30 Sep 2026
GMI Cloud raises $668M to expand GPU cloud across the U.S. and Asia
Mountain View AI-cloud company GMI Cloud said Wednesday it raised $668 million in new financing — $223 million Series B equity led by ARCHIV with NVIDIA participating, plus a $445 million credit facility led by CTBC — to expand GPU capacity in the United States, Taiwan, and Southeast Asia.
AI labs and apps still live or die by whether GPUs show up on time on both sides of the Pacific. GMI Cloud’s $668 million package, with NVIDIA on the equity side and a large Asia-bank credit line, is a bet that one operator can keep filling clusters from Taiwan’s supply chain into U.S. and Asia-Pacific demand. That is infrastructure money, not another chatbot launch.
On Wednesday, 30 September 2026, Mountain View-based GMI Cloud announced $668 million in new financing. The package is $223 million in equity for a Series B, plus a $445 million credit facility led by CTBC. A Series B is a later round of venture equity: investors put money in for a share of the company, after earlier funding. A credit facility is a loan line the company can draw. The two pieces add to $668 million. The wire headline says the company raised over $660 million. The body names the precise total as $668 million. The company site banner the same day says GMI Cloud raises $668 million, led by ARCHIV, with participation from NVIDIA. The dateline is Mountain View, Calif. The Business Wire dateline is dated Sep 30, 2026, and does not print an hour on that line. A Financial Content reprint of the same wire stamps the item September 30, 2026, at 1:56 p.m. Eastern. Those figures are GMI Cloud’s.
Who put the money in, as the release states it. The Series B was led by ARCHIV, a new investment firm based in San Francisco that specializes in AI and robotics. NVIDIA participated. Participation, in that sentence, means NVIDIA put money into the equity round. The release does not say NVIDIA will ship a GMI Cloud product, and it does not name NVIDIA as the operator of the clusters. The round also names investors across the Asia-Pacific: DSC Investment, Trend Micro, KB Investment, Kyobo Life, KT Corporation, and others. The release says “and others.” It does not print a full list. The $445 million credit facility is led by CTBC. The release does not spell out the letters CTBC, and it does not say how much of the facility is already drawn. Those lines are GMI Cloud’s.
What the money is for. The release says the funding is growth capital for capacity in the United States, Taiwan, and the rest of the Asia-Pacific. The bullet under the headline names the United States, Taiwan, and Southeast Asia. Capacity, here, means more graphics processing units, the chips that do the AI math, available for customers to rent. The new capacity builds on the company’s Taiwan AI Factory, which it says it announced in 2025 as its first facility in Asia, and on a Japan sovereign-AI initiative it says it announced earlier this year. Sovereign AI, in that sentence, is the company’s name for the Japan effort. The release does not describe the Japan project beyond that name. Beyond the buildings and the chips, the Series B is also for the continued growth of GMI Cloud’s inference services and for strategic hiring as the company scales. Inference is running a trained model so it can answer, as distinct from the earlier work of training it. The release does not print a headcount, a hiring target, or a date the new capacity will open. Those lines are GMI Cloud’s.
The company ties the round to a stretch of commercial growth, and these figures are its own. Contracted annual recurring revenue has reached more than 9 times its level at the end of 2025. Annual recurring revenue, or ARR, is the yearly value of contracts. The wire’s bullet line says contracted ARR tops $600 million. Live ARR in production, the part the company says is already running for customers, has grown more than 4.5 times over the same period. The release does not print the 2025 dollar baseline, so the nine-times line cannot be turned into a second dollar figure from the page alone. It says its inference platform now processes approximately 4 trillion tokens a week. A token is a small piece of text, or other model input, that the system handles. Four trillion a week is the company’s count of that traffic. The release does not say an outside auditor checked the revenue or the token count. Those lines are GMI Cloud’s.
Customers the release names: Fireworks, Higgsfield, Nous Research, OpenRouter, Reflection, Cartesia, Trend Micro, and Utopai Studios. Trend Micro is also named as an investor in the round. The release lists the names. It does not say what each one spends, or which ones sit on the new capacity. Those names are GMI Cloud’s.
