Money Stuff: AI-Backed Securities

Subscribe to Bloomberg.com for unlimited access to all our coverage.

The basic business of finance is slicing up cash flows. You have some stuff, the stuff generates cash flows, you put the stuff in a box, the box issues securities to investors, the securities have claims on the cash flows. Some investors want safe securities with senior claims on the cash flows: If the stuff is expected to generate $100 of cash flows, you can probably issue investment-grade debt with a claim on the first $20 of the cash flows. Some investors want high-risk, high-reward securities with junior claims on the cash flows: If the stuff is expected to generate $100 of cash flows, someone will probably pay something for a stock option with a claim on all of the cash flows above $120. Etc.

This works for many, many things that you might put in the box. If you own a widget business, you can put that business in a box called a “corporation,” issue bonds with senior claims on the business’s cash flows, and issue stock to investors who want the risky upside of the business. If you’ve got a rockets-and-satellites-and-enterprise-AI-and-social-media business, you can put that in a big corporate box called SpaceX and issue bonds and stock. If you own some office buildings, you can put them in a box called a “real estate investment trust” or “REIT,” issue some bonds with a senior claim on the rental income and issue stock to investors who want the upside from your rent. But you can also put purely financial stuff in a box. If you own some mortgage loans, or some credit card receivables, you can put those in a box and sell junior and senior claims on them. Etc.

People often use the word “securitization” to describe this process: You put stuff in a box, you issue securities, it’s securitization. It would not, I think, do huge violence to the English language to describe every public company as a “securitization”: SpaceX put its rockets-etc. business in a public company and issued securities to finance it so, why not, securitization. In practice, though, people do not use the term this way. A “securitization,” in normal usage, mostly means putting financial stuff — mortgages, credit-card receivables, car-part invoices, what have you — into a box and selling claims on them. (There is such a thing as a “whole business securitization,” but that’s weird.) 

In fact US securities law makes this distinction. Under the law, “securitization transactions” are subject to different rules from regular securities. (Most notably, securitization issuers are required to retain some “skin in the game” rather than selling all of the cash flows; there are also different disclosure rules.) A “securitization transaction” is defined to mean one involving “asset-backed securities.” Of course all securities are asset-backed: SpaceX stock and bonds are backed by its rockets and data centers, etc. But the technical term “asset-backed security” means “a fixed-income or other security collateralized by any type of self-liquidating financial asset (including a loan, a lease, a mortgage, or a secured or unsecured receivable) that allows the holder of the security to receive payments that depend primarily on cash flow from the asset.” That is, an asset-backed security is a security backed by a financial asset (not a factory or a business, but a financial claim) that is self-liquidating (that gets paid down over time until its value is zero). 

So a mortgage-backed security is the paradigmatic sort of ABS: You put some mortgages in a pot, you sell some securities with claims on them, they get paid off over time and the money is used to pay back the securities. A regular company is the paradigmatic sort of not ABS: You put a whole business in a pot, you sell some securities with claims on it, the business intends to grow and thrive and change over time, the pot (a corporation) has a perpetual life, etc.

These days, though, the stuff that everyone is interested in is not mortgages or businesses but data centers. You put a data center in a box, the box issues securities to investors, the securities have claims on the data center’s cash flows. You could imagine different ways to structure this. At one end, you could have a regular company that develops, builds and operates data centers, that finds tenants and power and chips for the data centers and tries to turn a profit, and that issues stock and bonds to finance itself. That approach exists; it more or less describes “neocloud” companies like CoreWeave Inc. [1]

At the other end, you could imagine building a data center, signing a long-term lease with Anthropic or OpenAI, and then putting the lease in a box. You have contracted to get, say, $1 billion a year from Anthropic for 10 years, you put those contractual cash flows in a box, you sell the first $100 million a year as AAA-rated bonds, etc. You don’t see a ton of this approach. Everything is too new and fast-changing. In 10 years, Anthropic might not exist, and you might have re-rented the space to someone else. You might have to tear out the chips and put in new chips. Etc. Running a data center is, at this point, like running a business; there are decisions to be made, and you can’t just do it on autopilot. 

And so a lot of what you see is somewhere in between. A big asset manager arranges the financing, a big tech firm provides the chips and guarantees their value, a big AI firm signs a lease, the cash flows are all reasonably predictable and they get packaged into rated bonds that are sold to investors. The stuff in the box is not just the lease, not just a series of contractual cash flows; the stuff in the box is the data center, the chips, the actual business assets that create the cash flows. The data center financing is colloquially a “securitization,” but it is not technically a securitization. 

