models does. The customer is buying compute, but it is also buying confidence that power, networking, cooling, hardware, and software will arrive as one usable system when the business needs it.

GMI Cloud says contracted annual recurring revenue has topped $600M and reached more than 9x its year-end 2025 level. It also reports that live ARR in production has grown more than 4.5x over the same period and that its inference platform processes about 4T tokens per week. These are company-reported measures, and contracted ARR is a commercial commitment rather than recognized revenue, but together they explain why GMI Cloud is raising equity and arranging credit at the same time.

The customer roster provides additional operating context. GMI Cloud names Fireworks, Higgsfield, Nous Research, OpenRouter, Reflection, Cartesia, Trend Micro, and Utopai Studios as customers. In the company release, Fireworks co-founder Chenyu Zhao described GMI Cloud as a strong and reliable provider across NVIDIA GB200 and GB300 NVL72 systems, tying the financing story to equipment that customers say they are already using.

The Capital Structure Matters

About two-thirds of the new package is a credit facility. That structure fits the economics of a neocloud, where large amounts of capital must be committed to servers and data-center capacity before utilization turns the assets into revenue. Equity can fund products, talent, and organizational growth, while credit can finance hardware against a more explicit schedule of repayment and operational performance.

The arrangement also leaves less room for sloppy capital accounting. The $445M facility is financing capacity, not a new equity valuation, and the public announcement does not say how much has been drawn. GMI Cloud's challenge is to convert contracted demand into live workloads quickly enough that the infrastructure begins earning before financing costs and hardware depreciation claim too much of the margin.

That pressure is familiar to GMI Cloud. Its 2024 Series A package already paired equity with debt, only at a much smaller scale. The 2026 round expands the same operating model while asking the company to coordinate more hardware, more geographies, and more customer commitments without letting a larger balance sheet become an excuse for late delivery.

What GMI Cloud Is Building

GMI Cloud describes itself as an AI-native inference cloud rather than a simple GPU rental service. Its current platform spans model APIs and inference services, Kubernetes-based orchestration, dedicated and on-demand GPU compute, and hardware deployed in company-operated data centers. Its infrastructure pages list NVIDIA H100, H200, B200, GB200 NVL72, GB300 NVL72, and Vera Rubin platforms.

The company reports more than 30,000 GPUs, 300+ AI-team customers, 99.99% platform availability, and up to 3.7x efficiency gains. It also says enterprise deployments can use SOC 2 and ISO 27001-compliant infrastructure. Those figures and certifications are presented by GMI Cloud, so they should be read as company claims rather than independent performance audits.

The financing will add GPU capacity in the U.S., Taiwan, and the rest of APAC, continue development of inference services, and support strategic hiring. It builds on a Taiwan AI Factory announced in 2025 and a $12B, 1GW sovereign-AI infrastructure initiative in Japan announced in March 2026. The Japan project is separate context and is not part of the $668M financing total.

Why Taiwan Sits Inside the Investment Thesis

GMI Cloud argues that its Taiwan relationships provide a shorter path from server manufacturing to deployed clusters. That matters because the AI infrastructure market is increasingly constrained by the coordination of chips, systems, power, cooling, networking, permits, and customer schedules. Access to a GPU allocation is useful, but a customer earns nothing until the complete cluster is installed, tested, and available.

The company's geographic pitch is designed around that handoff. U.S. AI companies want capacity near users and supply chains in Asia, while APAC enterprises and governments want infrastructure closer to local data and compliance requirements. GMI Cloud is trying to serve both sides through one operating platform instead of treating sovereign and commercial deployments as unrelated businesses.

What the Financing Still Has to Prove

The next evidence will come from conversion, not announcement size. Investors, customers, and lenders will need to watch how much contracted ARR becomes live revenue, whether customer concentration remains manageable, how quickly new capacity reaches production, and what utilization looks like after the hardware arrives. The facility's draw schedule, security package, and covenants also remain undisclosed.

GMI Cloud now has capital, supplier relationships, a named customer base, and a large set of infrastructure commitments moving at the same time. Alex Yeh's promise about delivery dates will be tested cluster by cluster, where manufacturing precision has to become customer capacity before somebody else's launch calendar reaches the date GMI Cloud agreed to keep.