GMI Cloud has announced $668 million in new financing, with NVIDIA taking part in the equity round. The GPU cloud and inference provider, founded in 2021 and headquartered in Mountain View, California, published the news through Business Wire on 30 September 2026. The headline says “over $660 million”. The first line of the release is more exact: $223 million of Series B equity and a $445 million credit facility led by the Taiwanese bank CTBC.
That split is the most important fact in the announcement. About two-thirds of the money is a loan facility, not an investment in the company’s shares. For a business whose main cost is buying NVIDIA hardware, that is a normal shape, but it changes what the headline number means.
This article sets out who invested, how the money divides, what the revenue claims do and do not say, where GMI Cloud sits among the fast-growing independent GPU clouds, and what buyers of AI capacity should take from it. For background on how chipmakers invest in their own customers, see our analysis of circular financing between NVIDIA and AI companies, and for the sector’s public-market test, our report on Nscale’s IPO plans.
Table of contents
- What GMI Cloud Raised on 30 September
- How GMI Cloud Split the $668M Between Equity and Debt
- Who Backed GMI Cloud’s Series B
- GMI Cloud’s Revenue Claims, Read Closely
- Where GMI Cloud Fits Among the Neoclouds
- How GMI Cloud Got Here
- What GMI Cloud’s Round Means for Buyers of GPU Capacity
- What We Still Don’t Know About GMI Cloud’s Round
- GMI Cloud FAQ
- References and Further Reading
What GMI Cloud Raised on 30 September
The release’s sub-headings summarise the deal: the round “was led by ARCHIV, an AI and Robotics investment firm, with participation from NVIDIA”, and the money “will expand GPU capacity across the U.S., Taiwan, and Southeast Asia as contracted ARR tops $600 million”.
Two instruments, one headline
The $668 million is the sum of two very different things. The $223 million Series B buys shares. The $445 million facility is a credit line arranged by CTBC, one of Taiwan’s largest banks, which GMI Cloud can draw on and must repay. No valuation, interest rate, term or security for the facility was disclosed.
What the money pays for
GMI Cloud says the funding supports capacity expansion in the United States, Taiwan “and the rest of APAC”, building on its Taiwan AI Factory announced in 2025 and a Japan sovereign AI initiative announced earlier this year. The Series B will also “support the continued growth of GMI Cloud’s inference services and strategic hiring”.
| Item | Detail |
|---|---|
| Announced | 30 September 2026, Business Wire |
| Total | $668 million (“over $660 million” in the headline) |
| Equity | $223 million Series B, led by ARCHIV, NVIDIA participating |
| Debt | $445 million credit facility led by CTBC |
| Valuation | Not disclosed |
| Revenue claim | Contracted ARR above $600 million, more than 9x end-2025 |
| Usage claim | About 4 trillion tokens processed per week |
| Uses | GPU capacity in the US, Taiwan and APAC; inference services; hiring |
How GMI Cloud Split the $668M Between Equity and Debt
By our arithmetic, $445 million is 67% of $668 million, so two-thirds of the round is debt and one-third is equity.
Why a GPU cloud borrows
A GPU cloud works much like an equipment leasing business. It buys expensive hardware, installs it and rents it out under contracts. Banks are comfortable lending against that kind of asset when there is a signed customer behind it, and borrowing is usually cheaper than selling shares. Equity pays for people, software and sales; the loan pays for racks. The business analysis site FourWeekMBA made the same point in its write-up, calling the split “the expected shape” for the model.
The risk that comes with it
Debt has a repayment schedule that does not move if demand slows. Equity investors can wait out a weak quarter; a lender expects to be paid. The more a GPU cloud relies on loans, the more its health depends on keeping its chips rented. In November 2025 GMI Cloud’s founder told Reuters its GPU utilisation was “almost full”, which is the condition that makes this structure work.
Even more debt last time
This is not new for the company. When GMI Cloud raised $82 million in its Series A in October 2024, the money was $15 million of equity and $67 million of debt, according to Data Center Dynamics, citing TechCrunch. That is 82% debt, by our arithmetic. So the new round is actually less reliant on borrowing than the last one.
Share of each GMI Cloud round that is debt (our arithmetic from disclosed splits)
Adding the two rounds
Across both disclosed rounds, GMI Cloud has raised $238 million of equity and $512 million of debt, a total of $750 million, by our sum. Data Center Dynamics also reported that the Series A brought total capital to “more than $93 million”, which implies roughly $11 million raised before it.
GMI Cloud’s two disclosed rounds by instrument, US$ millions
Who Backed GMI Cloud's Series B
The investor list mixes a brand-new lead, the chipmaker whose hardware GMI Cloud rents out, and a cluster of Korean and Taiwanese institutions.
ARCHIV, a new lead investor
ARCHIV is described in the release as “a new investment firm based in San Francisco that specializes in AI and robotics”. Little else about it is public yet. Leading a $223 million round as a new firm is unusual, and the release does not say how much ARCHIV itself contributed.
