NVIDIA (NVDA.US) is transforming from a pure hardware supplier into a "lender of last resort" for AI infrastructure, as the capital required to build next-generation artificial intelligence systems surpasses the limits of traditional Wall Street financing channels. The chip giant, with a market capitalization of approximately $5.4 trillion, is now using its own balance sheet to back customers' GPU purchases, a move unprecedented in the semiconductor industry. In a recent episode of the "All-In" podcast, renowned fund manager and founder of Atreides Management, Gavin Baker, asserted, "NVIDIA is becoming the central bank of the AI economy, or the Fed of AI."
This shift stems from an August 10 announcement where NVIDIA revealed strategic partnerships with six top Wall Street financial institutions: Apollo Global Management (APO.US), BlackRock (BLK.US), Blackstone (BX.US), Brookfield Asset Management (BAM.US), Goldman Sachs (GS.US), and KKR (KKR.US). Together, they will establish an independent computing power financing platform, aiming to unlock over $500 billion in third-party capital for AI infrastructure investment over the long term. Goldman Sachs plays a pivotal "bridging" role in this plan, currently engaging with potential investors including U.S. insurance companies, asset managers, and banks. This is not a simple financing collaboration but an experiment to create a new financial market centered around AI computing power. NVIDIA CEO Jensen Huang positions this initiative as a key step toward making AI computing power a standardized, sustainable investment asset class.
Core Mechanism: Residual Value Guarantees Plus Asset Securitization
The core logic of this financing model is to allow AI data center operators to obtain GPUs without relying entirely on their own cash, debt issuance, or equity raises. Instead, they can secure loans from financial institutions to complete construction. Previously, hyperscale cloud providers and emerging cloud service providers had to prepay for GPU costs. Now, they can borrow to purchase the systems and repay the loans using future revenue from inference and leasing. NVIDIA's role is to match computing power buyers with Wall Street capital and provide "residual value guarantees"—promising that these GPUs will still be leaseable and generate cash flow value after three to four years. NVIDIA offers up to a 25% residual value risk backstop in some projects. If a borrower defaults and the actual price of the GPUs liquidated by financial institutions is lower than expected, NVIDIA covers part of the value gap. Huang stated on social media platform X that the company can choose to assume up to 25% residual value support per agreement, equivalent to a maximum of $125 billion. Baker compared this model to mortgage securitization: NVIDIA's standardization of GPU system designs and lease contracts makes future cash flows potentially packageable into securities, creating a new investable and tradable asset class. Goldman Sachs Chairman David Solomon called it "a credit market backed by NVIDIA computing power."
However, controversy has not subsided. "The Big Short" investor Michael Burry recently issued another warning. On social media platform X, he shared a chart showing that between Microsoft (MSFT.US), Oracle (ORCL.US), Amazon (AMZN.US), Google (GOOGL.US), Meta (META.US), OpenAI, Anthropic, xAI (SpaceXAI), CoreWeave (CRWV.US), NVIDIA, and AMD (AMD.US), there are approximately $46 billion in direct equity investments and $879 billion in multi-year purchase commitments circulating among each other. Burry believes the same capital could be double-counted as revenue across multiple nodes in the chain, amplifying overly optimistic assessments of AI commercialization. The Bank for International Settlements also warned in its June annual report that debt growth among hyperscale cloud providers related to AI construction is outpacing the balance sheets supporting that debt. Huang has publicly dismissed the term "circular financing" as "absurd." Facing the latest criticism, he again sought to reassure the market, stating that NVIDIA will prudently evaluate each project. NVIDIA officials emphasize that project investment and financing decisions are entirely subject to independent review by partner financial institutions, with no enterprise-led credit allocation circular financing mechanism. Baker also clarified that this is not circular financing. He characterizes it as asset-backed lending based on real expected cash flows, not a method for suppliers to artificially inflate their customers' purchasing power.
Biggest Risk: "Dark GPU" Crisis
While NVIDIA attempts to redefine GPUs as financeable infrastructure assets, the biggest threat to this grand narrative is not insufficient demand, but an oversupply of computing power. David Sacks, a member of the President's Council of Advisors on Science and Technology, warned on the "All-In" podcast that a "dark GPU" scenario—where a large number of GPUs are idle and prices collapse—could cause systemic shocks to the entire AI infrastructure investment chain. Sacks drew an analogy to the "dark fiber" of the internet bubble era: in the early 2000s, telecom companies massively laid fiber optic cables, but demand fell far short of expectations, leaving much of the fiber unused. He said that if data centers are built based on spot price assumptions of $30 to $50 per watt (an estimate Musk had publicly made for AI computing power value), and those prices cannot be sustained, it will be a disaster for everyone. Yet, Sacks also offered a counterintuitive perspective: the current political resistance to data center construction might actually protect the market from overcapacity. The higher the barriers to building, the harder it is for supply to expand rapidly, lowering the risk of overcapacity.
Supply Chain and Industry Chain Reactions
Beyond NVIDIA itself, the impact of this financing architecture extends far. Analysts point out that, firstly, once institutions like Goldman Sachs, KKR, and Blackstone begin to price GPUs as assets akin to "aircraft financing," the market valuation logic for computing power facilities will shift from tech growth stocks to infrastructure assets. Secondly, competitors like AMD, lacking a comparable residual value guarantee system, will find financing terms a decisive variable when competing for the same large customers. Thirdly, the pace of data center construction will no longer be constrained by the capital expenditure budgets of hyperscale cloud providers. The order visibility of server manufacturers will become a leading indicator of whether this financing framework is truly taking hold. However, once capital is no longer a bottleneck, physical constraints like grid connection queues, transformer and turbine equipment delivery times, and land and permitting issues will emerge. Finally, Baker specifically noted that Anthropic's forthcoming S-1 filing will be the most important data point for the industry—he believes macro and value investors who underestimate the basic economics of AI will be proven wrong. Anthropic confidentially submitted a draft S-1 to the U.S. SEC on June 1, with the market expecting a Nasdaq listing as early as October. This prospectus will, for the first time, reveal the true economic picture of AI computing power to the public—whether it is a sustainable cash-flow business or a mirage built on stacked capital.
Comments