Nvidia Rounds Up Wall Street for a $500B AI Bet. The Fine Print Is Getting Scrutiny.

Nvidia just pulled off something Wall Street hasn't seen before: convincing six of the world's largest asset managers—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to sign onto a coordinated scheme aimed at funneling more than $500 billion into AI compute infrastructure. The chipmaker framed the Monday announcement as the moment "technology chips become an investable asset class." The market responded by shaving roughly 3% off Nvidia's share price, wiping out more than $60 billion in market cap in a single session. That gap—between the grandeur of the ambition and the skepticism of the reaction—is the story.

GPUs as Toll Roads

Jensen Huang, appearing in a rare joint CNBC interview alongside the CEOs of all six partner firms, laid out the core thesis: a GPU is no longer just a chip that depreciates on a balance sheet. It's an infrastructure asset, like commercial real estate or a power grid, that generates revenue and can be pledged as collateral. "These are revenue-generating assets now. They're productive, they're long-lived, they're fungible, they're flexible," Huang told CNBC's Becky Quick. "This is really the first time that technology chips have become an investable asset class." The logic hinges on a simple observation: AI compute demand hasn't stopped growing. GPU rental prices for Nvidia's H100 and B200 have climbed 25% to 48% over recent months, depending on the SKU, while top-tier cloud providers report utilization rates between 85% and 98% with waitlists stretching six months. Data center operators like CoreWeave have turned GPU-backed loans into investment-grade debt—its $8.5 billion delayed-draw term loan closed in March with an A3 rating from Moody's, backed by AI chips and cloud service contracts with Meta. That's the foundation of the "new asset class" pitch. Whether it's durable or cyclical is the open question.

The $500 Billion Question

The structure works like this: each of the six firms will raise its own capital pool and finance AI infrastructure projects—data centers, power, and GPU purchases—through special purpose vehicles that issue debt. Nvidia's role is limited to a residual-value backstop, capped at 25% of any individual deal. That means, in theory, the debt doesn't sit on the balance sheets of hyperscalers or AI labs. It sits with institutional investors. Critics quickly drew a parallel to 2000. The concern: if AI companies borrow heavily to buy Nvidia hardware, then pay Nvidia back via compute rental revenue that hasn't materialized yet, and Nvidia invests in those same companies (it has committed roughly $300 billion to OpenAI and up to $100 billion to Anthropic)—the whole thing starts to look like the circular financing that inflated the dot-com bubble. "Nvidia's stock dipped approximately 5% following reports of the OpenAI negotiations in July. The market is telling you something," read a widely-circulated Reddit post comparing the moment to the late-1990s telecom equipment playbook, where manufacturers lent customers money to buy their own gear. Michael Burry, of The Big Short fame, has publicly flagged the scale of circular spending. Jim Cramer, who lived through the dot-com crash, said earlier this month: "I lived through 2000, I don't want the sequel." Holger Zschaepitz, the financial commentator, put it more succinctly on social media: "Wall Street is committing $500 billion in capital to help NVIDIA's customers buy NVIDIA chips." Even the bond market is signaling fatigue. A debt issuance tied to Meta and BlackRock's own AI financing venture drew what dealers described as "light demand" when it priced—an early sign that even institutional money is getting picky. Nvidia's five-year credit default swaps have doubled in cost since May, trading around 77 basis points. That's the market pricing in the possibility that Nvidia's guarantee—however capped—carries real risk.

The Circularity Defense

Huang rejected the circular financing framing outright. "This initiative is designed to address that concern," he said. "We are bringing independent, long-term institutional capital into the AI infrastructure market. The demand is real: it comes from frontier AI labs, AI-native startups, enterprises, cloud providers, and countries building AI services. The capital providers independently underwrite each project." Morgan Stanley's analysts have largely backed this view, noting that the structure cuts off the "Nvidia lends to customer, customer buys Nvidia" loop by design. The 25% residual-value cap means independent third parties dominate the decision-making on deals. If the AI boom slows, the asset managers take the first hit, not Nvidia's balance sheet. But that's also the problem. If the AI boom slows, the asset managers do take the hit. That's the risk transfer. And the question of whether AI monetization can justify the capital that's piling into the sector is no longer hypothetical. Chinese financial media has gone further, warning that the structure could trigger a "subprime crisis" if AI returns fail to materialize. The SEC didn't help calm those nerves.