Alex Yeh, founder and chief executive, is quoted in the release. He said customers are scaling faster than ever and need infrastructure that keeps pace. He said AI is driving a new renaissance, and reliable compute is its foundation. He said the goal is to build that foundation across continents, with an ecosystem of products on top of it. A second quotation from Yeh says that in AI infrastructure a delivery date is a promise, because customers plan launches, hiring, and revenue around it, and that the company’s place in Taiwan’s supply chain is how it keeps that promise, cluster after cluster. Those quotations are his, in the company statement.
The release’s case for one company on both sides of the Pacific. It says demand for compute is no longer regional. U.S. AI companies and hyperscalers, the very large cloud operators, need capacity in both the United States and Asia. Enterprises across the Asia-Pacific, it says, want production AI built close to home, under local data and compliance rules. It says most AI clouds are built for one side of that, and that GMI Cloud is built for both, as a single platform so customers can place a workload where their users, data, and regulators are. A workload is the job a customer runs on the chips. The next paragraph is about supply. It says serving both markets starts with a secure supply chain. It says GMI Cloud’s ties to Taiwan, home to most of the world’s AI server manufacturing, give it close relationships with manufacturers and a more predictable path from order to deployment. It says the company brings that precision and operational discipline to customers worldwide. Those lines are GMI Cloud’s. They are the company’s account of its supply chain. They are not a count of servers delivered this week.
Chenyu Zhao, co-founder of Fireworks, is quoted in the same release. He said capacity that arrives late is capacity the company cannot use. He said GMI Cloud has been one of Fireworks’ strongest and most reliable providers across NVIDIA GB200 and GB300 NVL72 systems. Those names are NVIDIA’s rack-scale machines: a cabinet of the company’s newest AI chips, sold as one system. The quotation is his, as GMI Cloud printed it. It is a customer’s statement in the company’s release. It is not a measured test that a rack arrived on a named date.
The about box describes the company. GMI Cloud says it is an AI-native cloud infrastructure company, built as one cloud for compute, inference, and agents. An agent, here, is software that takes a next step on a task, not only a chat reply. The box says the company provides GPU clusters, optimized inference, and agent infrastructure on one cloud, so enterprises and developers can build, deploy, and scale production AI. It was founded in 2021 and is headquartered in Mountain View, California. It says it operates GPU infrastructure across the United States and Asia-Pacific. The media contact is louisa.g@gmicloud.ai. The site it prints is gmicloud.ai. Those lines are GMI Cloud’s.
The release is a financing announcement. It does not print a valuation. It does not print a price for an hour of compute. It does not say how much of the $445 million facility has been drawn. NVIDIA’s role in the round, as the release states it, is participation in the equity. The revenue and token figures are the company’s.
The picture is the GMI Cloud homepage on the day of the announcement. A lime bar across the top says GMI Cloud raises $668 million in fundraising, led by ARCHIV, with participation from NVIDIA, and offers a link to read more. Under it, the black navigation bar shows the GMI logo, the words Powered by NVIDIA, and an NVIDIA Preferred Partner badge. The large line on the page is one cloud for compute, inference, and agents. Behind the type is an aerial photograph of a construction site: a building frame, cranes, and rows of green containers. The banner names the raise. It is not a photograph labeled as a finished data center, and it does not print a calendar date.
In plain terms, GMI Cloud said on Wednesday that it raised $668 million to add GPU capacity in the United States, Taiwan, and Southeast Asia. Of that, $223 million is Series B equity led by ARCHIV, with NVIDIA participating, and $445 million is a credit facility led by CTBC. The company says contracted annual revenue is above $600 million and more than nine times the end of 2025, that the live portion has grown more than 4.5 times, and that inference traffic is about 4 trillion tokens a week. Those are the company’s figures. Founder Alex Yeh says the point is to keep delivery dates on both sides of the Pacific, using Taiwan’s server supply chain. Fireworks co-founder Chenyu Zhao is quoted calling GMI Cloud a reliable provider on NVIDIA’s GB200 and GB300 NVL72 racks. The announcement does not print a valuation, and it does not turn NVIDIA’s investment into a product deal.