Nvidia is looking to turn AI infrastructure into a massive investable asset class, giving Wall Street to supply the capital needed to build data centers that will ultimately create demand for Nvidia chips. A recent SEC interpretation may make some of these financing structures easier to create.
Nvidia’s Innovative Solution
First let’s understand the problem Nvidia is trying to solve. AI data centers are extremely expensive. Building an AI data center costs billions of dollars as companies have to pay for: GPUs, servers, networking equipment, buildings, electricity infrastructure, cooling systems, and construction.
Many AI companies don’t have enough cash to pay for all this themselves, or simply don’t want to.
Nvidia is trying to create a financing ecosystem where banks, private-equity firms, asset managers, pension funds, and other sources provide the money.
Recently, Nvidia announced collaborations with major financial firms like Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, with the goal of raising over $500 billion of third-party capital for AI infrastructure.
How Does The SEC Fit In The Picture
Imagine a company wants to borrow $1 billion to build an AI data center.
The debt can be packaged into debt securities and sold to investors. This is called securitization.
It’s similar to what happened with mortgages. A bunch of mortgages get pooled together, then securities get created and sold to investors.
Now instead of mortgages, it’s data center leases/loans.
Law firm Latham Watkins looked for guidance from the SEC regarding these data center financing structures. The SEC’s position reportedly means some of these structures certain securitization risk-retention requirements.
What Is “Risk Retention”?
Before the financial crisis of 2008, financial institutions could give out mortgages, package them into securities, and sell most of all of the risk to somebody else.
This created a potential problem: bonds can be supported by bad mortgages.
The incentive to underwrite the loan drops.
So Dodd-Frank brought risk-retention requirements for certain securitizations.
Now, if you create/package certain risky loans and sell them to investors, you have to keep some of the risk yourself.
What Does This Have To Do With Nvidia?
If certain data center debt structures aren’t subject to risk-retention requirements, the parties creating the financing don’t have to keep any risk.
This makes the financing structure cheaper, easier to execute, and easier to scale.
Nvidia doesn’t want to personally finance hundreds of billions of dollars of data center construction. It wants third-party capital to do it.
Here’s A Breakdown
Let’s say an AI company wants to build a $10 billion data center. It needs financing.
Old/more restrictive structure
A financial institution creates a securitization:
$10B data center debt.
If risk retention rules apply, the institution needs to retain a portion of risk.
That means the sponsor keeps some risk, while investors buy the rest.
This limits how much the financing the sponsor can create.
Potential New Structure
If the debt structure falls outside the relevant risk retention requirements then the sponsor can create the financing and investors can essentially buy all the securities.
The sponsor doesn’t have to keep any risk on the books, making it easier to create large amounts of financing repeatedly.
Why This Matters For Nvidia
Nvidia’s business depends on enormous demand for its GPUs.
But eventually there’s a bottleneck:
Who is going to pay for all these GPUs?
Microsoft, Meta and Google can spend billions.
But smaller AI companies and cloud providers may not have the balance sheets to purchase huge quantities of Nvidia hardware.
So Nvidia is essentially trying to help create a financial infrastructure for purchasing AI infrastructure.
The Wall Street Journal described the idea as creating a new asset class around financing Nvidia’s chips, somewhat analogous to how investors finance things like aircraft, credit-card receivables and mortgages.
What Investors Need To Understand
Third party financing could: speed up AI infrastructure construction, increase Nvidia GPU demand, allow smaller AI companies to obtain computing capacity, move some financing risk away from Nvidia, create a new market for institutional investors, turn AI infrastructure into an investable asset class.
There is a really obvious comparison to the 2008 financial crisis, however.
In 2008 mortgage securitization created enormous amounts of credit which led to weak underwriting and defaults creating the financial crisis.
This AI infrastructure financing could create enormous amounts of credit which could potentially lead to aggressive underwriting.
Now this doesn’t mean we’re headed for another crisis, data centers aren’t subprime mortgages.
But the concern is about incentives and leverage.
If investors become convinced that AI infrastructure is an incredibly safe asset, they might provide these massive amounts of debt.
It raises the question: Is there going to be too much money chasing AI infrastructure projects whose numbers haven’t been proven?
There are already talks of investors being hesitant to finance assets like GPUs because of their high depreciation in a rapid technological environment.