Or that’s the argument. Bloomberg’s Jack Trapanick and Scott Carpenter report:

The Securities and Exchange Commission has made it easier for data center owners to sell asset-backed securities, potentially opening the door for more debt sales as tech firms scour Wall Street for ways to pay for artificial intelligence.

The SEC said a major subset of data-center securitizations don’t need to have disclosures and investor protections that similar deals require. That includes risk retention, a requirement that companies issuing asset-backed securities retain some of the debt to better align their interests with investors.

In a letter late last month, staff wrote that data centers aren’t financial assets that liquidate over time, like loans or leases, and therefore bonds tied to them aren’t subject to the same rules as debt backed by car loans or home mortgages.

Here is the SEC’s interpretive letter, and here is the letter from Latham & Watkins LLP requesting it:

We respectfully request that the Staff concur with our view that fixed-income or other securities issued in data center securitizations of the type described in this letter (“Data Center Securitizations” or “DCS”) are not Exchange Act ABS because the securitized assets are not “self-liquidating financial assets” and therefore payments to investors do not depend primarily on cash flows from self-liquidating financial assets.

The basic point is that a data center is more like a business than like a mortgage or a lease:

Although a portion of the Securitized Assets include customer contracts (including in some cases leases) that generate revenue necessary for the Issuer to make payments due on the securities, the revenue generated is reduced by expenses to operate and maintain the data center facility (e.g., taxes, insurance, electricity, repairs and maintenance, security services, etc.). Consequently, the payments to the holders of the securities do not “depend primarily on cash flow from the [self-liquidating financial] asset” because the amount of cash available to pay investors also depends on the effectiveness of the operator or manager to minimize the operating expenses of the Issuer.

The investment opportunity in a DCS is most like an investment in a real estate investment company. In general, a real estate investment company acquires, owns, finances, manages, leases, and develops real estate and may raise capital to fund these operations. Similarly, an Issuer of a DCS acquires, finances, manages, and develops data centers and as discussed above, the proceeds from the issuance of DCS may be used for a variety of purposes. An investor in a real estate investment company is exposed to the entire operations and activities of the company and all assets and liabilities. In contrast, a DCS investor is isolated from the other activities of the operator or manager, a hallmark of securitization financing. However, this type of securitization directly funds the operations of a business, and the proceeds may be used for a variety of purposes, whereas Exchange Act ABS are issued to fund the purchase of a pool of self-liquidating financial assets. 

There are some diagrams. It’s not obvious to me that data-center financing has to always work like this. You could imagine a future in which, say, SpaceX builds data centers on its own balance sheet, leases them out long-term to Anthropic, packages the leases and sells them as asset-backed securities. In a world where “compute is an investable asset,” perhaps compute-backed bonds will in fact be asset-backed securities. But for now, building AI infrastructure is more like a business.

Donald Trump is selling early access to US policy decisions to hedge funds and trading firms for about $100,000 per month. (Technically his company, Trump Media & Technology Group Inc., is selling a feed of early access to policy announcements he makes on TMTG’s Truth Social site, but Trump is TMTG’s biggest shareholder.) The pitch for this service is that Trump’s policy decisions are market-moving, and if you get them a few milliseconds before everyone else, you will be able to trade on them and make money. This isn’t true for everyone, but it probably is true for an algorithmic trading firm; if you are paying for lots of fast data feeds, you might as well pay for a fast data feed of US government policy.

This raises at least two obvious problems:

1. It seems bad for the president of the US to sell early access to government decisions for his personal profit? Just, like, as a citizen, I find that weird?
2. It’s maybe insider trading? Arguably this information belongs to the US government, and arguably Trump is misappropriating it for his own benefit and selling it for cash. At the time it is sent to hedge funds, it is arguably non-public [2]  — they get it a few milliseconds faster than anyone else — and arguably material. In general, if you pay a corporate or government insider for material nonpublic information and use that information to trade, that’s illegal insider trading. That does seem to more or less describe what is happening here, though there are counter-arguments. [3]

What can you do about it? The first problem is, it seems to me, basically a political problem: If the president is using his government policy decisions to profit personally, the remedy is for Congress to say “hey maybe stop doing that.” I would not count on that happening. “Government watchdogs and congressional Democrats have criticized the service,” notes the Wall Street Journal, sure.

