The SEC Just Lifted the 2008 Guardrail From the AI Boom

On Monday, the Securities and Exchange Commission exempted data-center asset-backed securities from the risk-retention requirements and investor-disclosure rules that were written after 2008 to prevent securitization-driven concentration from threatening the financial system. The action came in response to a query from the law firm Latham & Watkins ^[1].

There was a growing need for this relief. — Kevin Fingeret

The financial architecture of the AI industry has, over the past several months, assembled itself into a shape that would have been familiar to anyone who studied the mortgage-securitization chain of the mid-2000s — except this time the asset at the center is not a house but a graphics processing unit, and the institution at the center is not a bank but the company that makes the chips. That company is Nvidia. Over the past year, it has built something no semiconductor company has ever attempted: a vertically integrated financial stack that reaches from the silicon to the securitization. It manufactures the GPUs. It finances their purchase through a revenue-sharing program that lets startups and cloud providers acquire compute without upfront capital, in exchange for a cut of their future revenue ^[2]. It takes equity stakes in those same customers — it holds $3.6 billion of CoreWeave stock, making it simultaneously supplier, lender, and shareholder in its own client ^[3]. And yesterday, it launched the capstone: a $500 billion financing platform, built with a coalition of Wall Street asset managers including Apollo, Blackstone, Goldman Sachs, BlackRock, Brookfield, and KKR, structured through a special-purpose vehicle designed to transform GPU-backed debt into what Jensen Huang called something no one had ever said about a microchip before ^[4].

the first time that technology chips have become an investable asset class. — Jensen Huang

Mark Cuban put the same observation more bluntly.

the idea that it is circular is, it’s ridiculous. — Nvidia

The chain does not stop at securitization. In late July, Nvidia was reported to be leasing the Beacon Point data center in Texas for $19.6 billion over 15 years, with plans to sublease capacity to its neocloud partners ^[5]. A chipmaker is now a data-center landlord. The vertical integration is complete: silicon, vendor financing, equity, securitization, real estate. Every link in the chain that turned mortgages into a systemic crisis has an analogue in the AI capital stack, and the same company sits at every node. Huang has dismissed the circular-financing concerns as "ridiculous."

We’re basically holding the planet together—and it’s not untrue. — Jensen Huang

CoreWeave is the concrete proof of how circular this structure has become. The company, Nvidia's largest neocloud partner, reported 112% year-over-year revenue growth to $2.1 billion in the first quarter of 2026 — and a net loss of $740 million. It carries $25 billion in debt and spent $15 billion on capital expenditures in just two quarters ^[3][6]. The assets backing that debt are Nvidia's own chips, whose residual value is set by the same company that supplied them — a concentration no outside party independently prices. The revenue that services the debt comes, in significant part, from selling access to those same chips. And the equity holder that benefits if the structure holds is, again, Nvidia. The warnings have been accumulating for months, and what distinguishes them now is not their volume but their source. In late June, the Bank for International Settlements — the central bank of central banks, an institution that does not traffic in hyperbole — compared the AI investment boom to the dot-com bubble and the 1830s canal mania, warning that such episodes "ended with an eventual reversal in investment, inducing economy-wide recessions" and that the debt-heavy nonbank structures now financing AI could produce cascading defaults if investor optimism faded ^[7].

These episodes ended with an eventual reversal in investment, inducing economy-wide recessions. — Bank for International Settlements

Moody's extended that historical analogy into present-tense operational reality days later, warning that the financial system had developed a "systemic dependency" on a small set of AI providers — the same providers whose debt the banks are underwriting — and that a model outage at one major provider "could potentially spread quickly across customers and sectors" ^[8]. Goldman Sachs chief credit strategist Amanda Lynam then added a time dimension, flagging 2028 maturity walls as the point where concentration risk becomes refinancing risk ^[9]. Apollo chief economist Torsten Sløk added speed: the AI investment cycle is building at nearly twice the pace of the 2000s housing boom, and a cycle that builds that fast can unwind at a similar pace ^[10]. Société Générale strategist Albert Edwards connected the AI credit cycle to the broader economy, noting that the AI wealth effect was driving consumer spending and that the credit intensity of GDP had reached a 70-year high of 3.73 — making the economy, in his words, like the Wile E. Coyote character suspended in thin air before the collapse ^[11]. Then, in early July, career analysts at the U.S. Treasury authored a draft report warning that the AI market was forming a bubble more deeply entrenched than the dot-com era. The Treasury spokesperson dismissed the draft as unvetted, but the report's existence — and its authors' identity as nonpolitical staff — meant the warning had reached the government's own balance-sheet analysts ^[12]. Ray Dalio and Jeremy Grantham arrived at the same conclusion from the valuation side: Dalio identified the critical risk as investors borrowing against overvalued assets, the precise mechanism that turned the 2008 housing correction into a financial crisis, while Grantham called the current environment a "bubble within a bubble" ^[13]. Which brings us back to the SEC exemption. The risk-retention rules that were lifted from data-center asset-backed securities were the central regulatory response to 2008 — the requirement that loan originators keep skin in the game so they could not simply originate, package, and offload bad debt. The SEC has now removed that requirement from the exact instruments at the end of Nvidia's chain. The guardrail is down. And the people who would decide whether to raise it are divided. New York Fed President John Williams says he does not see a bubble. "I'm not as worried about the financial stability from the leverage right now," he said, pointing to high corporate earnings ^[14].

I’m not as worried about the financial stability from the leverage right now. — John Williams

Kansas City Fed President Jeff Schmid sees something different. "We have to correlate what's happening in AI, just from a pure scale standpoint, to some of the other experiences we've had that could, in fact, create a systemic problem," he said, explicitly invoking the "too big to fail" framing ^[15].

We have to correlate what's happening in AI, just from a pure scale standpoint, to some of the other experiences we've had that could, in fact, create a systemic problem. — Jeffrey Schmid

Two Fed presidents, the same institution, opposite reads. BlackRock, the world's largest asset manager, splits the difference: it maintains its overweight AI position and says "the fundamentals look good," but acknowledges that "risks have risen as well the last few weeks" and has declined to add to its conviction ^[16]. This is not a disagreement that will be resolved by more data. It is the same internal split that preceded the last crisis — between those who saw a contained correction in a specific asset class and those who saw a systemic threat running through the financial architecture itself. The pattern is recognizable. The chain is built. The guardrail is down. And the people who would act are divided.