Larry Fink, the man who helped turn the home mortgage into a global financial product, has a new obsession. He recently compared the latest Nvidia-led consortium to the birth of mortgage-backed securities in the 1970s. It is a comparison that should make any investor sit up straight, not because it promises a golden age of stability, but because it signals a fundamental shift in how we value the physical guts of the internet.
This is Part Seven of the Compute Landlord Thesis. If the previous chapters tracked the rise of equity plays and vendor-financed exposure, this is the moment the landlord moves from collecting rent to securitizing the building itself. Nvidia, having spent years convincing the world that its chips are the new oil, is now convincing Wall Street that its chips are the new collateral.
On August 10, Nvidia announced a series of Memorandums of Understanding with a heavy-hitting roster of financial institutions: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. The goal is to mobilize over $500 billion in third-party capital for AI infrastructure. To put that number in perspective, it is 2.5 times the size of Google’s financing network and five times larger than the AI Infrastructure Partnership announced in September 2024. It is a staggering amount of money, though it is worth noting that these are currently non-binding MOUs. The actual deployment of this capital remains subject to the execution of final agreements, a detail that the market seemed to digest with a mix of awe and anxiety.
Nvidia’s shares fell roughly 2.9% on the news, erasing about $60 billion in market capitalization. This was not necessarily a rejection of the ambition, but rather a reflection of the market’s skepticism regarding the leverage involved. When you turn compute into a bankable asset class — treating it like commercial real estate or toll roads — you are essentially building a financial skyscraper on top of a rapidly evolving technological foundation. If the underlying tech shifts or one major player hits a wall, the ripple effects through the AI ecosystem could be significant.
Jensen Huang’s framing of this shift is characteristically precise: Nvidia is no longer just a hardware vendor; it is an infrastructure architect. By standardizing compute as a collateralized financial instrument, Nvidia is effectively creating a new class of productive, investable infrastructure: the AI factory. This is the institutional-grade third phase of the compute landlord thesis. We have moved past the era of simple equity investments, like the $3 billion Nvidia-Lancium power play or the $10 billion Volta Infra deal, into an era of massive, structured credit facilities.
The competitive implications are stark. This scale creates a moat that is almost impossible to cross for smaller players. It favors firms with massive, flexible, long-term capital bases — the kind of capital that Apollo’s Jim Zelter or Blackstone’s Jon Gray specialize in deploying. We have already seen the precursors to this: the $35 billion Apollo-Blackstone loan to Broadcom, and the Aligned Data Centers acquisition. These were the warm-ups. The $500 billion consortium is the main event.
The structural mechanics here are what matter most. Goldman Sachs’ David Solomon noted the opportunity to create a market for credit backed by Nvidia compute. This is the core of the thesis: compute is becoming the essential layer of infrastructure, as Brookfield’s Bruce Flatt put it. When you turn a GPU cluster into a bond, you are betting that the demand for that compute will remain as constant as the demand for electricity or highway access. It is a bet on the permanence of the AI factory.
However, the risks are as structural as the rewards. As Axios pointed out, the interconnectedness of these deals means that a failure at one node could trigger a systemic cascade. We are moving from a world where companies buy chips to a world where financial institutions finance the entire stack, from power generation to the final inference rack. If the financial engineering outpaces the actual utility of the compute, we are looking at a classic mismatch of duration and risk.
For investors and builders, the next phase is not about the headline number. It is about the execution. Watch for the transition from these non-binding MOUs to final, binding agreements. Watch how the credit terms are structured and whether the market begins to price in the specific risks of AI infrastructure as distinct from traditional data center assets. The compute landlord has arrived, and he is looking for a mortgage.