Nvidia just helped a 5-year-old startup raise $668 million

By the numbers: GMI Cloud raised an $82 million Series A round in October 2024, a substantial sum on its own for a company that had spent its early years focused on Bitcoin mining infrastructure rather than AI. Going from that round to a $668 million raise roughly two years later represents more than an eightfold jump in a very short window.

Why it matters: that trajectory reflects just how fast capital has been flowing into GPU cloud infrastructure as AI demand has scaled. A company doesn't typically multiply its fundraising by that much in two years unless investors, and in this case its own chip supplier, see clear evidence of real, sustained customer demand behind it.

Why it matters: the funding is earmarked for a $500 million AI data center in Taoyuan, Taiwan, expected to be operational by March 2026. Once complete, the facility will house roughly 7,000 Nvidia GB300 GPUs spread across 96 server racks.

• Equity raised: $223 million, led by ARCHIV, with Nvidia participating.
• Credit raised: $445 million, led by Taiwan's CTBC Bank.
• Data center cost: $500 million.
• Expected GPU capacity: about 7,000 Nvidia GB300 chips across 96 racks.
• Target completion: March 2026.

The catch: this arrangement sounds circular because, in a real sense, it is. Nvidia sells GPUs to companies like GMI Cloud, those companies rent that computing power out to AI developers, and a portion of that rental revenue flows back toward Nvidia, sometimes directly through investment returns on stakes like this one.

Background: this isn't a one-off. Nvidia has made similar moves across what the industry calls "neoclouds," smaller companies built specifically around renting out GPU computing power, rather than offering the full range of services a traditional cloud giant like AWS or Azure provides. Nvidia's $2 billion investment in CoreWeave, another prominent neocloud, follows the same basic pattern.

In real life imagine a car manufacturer investing directly in a rental car company that exclusively rents out that manufacturer's own cars. The manufacturer sells more cars either way, but now it also profits from every rental, and it gets a say in how fast that rental company grows.

Who's affected: part of Nvidia's reasoning appears to be about balance of power. Relying too heavily on a small number of giant hyperscaler customers, Microsoft, Amazon, Google, gives those companies significant leverage, including the ability to design their own competing chips over time.

Funding smaller, independent GPU cloud providers gives Nvidia more customers with a direct stake in Nvidia's own chips succeeding, spreading out that dependency while also creating competition for the hyperscalers themselves in the broader cloud computing market.

What's next: the $445 million credit piece isn't a standard unsecured loan. It's backed specifically by GMI Cloud's existing GPU computing-capacity leasing contracts with its own customers, essentially using already-signed future rental revenue as collateral.

That structure matters because it reflects real, contracted demand rather than pure speculation about future growth. Banks extending this kind of financing are betting on revenue GMI Cloud has already locked in through customer agreements, not just projected market growth.

This kind of GPU-backed lending has become increasingly common across the neocloud sector generally, as banks and other lenders look for ways to finance expensive hardware purchases without taking on the full risk of unproven, speculative future demand.

Background: the company's origin story is a genuine pivot, not a straight line into AI. Founder Alex Yeh, previously a partner at a Taiwan-based crypto investment fund, was running a large-scale Bitcoin mining operation in Inner Mongolia when China abruptly banned Bitcoin mining entirely in 2021.

Yeh responded quickly by relocating, building new data center infrastructure in Arkansas and Texas originally meant purely to service Bitcoin mining computing nodes at scale. GMI Cloud officially launched in 2022 still oriented around that crypto-mining infrastructure, before pivoting toward AI as demand for GPU computing exploded.

Why it matters: that history explains something important about GMI Cloud's underlying capability. Bitcoin mining and AI computing share a surprising amount of technical overlap, both require large-scale, power-hungry hardware operating continuously, and companies with existing data center infrastructure built for one were often well positioned to pivot toward the other once demand shifted.

The company's current customers reflect how far that pivot has gone. Reported clients include AI-native companies like Higgsfield, HeyGen, and Eigen AI, businesses whose products depend directly on fast, reliable GPU access for real-time AI inference.

Taiwan brings some genuine practical advantages for this kind of facility beyond just mere corporate strategy alone. The island has deep, decades-long semiconductor industry infrastructure and technical expertise, much of it tied to chip manufacturing giant TSMC, making it a natural location for large-scale AI computing buildouts that require both technical talent and reliable power infrastructure.

Who's affected: GMI Cloud's stated push to expand across Asia positions this Taiwan facility as a regional hub, giving AI companies operating in Asian markets a local alternative to routing all their GPU computing needs through US-based data centers.

• This is infrastructure funding, not a product you'll use directly. It shapes where and how AI computing capacity gets built, not what shows up in your apps this week.
• It's a useful example of how concentrated the AI hardware supply chain has become, with the same company, Nvidia, appearing as supplier, customer relationship, and now investor across multiple parts of the industry.
• Watch for more of these deals. Nvidia's pattern of backing neocloud providers, not just hyperscalers, suggests this kind of financing arrangement will keep showing up as AI infrastructure demand keeps growing.
• Regional GPU capacity, like this Taiwan facility, matters for AI companies operating outside the US, potentially reducing latency and dependency on American data centers.