Nvidia has announced a $500bn programme of AI infrastructure financing alongside Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. The announcement came just a fortnight after a Securities and Exchange Commission staff response clarified a key regulatory question for data centre securitisation. The exchange took six days to produce a letter that could reshape the way compute-heavy assets are funded.

At issue is whether data centre securitisations fall outside the Exchange Act definition of an asset-backed security. That definition matters because it triggers the risk retention rules written into Dodd-Frank after the 2008 financial crisis. Those rules require a deal sponsor to keep some risk on its own books, aligning incentives with investors and preventing the kind of lax underwriting that pushed residential mortgage securitisation into crisis.

What was asked, and what came back

Law firm Latham & Watkins wrote to the SEC on 23 July. It asked the staff a narrow question: do data centre securitisations fall outside the Exchange Act definition of an asset-backed security? The SEC’s Office of Structured Finance, chaired by Kayla Roberts, replied on 29 July. The staff agreed with Latham’s view.

The response is not a formal rule, but it carries practical weight. Sponsors and investors look to SEC staff guidance for comfort in structuring novel transactions. A written articulation of the staff’s view removes some of the uncertainty that had surrounded the fast-growing data centre securitisation market.

The argument Latham made

The Latham letter turns on a phrase. An asset-backed security, by definition, rests on a self-liquidating financial asset. Since 1992, the SEC has read that as one converting into cash within a finite period. A mortgage qualifies because repayment extinguishes the loan. Latham argued that a data centre does not. The facilities are tangible and physical, they endure beyond the life of the securities, and they may appreciate in value. When the notes are repaid, the issuer still owns the building.

That distinguishes a data centre from a single-asset commercial mortgage deal. In the mortgage deal, the issuer holds only the loan and ends up with nothing once it is repaid. A data centre, by contrast, retains productive value long after the securitised debt has been retired. On that comparison, the reasoning holds together, and the SEC staff accepted it.

Latham has worked on data centre securitisation since the first deal in 2018. It told the SEC that the market has since passed $50bn in cumulative debt issuance. The firm also described how the market had behaved: participants complied with the asset-backed rules throughout, the letter says, “out of an abundance of caution” rather than because the definition required it.

What the letter covers

The letter describes the securitised assets in detail. Buildings and data halls, electrical and backup power systems, cooling, network connectivity, physical security, land, and the contracts needed to run the facilities. Notably, it does not mention chips or graphics processing units. This is a narrow carve-out for the real estate and infrastructure that house compute, not the compute itself.

The letter also sets out the shape of these deals. Loan-to-value tops out at 70% of appraised value. Notes carry an anticipated repayment date of around five years. Final maturity runs 25 to 30 years. Nearly all use a master trust, the letter says. That structure lets sponsors issue further securities later, add data centres, and in some cases dispose of or substitute assets.

Investors generally have no recourse to the sponsor or the operator. The letter records the usual exceptions as fraud, wilful misconduct and gross negligence in managing the sites. That means the securitisation is intended to be genuinely non-recourse, with the physical asset and its contracts serving as the primary source of repayment.

What the letter says about itself

The SEC response sets its own limits. It reflects the views of the staff of the Division of Corporation Finance, not the Commission. The Commission has “neither approved nor disapproved its content”. It is not a rule or a regulation and has “no legal force or effect”. The staff add that their views rest on the representations in Latham’s letter, and that “any different facts or conditions might require the Division to reach a different conclusion”.

This self-limiting language is standard for SEC staff interpretative guidance. It gives industry actors a reasoned basis for moving forward, but it does not bind courts or future Commissions. A change in the underlying facts, or a new SEC leadership, could alter the analysis.

What lawyers say it means

Securitisation lawyers have welcomed the response. Orion Mountainspring, a lawyer at Orrick, said the response gives sponsors something specific: the chance to push down the equity required in a deal over time. He called it good news for them. More flexible structures are now likely, according to B.K. Lee at Alston & Bird, and more deals should materialise now that the guidance exists in writing.

Seth Messner of Katten Muchin Rosenman said Latham had asked the SEC to put these deals outside the risk retention rules, and the SEC basically agreed. Messner was more cautious on Nvidia, noting it is not clear whether its agreements are designed for securitisation, only that the guidance sounds applicable if they are.

Katten’s data centre partners placed the rules in context. They date from after the 2008 crisis, when securitisations of poorly underwritten residential mortgages set off a global financial collapse. Risk retention was designed to make sponsors share in the losses. Excluding data centres from that regime signals that the SEC views them as different in kind from the mortgage pools that caused the crisis.

Nvidia’s position and release cycle

Nvidia has not said whether the platforms it built with six finance giants will securitise anything. It agreed those arrangements through memoranda of understanding. Their stated purpose is to pool capital so AI labs, enterprises and cloud providers can reach compute hardware without drawing on their own balance sheets.

Nvidia has described its hardware as a revenue-generating asset, reaching for four adjectives: productive, long-lived, fungible and flexible. The wording suggests the company wants the market to think of GPUs as infrastructure assets that can be financed and refinanced in the same way as real estate. The SEC letter, however, only covers physical facilities. It explicitly does not address hardware.

Nvidia’s release cadence is a matter of record. The company has grown by moving large customers to its newest hardware close to annually. At Computex in 2024, it shortened its release cycle from two years to one. That raises a question framed by analysts: Nvidia wants customers to buy as many next-generation chips as possible, but it also wants them to know their old chips will stay valuable. Can it manage both?

The broader context is also worth noting. Reports have tied Nvidia to backing the buildings behind an OpenAI data centre in Ohio, to the value of $105bn. And prominent investors such as Jeff Gundlach have said turning compute into an asset class looks like a market top. The securitisation debate is therefore taking place at a moment of maximum enthusiasm for AI infrastructure spending.

What would settle it

Four things would clarify the picture, all checkable. Whether Nvidia discloses that any of the $500bn involves securitisation. Whether the staff position ever reaches hardware rather than facilities; the letter does not address that either way. Whether ratings agencies treat compute-linked collateral as they treat buildings. And where the risk finally sits. Five tech giants already hold $1.65tn off balance sheet, and lenders already issue GPU-backed debt.

The SEC staff letter is a meaningful development for data centre finance, but it is narrowly drawn. Physical data centres may now be funded outside the risk retention regime, yet the boundaries of that exclusion remain largely untested. As AI infrastructure spending accelerates, the distinction between a building and a chip could become one of the most consequential questions in structured finance.

Source:TNW | Nvidia News