The second problem is one of insider trading. The normal remedy there would be for the US Securities and Exchange Commission, or federal prosecutors, to say “hey don’t do that” or bring enforcement actions. This, for reasons, is not going to happen any time soon. [4]  

Another possible remedy is for investors to sue. Trump announces a new war, a few hedge funds sell stock a millisecond before the war is public, stocks drop when the war is public, and people who bought stock in that millisecond sue. They sue Trump Media, I guess: “We bought stocks at inflated prices because we didn’t know about the war, but then the war news came out and stocks dropped. But you sold early news about the war to the market-making firms who sold us the stock; they were insider trading and you owe us money.”

There are problems with this theory. For one thing, I suspect that many subscribers to the Truth Social feed will be market makers who use it defensively to avoid trading when a war is announced. Even if they are trying to trade during their millisecond head start, the probability that you will trade with them is low; there will not be a lot of victims here. (The victims might be market makers.) Even if there are victims, the normal move is to sue the insider traders (the people who traded on the news), not the tipper (Trump Media). [5]

It’s just an interesting little gap. “Donald Trump is selling advance notice of policy decisions to financial firms so they can trade” does sound like it might be an insider-trading problem, but it’s not clear anyone can do anything about it. 

Anyway here’s a different theory:

President Donald Trump was sued by two media organizations over Trump Media & Technology Group Corp.’s plan to sell faster access to his posts on Truth Social.

The Intercept Media, a news organization, and the non-profit Freedom of the Press Foundation asked a court in New York Wednesday to block the plan, calling it “extraordinary, corrupt, and unconstitutional.”

Well, sure, but why do you get to sue? Here is their complaint:

Plaintiffs are harmed by the President’s illegal scheme. The Intercept Media, Inc. is an award-winning nonprofit investigative news organization that frequently covers the President’s posts. Freedom of the Press Foundation is a nonprofit organization that supports public interest journalism and operates the “Trump Anti-Press Social Media Tracker” database, which aggregates and categorizes the President’s Truth Social posts attacking the media. Both face indefinitely delayed access to the President’s latest posts, and face permanent bars to the President’s archived posts, making it harder for both organizations to do their jobs.

The Constitution guarantees that public officials “can themselves have no pecuniary interest or proprietorship, as against the public at large, in the fruits of their [official] labors.” … The President is profiting by selling government information. That is illegal. Plaintiffs bring this case to stop it.

I guess their theory is that, as journalists, they are entitled to get Trump’s policy decisions at the same time hedge funds do? Maybe?

The basic situation is that US states have historically regulated sports gambling, either banning it or allowing it in regulated forms. But recently prediction markets like Kalshi Inc. have started offering sports gambling and claiming that, as federally regulated commodities exchanges, they are exempt from state regulation. In the past, the federal regulator, the US Commodity Futures Trading Commission, had not allowed prediction markets to offer sports gambling, because the CFTC’s view was that it was not in the public interest to offer sports gambling on federal commodities exchanges. But the current CFTC has reversed this position. Now its view is not only that federal law allows sports gambling, but also that it prohibits state regulation of sports gambling, and the CFTC has gone to court to argue that states can’t interfere with Kalshi’s sports betting operation.

This is still working its way through the courts. We talked last week about a lawsuit that New York brought against Kalshi, arguing that it offers sports bets in violation of New York gaming regulation. This is obviously true, but also perhaps irrelevant; Kalshi’s response was that “states can’t just shut down a federally licensed exchange.” Federal courts have largely agreed with Kalshi about that, but there are exceptions, and who knows what will happen here.

In the meantime, though, Kalshi and the CFTC have declared an emergency. The emergency is that a court might agree that Kalshi is violating the law. The CFTC says:

The [Commodity Exchange Act] requires the Commission to provide a uniform national market in derivatives transactions. As part of this obligation, the CFTC ensures public confidence in its markets by safeguarding market resilience and orderliness. The Commission is also tasked with providing competitive, fair, and efficient markets that protect the price discovery process of trading in the centralized derivatives markets. Major market disruptions hamper these efforts. 

And so it is, I think, ordering Kalshi to ignore any court orders? The CFTC’s order says:

Under the Commission’s statutory emergency powers, it may direct Kalshi and its affiliates to continue to perform its functions as an exchange in accordance with the CEA’s Core Principles and its normal practices. This exercise of the Commission’s emergency authority will give market participants the necessary assurances that a CFTC-registered [designated contract market] cannot be shut down by a single State and that the trades they execute will be duly cleared and fulfilled. 

NOW THEREFORE:
IT IS HEREBY ORDERED that, pursuant to Section 8a(9) of the CEA, the Commission having reason to believe that the threat to its market justifies the exercise of its statutory emergency power, Kalshi shall continue to perform its functions as an exchange in accordance with the CEA’s Core Principles and its normal practices. 

