The Era of Compute Securitization Has Arrived: Will AI Repeat the 2008 Subprime Crisis?

💡Nvidia's US$500 billion financing initiative with six Wall Street financial institutions signals the start of computing power securitization by mobilizing third-party capital for AI infrastructure and notably, Nvidia will provide a residual value guarantee of up to 25% for certain financing projects. This guarantee reduces financial institutions' concerns over rapid GPU technology cycles and depreciation, thereby unlocking greater inflows of financial capital, effectively establishing a final credit backstop for GPU assets and marking the official beginning of the 'era of computing power securitization' alongside CME's launch of computing power futures.

💡CME's new computing power futures contracts, planned to launch on October 5 and track the rental costs of Nvidia's H100 and B200 GPUs using Silicon Data's hourly rental prices, play a crucial role in the financialization of AI infrastructure by providing a tool for managing fluctuations in GPU rental costs for operators. For financial institutions, these futures enable more transparent assessments of the returns and risks associated with computing power assets, ultimately increasing their willingness to allocate capital to such assets and supporting the securitization trend.

💡Nvidia and CME are prioritizing the securitization of computing power assets due to two main factors: "scale" and "speed." The capital expenditure required for AI infrastructure is rapidly expanding, exceeding what technology giants and cloud providers can support through their operating cash flows, with spending expected to exceed US$1 trillion by 2027. "Speed" is crucial as AI infrastructure aims to deploy nearly US$3 trillion in investment within five years, a pace significantly faster than historical infrastructure cycles, necessitating securitization to monetize future long-term rental income in advance and shorten capital recovery periods.

💡Nvidia's 'AI Federal Reserve' role involves providing a residual value guarantee of up to 25% for certain financing projects that fund AI infrastructure. This guarantee is crucial because it reduces financial institutions' concerns over the rapid GPU technology cycles, equipment depreciation, and future resale prices. By absorbing part of the value shortfall when actual residual values fall below expected levels, Nvidia establishes a final credit backstop for GPU assets, thereby unlocking greater inflows of financial capital and acting as a guarantor of asset value rather than directly providing liquidity.

💡For computing power to become a capitalizable commodity, financial institutions must address four fundamental questions: the economic life of the asset, its estimated future residual value, its standardization and transferability between different holders, and its compliance with existing regulatory requirements. Historically, significant uncertainty surrounding these aspects made GPUs and AI data centers difficult to use as collateral for large-scale financing. However, collaborations involving Nvidia, CSPs, AI supply chain participants, and financial institutions are gradually clarifying these barriers, making securitization more viable.

💡Nvidia's AI Factory Reference Architecture addresses the standardization challenge in data center financialization by integrating computing, power, cooling, networking, and data center infrastructure into a replicable architecture. This initiative aims to reduce the significant variation across data center facilities, which traditionally made consistent value assessment and ownership transfer difficult for financial institutions. By evolving data centers from standalone customized projects into modular, replicable, and verifiable AI factories, Nvidia's architecture makes these assets more predictable and easier for financial institutions to evaluate and securitize.

💡Regulatory exemptions make data center asset-backed bonds more appealing for AI infrastructure financing because, according to a recent letter from the US Securities and Exchange Commission (SEC), these bonds do not fall under the traditional definition of "asset-backed securities (ABS)." This exemption frees them from the stringent securitization regulations implemented after the 2008 financial crisis, reducing both bond issuance costs and complexity. Consequently, this regulatory clarity allows more fixed-income capital to participate in AI infrastructure financing, lowering barriers for institutional investment.

💡The 'Prisoner's Dilemma' that technology giants face in the AI financing race is a "move forward or fall behind" scenario. If they believe the AI industry will expand, they collectively increase spending to avoid losing market share to competitors. Conversely, if growth falls short, a company continuing to expand risks market punishment. This dynamic leads to a collective strategy that can instantly shift from aggressive computing power acquisition to collectively cutting spending and demonstrating capital discipline, indicating little room for a gradual transition between expansion and contraction.

💡Current bank reserves and money market fund assets differ significantly from those before the 2008 financial crisis; the Federal Reserve has shifted from a scarce-reserves regime to an ample-reserves regime, with bank reserves now approximately US$3 trillion, compared to extremely low levels prior to 2008. Additionally, US money market funds currently hold around US$8 trillion in assets, indicating a vast abundance of market liquidity. This ample liquidity suggests that the primary constraint on bank financing today is not a lack of lendable funds, but rather banks' willingness to assume additional credit risk.

💡Today's technology giants possess enormous core businesses like cloud services and digital advertising, generating stable operating cash flows, making their financial profiles fundamentally stronger than subprime borrowers in the 2008 crisis. While credit default swap (CDS) spreads for technology giants have surged, the largest increases are concentrated among newer GPU-centric cloud providers (Neoclouds), not the four major CSP giants or Nvidia. Their implied five-year cumulative default probabilities remain manageable, ranging from 3.5% to 6.6%, indicating well-capitalized entities with greater resilience compared to credit-impaired individuals.