$500 Billion AI Finance: Off-Balance Sheet Debt Grows

The funding sources for AI infrastructure investments are expanding from operating cash to bonds, leases, and private equity. As the pace of building data centers and graphics processing units (GPUs) accelerates, concerns have arisen that relying solely on the debt recorded in financial statements makes it difficult to gauge the actual burden.

NVIDIA ($NVDA) announced on the 10th that it is pushing forward with a financial platform for AI compute infrastructure exceeding $500 billion (approximately 710 trillion won) in collaboration with Apollo, BlackRock, Blackstone, Brookfield, KKR, and Goldman Sachs. Jensen Huang, CEO of NVIDIA, stated in the announcement, "In AI, compute is revenue," indicating that AI compute should be viewed not just as a simple server purchase but as an infrastructure asset capable of generating long-term cash flow.

The core of this initiative is a structure where AI research labs and cloud service providers utilize large-scale compute facilities, while financial institutions create dedicated capital pools. NVIDIA provides the chip and software ecosystem. Previously, we reported on the debate surrounding NVIDIA's role in AI infrastructure finance.

AI factories refer to data center-type infrastructure that combines GPUs, networks, power, cooling, and software to provide large-scale AI computations. The structure involves securing power and land first, deploying equipment, and then recovering revenue from customer usage and long-term contracts. This results in a large initial investment and a delayed point of profitability.

PIMCO projected in a podcast released on July 16 that the scale of AI facility investments in 2026-2027 would exceed $1.6 trillion (approximately 227.2 trillion won). PIMCO indicated that a significant portion of this could be sourced from the bond market. In a separate report, it analyzed that capital expenditures related to AI are shifting from a structure reliant on operating cash flow to a debt-based investment cycle.

The burden outside the financial statements has also increased. PIMCO reported that as of May 15, the estimated facility investment for major hyperscalers in 2026 is $690 billion (approximately 980 trillion won) and is expected to rise to $870 billion (approximately 1,235 trillion won) in 2027. It pointed out that future lease commitments not yet recognized on the books amount to $822 billion (approximately 1,167 trillion won).

Leases and purchase commitments may not be immediately recorded as debt but can later translate into cash outflow pressures. The Financial Times reported that the purchase commitments of hyperscalers and lease obligations identified by Goldman Sachs are each around $1.5 trillion (approximately 2,130 trillion won). However, these figures represent different definitions, making simple comparisons in terms of risk difficult.

The market structure is also changing. Until now, large tech companies have largely covered data center investments through their own cash generation capabilities. However, as competition in AI models drives up the demand for power, land, servers, memory, and network equipment simultaneously, the financing methods are expanding to include bond issuance, long-term leasing, special purpose vehicles, and private loans.

This trend aligns with previous reports indicating that big tech's AI infrastructure investments are shifting from self-financing to a combination of bonds, private equity, and long-term leases. Companies like Google and Meta have been diversifying their infrastructure burdens not only by directly owning data centers but also by utilizing guarantees, leases, and partnerships.

The connection with South Korea is also significant. NVIDIA and SK Group announced on July 24 a collaboration plan for AI data centers and memory exceeding $500 billion (approximately 710 trillion won). NVIDIA stated that SK Telecom plans to build an NVIDIA Vera Rubin DSX AI factory with a capacity of up to 2 gigawatts and aims to operate the first AI factory by 2027.

On the same day, Naver, NVIDIA, and Brookfield revealed plans to expand the scale of the DSX AI factory at the Sejong Data Center from 55 megawatts to 200 megawatts by 2028. This indicates that the domestic memory supply chain and data center investments are moving in tandem with the global AI infrastructure finance structure.

Industry perspectives are divided. One side views the AI infrastructure, which bundles large-scale power, data centers, and GPUs, as a new asset class capable of attracting long-term capital. The other side raises concerns about circular financing and vendor financing, where suppliers enhance customers' purchasing power.

PIMCO believes that the AI infrastructure investment cycle is more disciplined and has higher procurement potential than the telecommunications network investments of the 1990s, but emphasizes that the actual speed of monetization is key. AIMA acknowledged that leverage is a core tool for hedge funds but stated that it is difficult to generalize this as systemic risk immediately.

Leverage risk has first emerged in individual investor positions. Reuters reported that the AI-focused hedge fund Situational Awareness transferred most of its publicly held stocks to Citadel after losses in tech stocks and margin call pressures. While this case alone cannot definitively categorize the entire AI infrastructure as a systemic risk, it illustrates that AI positions built on leverage can turn into selling pressure during market downturns.

The key issue in AI infrastructure finance is verifying cash flows rather than the size of demand. Data centers and GPUs are built first, and revenues are validated later. This is why it is necessary to separate recorded debts, lease commitments, private loans, and purchase commitments.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.