NVIDIA on both sides
NVIDIA participated in the equity. It is also the supplier of every GPU GMI Cloud rents out, and in July 2026 the two announced a “strategic collaboration” alongside GMI Cloud’s $500 million capital spending commitment. WOWTALE reports that GMI Cloud is one of a handful of NVIDIA Reference Cloud Partners worldwide. A chipmaker investing in a customer that then buys its chips is common in this market, and it is disclosed here, but it means NVIDIA benefits whether it is paid as a supplier or as a shareholder.
Trend Micro as investor and customer
Trend Micro, the cybersecurity company, appears on both the investor list and the customer list. WOWTALE adds that GMI Cloud, Trend Micro and the Taiwanese electronics maker Wistron run a joint venture called Magna AI to build sovereign AI factories on NVIDIA’s Vera Rubin NVL72 platform.
A Korean block
Four Korean institutions joined: DSC Investment, KB Investment, the insurer Kyobo Life and the telecoms group KT Corporation. WOWTALE calls that “unusual for a Taiwanese startup raise”. KT has been building its own AI data centres and has a large AI and cloud partnership with Microsoft; DSC Investment previously backed the Korean AI chip company FuriosaAI. WOWTALE’s list also includes Redwood, which the release does not name.
| Investor | Role | Base |
|---|---|---|
| ARCHIV | Series B lead | San Francisco |
| NVIDIA | Participant; also GPU supplier | US |
| CTBC | Lead lender on the $445M facility | Taiwan |
| Trend Micro | Participant; also a customer | Japan and Taiwan |
| DSC Investment, KB Investment | Participants | South Korea |
| Kyobo Life, KT Corporation | Participants | South Korea |
| Redwood | Participant per WOWTALE; not in the release | Not stated |
GMI Cloud's Revenue Claims, Read Closely
The release gives two growth figures, and they are worth separating.
Contracted versus live
“GMI Cloud’s contracted ARR has reached more than 9x its year-end 2025 level,” the release says, and the sub-heading puts contracted ARR above $600 million. Separately, “Live ARR in production today has also grown more than 4.5x over the same period.” Contracted annual recurring revenue counts deals that are signed; live ARR counts revenue actually being billed from running systems. Only the contracted figure has a dollar amount.
What the gap means
Dividing $600 million by nine gives about $67 million, a rough guide to contracted ARR at the end of 2025; the company did not publish that base. Live revenue growing at half the multiple of contracted revenue is normal when customers sign before racks are installed. It is also exactly the gap a lender cares about, because the loan is repaid from live revenue, not from signatures.
Growth since the end of 2025, as a multiple (company figures, both stated as minimums)
Customers named
The release lists Fireworks, Higgsfield, Nous Research, OpenRouter, Reflection, Cartesia, Trend Micro and Utopai Studios, and says the inference platform “now processes approximately 4 trillion tokens per week”. Fireworks co-founder Chenyu Zhao called GMI Cloud “one of our strongest and most reliable providers across NVIDIA GB200 and GB300 NVL72 systems”. We look at what that volume means in our separate article on the 4 trillion tokens a week.
Where GMI Cloud Fits Among the Neoclouds
“Neoclouds” is the industry’s name for independent GPU clouds that compete with Amazon, Microsoft and Google for AI workloads. The past month has been busy for them.
The recent raises
WOWTALE’s round-up lists Nscale raising $3.36 billion in convertible notes, FluidStack $1.5 billion at an $18 billion valuation and Finland’s Verda $189 million. Crusoe recently raised a $3.9 billion Series F at a $30.9 billion valuation, which we covered in our report on Crusoe’s $3.9 billion round. WOWTALE names Permir, an Asia-Pacific neocloud that raised $505 million from Coatue, NVIDIA and others at a $5.5 billion valuation, as GMI Cloud’s closest regional peer.
Recent neocloud raises, US$ millions (instruments differ; figures as reported by WOWTALE)
Different weight classes
On those numbers, GMI Cloud sits in the middle tier, well below Crusoe and Nscale but above most regional players. At the very top, WOWTALE notes, CoreWeave took a $2 billion investment from NVIDIA toward a 5-gigawatt “AI factory” target by 2030. Others chose a different chip supplier entirely, such as TensorWave, which runs on AMD.
The Taiwan supply chain pitch
GMI Cloud’s argument for itself is delivery rather than price. “In AI infrastructure, a delivery date is a promise,” chief executive Alex Yeh said in the release. “Our place in Taiwan’s supply chain is how we keep that promise, cluster after cluster globally.” WOWTALE quoted him saying that “more than a quarter of expected data center capacity missed its completion date” last year, and that “every cluster we’ve committed has come online on schedule”.
How GMI Cloud Got Here
The company’s history explains both its Taiwan base and its appetite for debt.
From crypto mining to GPUs
According to WOWTALE, Yeh was running a Bitcoin mining operation in Taiwan in 2021 when China’s crypto-mining ban forced a change of direction, and he turned it into a GPU cloud business in 30 days. He previously managed a $200 million portfolio at CDIB’s China Life private equity fund and was a partner at Infinity Ventures Crypto.
