GMI Cloud Raises $223M Series B and Adds $445M Credit Facility

GMI Cloud has secured $668 million of financing, but only part of that headline figure is venture equity. The AI-infrastructure provider raised a $223 million Series B led by ARCHIV and added a $445 million credit facility led by CTBC Bank.

NVIDIA participated in the equity round alongside DSC Investment, Trend Micro, KB Investment, Kyobo Life, KT and other investors. The capital will support GPU capacity in the United States, Taiwan and other Asia-Pacific markets, expand inference services and fund hiring.

Financing snapshot

The distinction is essential: GMI Cloud did not raise a $668 million Series B. Roughly two-thirds of the package is credit, which can finance servers and data-center capacity but also creates repayment obligations.

What GMI Cloud does

GMI Cloud provides GPU infrastructure and inference services for organizations building and running AI models. Its pitch is broader than renting individual accelerators: the company combines access to compute, workload orchestration and inference software across regions where customers may have data-residency or latency requirements.

The company says contracted annual recurring revenue now exceeds $600 million, nine times its year-end 2025 level, while live annual recurring revenue has grown 4.5 times. It also says its systems process approximately four trillion tokens each week. These are company-reported operating measures, not audited revenue disclosures.

Contracted ARR is especially important to interpret carefully. It can indicate future demand, but the timing, cancellation rights, minimum commitments and conversion into recognized revenue are unknown. Investors and lenders are financing GMI Cloud on the expectation that signed demand will become deployed and utilized GPU capacity.

Why the capital structure matters

AI clouds require large upfront spending on GPUs, networking and data-center capacity. Equity absorbs early execution risk, while credit can reduce dilution and match financing with revenue-generating equipment. GMI Cloud’s package reflects that logic.

It also exposes the company to the same balance-sheet tension affecting the wider “neocloud” market. Capacity must be delivered on schedule, customers must consume what they contracted, and pricing must stay high enough to cover hardware depreciation, power, operations and interest expense.

NVIDIA’s participation is strategically useful because chip allocation and system integration are core constraints. It should not be read as a guarantee of customer demand or profitability.

The analytical read

GMI Cloud is pursuing a geographically differentiated alternative to the largest hyperscalers. Taiwan provides proximity to the semiconductor supply chain, while expansion in Japan and other markets can appeal to customers seeking sovereign or regional AI capacity.

The strategy has three major tests:

• Utilization: installed GPUs need consistently high paid usage, not merely contracted capacity.
• Funding discipline: the $445 million facility increases fixed obligations and magnifies execution risk if deployments slip.
• Software differentiation: inference tooling must create durable value beyond access to scarce hardware.

The company’s disclosed demand indicators are impressive, but they remain self-reported. The decisive evidence will be how much contracted ARR becomes live revenue, the gross margin after infrastructure costs, and the concentration of the customer base.

What to watch

Watch for facility terms, collateral details, capacity additions, customer concentration and the conversion rate from contracted to live ARR. Those disclosures will show whether the financing is building a durable AI cloud or simply accelerating a capital-intensive capacity race.