$223m of equity, $445m of debt
GMI Cloud closed $668m in new financing on 1 October, split between a $223m Series B and a $445m credit facility. The equity round was led by ARCHIV, a San Francisco firm that invests in AI and robotics, with Nvidia participating; the debt was led by the Taiwanese bank CTBC.
The split is the story. Two thirds of the money is borrowed, which is now the standard shape for a company whose business is buying accelerators and renting them out. Equity buys the company; debt buys the hardware.
Asia-Pacific investors took most of the rest of the equity. The Next Web reported that DSC Investment, Trend Micro, KB Investment, Kyobo Life and KT Corporation all participated, alongside Nvidia.
The revenue number the round is priced on
GMI Cloud says its contracted annual recurring revenue now exceeds $600m, more than nine times what it was at the end of 2025, and that production revenue grew more than 4.5 times over the same period. Its inference platform processes roughly 4 trillion tokens a week.
Those are the company’s own figures and carry the usual caveat: contracted ARR is what customers have signed for, not what has been invoiced. The named customers are mostly AI-native: Fireworks, Higgsfield, Nous Research, OpenRouter, Reflection, Cartesia, Trend Micro and Utopai Studios. Trend Micro appears as both investor and customer.
Founder and chief executive Alex Yeh framed the pitch in delivery terms: “In AI infrastructure, a delivery date is a promise.” That is a reasonable thing to lead with in a market where capacity is routinely sold before it exists.
Why Taiwan
GMI Cloud’s distinguishing feature is where its capacity is. The money goes to expanding GPU capacity in the United States, Taiwan and the wider Asia-Pacific, and the debt came from a Taiwanese bank rather than a US private credit fund.
That matters because almost every other large GPU cloud financing this year has been American money buying American capacity. A Taiwanese bank underwriting $445m against accelerators is a different pool of capital entering the same trade, and it is the pool sitting closest to where the chips are manufactured.
What to watch is the ratio. A company borrowing twice its equity against hardware that depreciates is making the same bet Lambda, CoreWeave and now Amazon are making in their different ways, and the thing that settles it for all of them is what a two-year-old accelerator is worth when the lease ends.