The Series A partners
The October 2024 round was led by Headline Asia, with the Thai energy company Banpu and the Taiwanese electronics maker Wistron taking part. According to Data Center Dynamics, citing TechCrunch, Banpu agreed to supply power to the company’s operations and Wistron planned to co-develop products with it. Those are strategic partners rather than purely financial ones: one helps with electricity, the other with servers, the two things a GPU cloud cannot run without.
The Taiwan AI Factory
In November 2025, Reuters reported that GMI Cloud would build a $500 million AI data centre in Taiwan with about 7,000 NVIDIA Blackwell GB300 GPUs in 96 racks, drawing about 16 megawatts and due online by March 2026. At the time the company said it was raising $400 million from several Taiwanese banks for the project, planned a 50-megawatt data centre in the United States, and was considering a stock market listing in two to three years. The release does not say whether the new CTBC facility is linked to that earlier bank financing.
| Date | GMI Cloud milestone |
|---|---|
| 2021 | Founded; pivot from crypto mining to GPU cloud |
| October 2024 | $82 million Series A: $15M equity, $67M debt |
| November 2025 | $500 million Taiwan AI factory announced |
| Earlier in 2026 | Japan sovereign AI initiative announced |
| 22 July 2026 | $500 million capital spending commitment; NVIDIA collaboration |
| 30 September 2026 | $668 million: $223M Series B plus $445M credit |
What GMI Cloud's Round Means for Buyers of GPU Capacity
Few businesses will sign a contract with GMI Cloud directly. Many will use a service that runs on it, or on a rival like it. The round has three practical lessons for buyers of AI capacity and cloud infrastructure.
Ask who owns the hardware
Inference platforms and model routers often rent GPUs from neoclouds. GMI Cloud’s customer list includes Fireworks and OpenRouter, which many developers use without knowing where the chips sit. For data residency and outage planning, ask your provider which infrastructure partners it uses and in which countries.
Check the provider’s funding shape
A provider funded mostly by loans needs its capacity full. That can mean firm minimum commitments in your contract and less flexibility to scale down. Read the commitment and exit terms as carefully as the price.
Questions to put to any GPU provider
The table below turns those points into questions a buyer can ask before signing, whichever provider is involved.
| Question | Why it matters |
|---|---|
| Whose data centres will run our workload, and in which countries? | Data residency and outage exposure |
| Is capacity live today or contracted for a future date? | Signed capacity can arrive late |
| What minimum commitment and term does the contract require? | Debt-funded providers need full racks |
| Which GPU generation and rack type is included? | GB300 NVL72 and older systems perform very differently |
| What happens to our data and workloads if the provider is sold or fails? | Exit and portability planning |
Asking these questions does not require any view on whether a particular provider is sound. It simply makes the risks visible before they become your problem.
Regional capacity is growing
GMI Cloud pitches capacity near Asia-Pacific customers “under local data and compliance requirements”. Firms with users in Taiwan, Japan or Korea now have more regional options for AI workloads and wider cloud computing needs, rather than relying only on US hyperscalers.
What We Still Don't Know About GMI Cloud's Round
The release is clearer than many funding announcements, but important details are missing.
Price and terms
There is no valuation, and nothing on the credit facility’s interest rate, term, security or covenants. Those terms decide how much of each dollar of revenue goes to the bank.
Live revenue in dollars
Live ARR is given only as a multiple. Without a dollar figure, outsiders cannot tell how much of the $600 million in contracts is already paying.
Hardware and utilisation
The release gives no GPU count, no current utilisation rate and no gross margin, the three numbers that would show whether the loan is comfortably covered.
GMI Cloud FAQ
How much did GMI Cloud raise?
$668 million, made up of $223 million of Series B equity and a $445 million credit facility led by CTBC. The release was published on 30 September 2026.
Who led GMI Cloud’s Series B?
ARCHIV, a new San Francisco investment firm focused on AI and robotics, led the equity round. NVIDIA participated, along with DSC Investment, Trend Micro, KB Investment, Kyobo Life, KT Corporation and others.
Is NVIDIA an investor in GMI Cloud?
Yes. NVIDIA participated in the Series B and is also the supplier of the GPUs GMI Cloud rents out.
What does GMI Cloud do?
It rents out NVIDIA GPU clusters and runs AI models as an inference service, with data centres in the United States and Asia-Pacific. It was founded in 2021 and is headquartered in Mountain View, California.
What is GMI Cloud’s revenue?
The company says contracted annual recurring revenue is above $600 million, more than nine times its end-2025 level. It has not published live revenue in dollars.
References and Further Reading
Business Wire: GMI Cloud Raises Over $660 Million to Accelerate Global AI Infrastructure Expansion
WOWTALE: Taiwan’s GMI Cloud Raises $668M, With Korean Capital Piling In
FourWeekMBA: GMI Cloud’s $668M Round Is Two-Thirds Debt
Data Center Dynamics: GMI Cloud secures $82m in Series A funding for AI cloud
GMI Cloud: GMI Cloud Commits $500 Million to Expand AI Infrastructure