A Regulatory Loophole

The same day Nvidia made its announcement, the U.S. Securities and Exchange Commission quietly issued an internal letter exempting AI data center-related asset-backed securities from core investor protection rules introduced after the 2008 financial crisis. The rules—which require issuers to retain some debt and disclose more information—will no longer apply to ABS tied to AI data centers. The SEC's rationale, responding to a query from law firm Latham & Watkins, is that a data center is a physical asset that doesn't liquidate over time like a car loan or mortgage. Therefore, the securities backed by it shouldn't be regulated like other ABS. Critics see it differently. The exemption opens the door for AI infrastructure debt to be securitized without the same safeguards, potentially spreading risk through the financial system in ways that echo—though not yet replicate—the mortgage-backed securities crisis. Data center ABS issuance has already grown from $2.4 billion in 2020 to $15.5 billion in 2025. The new regime could accelerate that growth substantially.

What the Market Actually Believes

For all the hand-wringing, the financing platform is a real vote of confidence from four of the most sophisticated capital allocators in the world. Investing.com's analysis noted that the structure—built around equity co-investment from managers whose mandates already price illiquidity and long-duration risk—is "a considerably more conservative way to fund exactly the same physical assets" compared to unsecured corporate debt. Still, the market's reaction was telling. Nvidia's forward P/E has dropped to around 16x on FY2027 earnings estimates—its lowest level in a decade. That suggests investors are no longer pricing in unchecked AI growth. They're pricing in uncertainty. The core issue isn't whether AI compute demand exists. It does. AI startups covered by CLSA's database now generate a combined $178 billion in annual recurring revenue, up 31% sequentially. The issue is whether that demand can grow fast enough to justify the scale of capital entering the space. Here's a sobering data point: Big Tech's combined free cash flow has fallen to its lowest level in a decade—around $7 billion—after years of unprecedented capital expenditure. Morgan Stanley projects the largest cloud companies will spend $3.5 trillion on AI infrastructure between 2026 and 2028. Nvidia's own bond sale in June—$25 billion, which drew $85 billion in demand—was a test run for exactly this kind of market appetite.

The Asset That Doesn't Depreciate

Huang's argument rests on a unique feature of AI hardware: it doesn't depreciate like traditional IT equipment. A 3-year-old H100 still retains roughly 84% of its original value on the secondary market—a remarkable figure for a computer chip. CoreWeave re-leased H100s from a 2022 contract at 95% of original price when the contract expired. Nvidia's A100, launched in 2020, is still in mass commercial use globally. "Prices are insane," one hardware expert said flatly. "Cars depreciate faster than this." Goldman Sachs has run the sensitivity analysis: if chip lifespans shorten from 5 years to 3, implied annual depreciation across the sector jumps from roughly $3 trillion to nearly $4 trillion between 2026 and 2031. That's a $1 trillion swing hidden inside an accounting assumption. Insurers are taking notice too. U.S. state insurance regulators have begun reviewing whether the data center assets in insurers' portfolios—and the credit ratings supporting them—are actually justified.

The Bottom Line

There's a version of this story where the $500 billion financing platform becomes a powerful accelerator for GPU sales, creating an annuity-like revenue stream that pushes Nvidia's margins even higher. Morgan Stanley's model suggests the company could generate over $51 billion in incremental annual revenue through revenue-sharing agreements on the financed compute, pushing earnings per share up more than 10% by fiscal 2029. There's also a version where this becomes the defining cautionary tale of the 2020s—where cheap institutional capital, chasing a new asset class it didn't fully understand, created a debt spiral that ended in a bust. Huang is confidently betting on the first scenario. The market, trading at decade-low valuations with rising CDS spreads, is clearly hedging for the second. Nvidia reports its next quarterly results on Aug. 26. That report won't settle the debate—but it will give investors the clearest look yet at whether the compute demand underpinning this entire financial edifice is still accelerating, or starting to plateau. Until then, the $500 billion question remains open.