Nvidia backs a $635M GPU-backed AI loan — first of its kind in Asia

A $635 million GPU-backed AI loan sought by GMI Cloud — with Nvidia’s backing — signals something larger than one company’s balance sheet ambitions. It marks a structural shift in how AI infrastructure gets financed, and it arrives on the same day that United Microelectronics Corporation quietly crossed a manufacturing threshold that could reshape semiconductor supply chains for years.

Key takeaways

• GMI Cloud is seeking a $635 million multi-tranche loan backed by GPU assets and customer contracts, with Nvidia’s support — one of the first such financings of its kind in Asia.
• The company operates over 30,000 Nvidia GPUs — including H100, H200, and Blackwell chips — in US data centers, focused on AI inference workloads.
• Financing terms, lender identities, and a closing timeline have not been publicly disclosed.
• UMC has begun mass production of 12-inch silicon photonics wafers at its Singapore fab, supporting SILITH Technology’s 1.6T platform.
• UMC’s Singapore manufacturing base reduces geopolitical risk tied to Taiwan’s semiconductor supply chain.

GMI Cloud Pursues $635 Million GPU-Backed Loan with Nvidia Support

GMI Cloud, a US-based data center operator backed by Taiwan’s GMI Technology Inc. and Realtek Semiconductor Corp., is pursuing a NT$20.45 billion ($635 million) multi-tranche loan deal supported by customer contracts for graphics processing units. According to people familiar with the matter, the deal includes a five-year term loan tranche of NT$13.9 billion launched in Taiwan’s syndication market — making it one of the first GPU-backed financings of its scale in Asia.

The Nvidia connection matters here. GMI Cloud was named one of only six global Reference Platform Cloud Partners by Nvidia — a designation that certifies the provider meets Nvidia’s standards for running its hardware at scale. The company was also identified as an early contributor to Nvidia DGX Cloud Lepton, Nvidia’s managed cloud AI service. That relationship gives the loan an implied credibility that few pure-play cloud providers could match.

Scale and Focus of GMI Cloud’s AI Operations

GMI Cloud operates more than 30,000 GPUs housed in US data centers, running Nvidia’s H100, H200, and Blackwell chips. The company focuses specifically on AI inference workloads — the production side of AI where deployed models serve real-time predictions to users, as opposed to the training phase that dominates most headlines.

That distinction matters competitively. Inference is widely considered the faster-growing segment of AI compute demand, driven by the explosion in deployed AI applications. Positioning squarely in that segment, with certified Nvidia hardware at scale, gives GMI Cloud a credible claim on enterprise customers who need reliable, high-performance inference capacity without building their own data centers.

Undisclosed Loan Terms and Market Impact

Neither GMI Cloud nor Nvidia has publicly disclosed the financing terms, the identity of potential lenders, or a closing timeline. That silence is itself informative. The interest rate and loan-to-value ratio — once revealed — will function as a market signal, showing how lenders are pricing GPU collateral risk across the AI infrastructure sector.

If the $635 million financing closes, the capital would presumably fund additional GPU procurement and data center expansion, allowing GMI Cloud to compete more aggressively against CoreWeave, Lambda, and the major hyperscalers. The stakes go beyond GMI Cloud’s own growth story: every AI infrastructure company weighing debt financing over equity dilution will be watching closely to see what terms this deal ultimately lands on.

Emerging Trends in GPU-Backed Lending for AI Infrastructure

GPU-backed lending is gaining traction as a distinct asset class precisely because enterprise-grade chips carry concrete, defensible value. Nvidia’s H100 and Blackwell chips carry list prices that run into the tens of thousands of dollars per unit. When 30,000 of them are deployed in revenue-generating data centers under customer contracts, they form an asset base that lenders can actually underwrite.