I don’t even think the legal analysis is really wrong; I just think it’s amazing that a federal regulator has concluded that any effort to shut down sports gambling would be an emergency for financial markets.

Two theories about the financial industry are:

1. The financial industry creates a massive waste of talent: So many of the best and brightest scientists end up working at high-frequency trading firms, making a bit of extra money for those firms in a largely zero-sum competition instead of curing cancer or discovering cold fusion or whatever.
2. No, actually, it is good that the financial industry sucks up all of the scientists and pays them really well to work on trading systems. For one thing, they make capital allocation more efficient and that’s good. For another thing, this system makes scientific training lucrative: If getting an astrophysics PhD is a good path to quant-trading riches, more people will get astrophysics PhDs and some of them will end up doing astrophysics. Also, though, the problems that they work on — moving data around faster, analyzing it better — often have more general applicability. Many advances in communication are motivated by getting stock prices faster, and Brett Harrison posted on X that “many technologies now commonly associated with AI were created for” high-frequency trading firms. 

There are right now any number of quantum physicists who are working on extracting information from corporate earnings calls, because hedge funds can make a lot of money if they can trade quickly on corporate earnings calls. In one sense, putting all those quantum physicists to work analyzing earnings calls seems kind of sad and drab; shouldn’t they be doing cool quantum physics stuff? In another sense, though, it seems obvious that techniques developed for extracting information from corporate earnings calls would have broader applications. Extracting information from natural-language text, extracting information from natural-language audio, analyzing tone and body language: These are all things with immediate and lucrative application to stock trading, but they are also at the core of a much larger boom in artificial intelligence. The financial industry can subsidize more general breakthroughs.

And if all those quantum earnings-call analyzers get really bored they can make an AI music app? Apparently? Bloomberg’s Lucas Shaw and Ashley Carman report on “Suno, a generative artificial intelligence startup in Cambridge that lets people be rock stars without ever picking up an instrument.” Apparently Suno’s founders started out by generating corporate earnings calls using AI, but then realized that the same skills could be applied to songs:

[Suno CEO Mikey] Shulman was in the final year of getting a doctorate in quantum physics at Harvard University when he interviewed at Kensho, a software company in Cambridge that creates tools for financial companies. While working there part time, he befriended future Suno co-founders Georg Kucsko, Martin Camacho and Keenan Freyberg. They worked on a product that used machine learning to automate corporate earnings calls. Toward the end of 2021, they had two realizations: They enjoyed working together, and they thought there was a lot more they could do in AI audio.

They quit Kensho and started Suno, which means “listen” in Hindi. At first the company developed technology for turning text prompts into audio, which it called Bark. The outside developers who started to play with Suno’s software, though, wanted to make music. Shulman and his co-founders were happy to shift the company in that direction; they’d often had jam sessions in Camacho’s basement while at Kensho. In November 2023, Suno released a product that made writing and creating songs as easy as taking photos on a phone.

I suppose things could come full circle. A core thesis of this column is that you should never participate in a corporate rap video, but does that apply to Suno? If you are a the chief executive officer of a public company, and you want a punchy rap song for your executive offsite or for, heaven forbid, your earnings call, now I guess Suno can make you one.

Nvidia’s Show of Financial Force Soothes Credit Markets. Why Wall Street and Nvidia Are Building an Exotic Money Pipeline for the AI Boom. AI Looms Over Software Companies — and the Investors Who Piled Into Them. Goldman Sachs to Acquire ETF Provider Neos in $2.3 Billion Deal. Most Power Sought for US Data Centers Will Never Materialize. Citadel Securities Says Prepare for a Reload on Levered Stock Bets.  Jane Street Repaying $5.5 Billion Loan in Overhaul of Debt Load. Blue Owl Raises $750 Million to Pay Down Credit Lines. Energy Startup Raises $750 Million for Rust-Powered Batteries. Josh Kushner, Bob Iger Buy LA Lakers for $12.5 Billion. Norway wealth fund CEO says fund’s entire value could be lost if market collapses. Wealth Management Has a $3 Trillion Problem: Investors Are Keeping Too Much Cash. Former Hedge Fund CFO Pleads Guilty to Embezzling $3 Million. ‘Call Her Daddy’ Host Alex Cooper Raises Money at $500 Million Valuation. Harvard Students Are Already Stressing Over the Crackdown on A’s. Chinese robot maker’s IPO 5,500 times oversubscribed by retail investors. The Largest Tweetable Number.

If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Or you can subscribe to Money Stuff and other great Bloomberg newsletters here. Thanks!