Valuing Enterprise-Grade GPU Assets for Lending

The key underwriting variables for GPU collateral are utilization rates, the duration and quality of customer contracts, and the expected useful life of the hardware before the next generation erodes its competitive value. This last factor — hardware obsolescence risk — is the genuine unknown that lenders must price into any GPU-backed loan, since Nvidia’s product cadence has been accelerating.

What makes GMI Cloud’s deal structurally interesting is that the loan is backed not just by the chips themselves but by the customer contracts generating revenue from those chips. That layered collateral structure — hardware plus contracted cash flows — reduces lender exposure compared to a simple asset-backed loan and could serve as a template for the broader AI infrastructure financing market.

UMC Advances Silicon Photonics Production in Singapore

On July 14, United Microelectronics Corporation delivered its first mass-produced silicon photonics wafers from its 12-inch Fab 12i facility in Singapore — a milestone that moves the technology from R&D ambition into commercial supply chain reality.

Mass Production of 12-Inch Silicon Photonics Wafers

The jump from 8-inch to 12-inch wafer production is meaningful in manufacturing economics. More chips per production run translates directly into lower per-unit costs and the kind of scale that makes large customers comfortable placing substantial orders. The wafers coming out of UMC’s Singapore fab support SILITH Technology’s 1.6T silicon photonics platform — designed for the high-speed optical interconnects that hyperscale data centers need to move massive volumes of data between AI servers.

The UMC-SILITH partnership took roughly 18 months to move from initial development to full-scale manufacturing, a timeline that reflects both the technical complexity involved and the urgency of getting commercial-grade silicon photonics into the AI data center supply chain.

Partnerships with SILITH Technology and imec

UMC laid critical groundwork in December 2025 by licensing imec’s iSiPP300 silicon photonics process platform. Rather than building from scratch, UMC accessed a proven technology stack — a strategic shortcut that accelerated the path to manufacturing scale and cost efficiency. Risk production using the advanced imec-derived process is targeted for the 2026-2027 timeframe.

Citi analysts have reportedly viewed UMC’s silicon photonics push as a positive signal for the company’s outlook, suggesting the move could help differentiate UMC from competitors fighting over increasingly commoditized legacy chip manufacturing.

Geopolitical Advantages of Singapore Manufacturing Base

UMC has historically played second fiddle to TSMC in the foundry business, focusing on mature and specialty process nodes. Silicon photonics represents a deliberate bet on a segment where UMC can lead rather than chase. The Singapore manufacturing location adds a geopolitical dimension that amplifies the strategic logic: with ongoing tensions around Taiwan’s semiconductor supply chain, production capacity in Singapore reduces geographic concentration risk in a way that customers and investors increasingly value.

That said, the first-mover advantage won’t hold indefinitely. GlobalFoundries has its own silicon photonics program, and TSMC has signaled interest in the space. The question for UMC is how quickly it can deepen customer relationships and lock in volume commitments before larger competitors decide to invest aggressively in 12-inch photonics capacity of their own.

FAQ

What is the scale of GMI Cloud’s GPU infrastructure?

GMI Cloud operates over 30,000 GPUs, including Nvidia’s H100, H200, and Blackwell chips, primarily in US data centers.

What type of AI workloads does GMI Cloud focus on?

GMI Cloud specializes in AI inference workloads, which involve serving AI model predictions to users in real time — the production side of AI deployment rather than model training.

What are the known details about GMI Cloud’s $635 million loan?

GMI Cloud has not publicly disclosed the financing terms, lender identities, or closing timeline for the $635 million GPU-backed loan. The deal includes a five-year term loan tranche of NT$13.9 billion launched in Taiwan’s syndication market.

How does UMC’s silicon photonics production impact the semiconductor supply chain?

UMC’s mass production of 12-inch silicon photonics wafers in Singapore supports advanced AI data center interconnects through SILITH Technology’s 1.6T platform, while the Singapore location reduces geopolitical risks tied to Taiwan’s semiconductor supply chain tensions.

Article produced with the assistance of artificial intelligence and reviewed by the editorial team.