AI Bankers Gone Wild

There is growing evidence that banks are throwing caution to the wind in the race to be AI’s bankers. This brief article originally started about traditional banks, but the blockbuster $500B “deal” Nvidia announced this week with 6 non-traditional bankers (Goldman aside) demanded it be addressed.

I will walk through several major related developments recently in the news. This is not all encompassing but focuses on more recent, somewhat unhinged funding announcements without any due regard for whether the revenue will materialize to justify the debt bacchanal. There are also the early seeds of growing caution which could in itself present risks to the AI financing ecosystem.

Turning to perhaps the biggest example of bankers gone wild is the $40B bridge loan commitment to SoftBank to buy OpenAI shares due in March 2027. Few seem to be aware how precariously the AI bubble rests on the narrow shoulders of Masayoshi (AKA Masa) Son, founder of SoftBank. You see, the thing is, Masa doesn’t have the cash. While some of his OpenAI commitments were funded by selling stakes in other holdings and going in hock with others (eg margin finance) the latest commitment was so large he needed bankers to fill the gap.

The banks rolled the dice with their $40B bridge loan, but they were really, really hoping (is praying too strong a word?) that OpenAI would IPO before the loan came due in March 2027 and they would get repaid from IPO proceeds. But when OpenAI’s financials leaked and investors saw how truly grim the numbers looked, the IPO was put on hold.

On around June 25, it was clear the IPO was going to happen. One month later, once the banks realized that the IPO wasn’t going to take them out in time for the March 2027 deadline, the 6 original lenders decided to share the wealth/risk and broadened the syndicate to 21 added bagholders banks.

Unfortunately for Sam Altman, OpenAI didn’t have the same ability to spin a fanciful tall tale of revenue opportunities like SpaceX – colonizing Mars, extending intelligence to the stars and mining asteroids. When your business plan is straight out of a Star Trek script, 20 Wall Street analysts will giddily make up numbers to validate Elon’s SciFi fever dream and desired IPO validation.

OpenAI’s Sam Altman could only offer prosaic terrestrial opportunities and a really ugly financial picture. It’s a lot easier to produce fictitious revenue forecasts from datacenters in space than enterprise AI contracts or consumer ad revenue (the latter of which is 90% below OpenAI’s projections as I previously reported).

As for the $40B loan, what happens in March 2027 if there is no IPO and likely the bond market is less willing to roll the dice? Will the banks extend, transfer risk to others EG via SRTs (see below for more on SRTs) or perhaps call the loan due and payable thereby revisiting on Masa the post WeWork nightmare.

Subscribe to my Substack for an upcoming more detailed fascinating look at SoftBank and their unique role in the AI boom.

I first discussed this debt in my SpaceX pre-IPO article. Banks extended a $20B bridge loan to SpaceX in the hopes that they would be paid back from the IPO proceeds. That didn’t happen despite the S1 registration statement indicating it would. Thanks to pliant rating agencies and the willingness to provide a singularly insane BBB rating to SpaceX, banks found other bagholders, who may be insurance companies who love high yielding “investment grade” ratings (yes, it’s true)).

And then there was the recent FT story where Morgan Stanley was looking to refinance $15B in loans in the increasingly saturated and turbulent bond market. The Morgan Stanley led debt is for ONE data center in Texas for ONE client Anthropic. A client with no credit rating and no near term plausible revenue path and a single data center.

Though Google provides the guarantee for the data center, the bonds are expected to be speculative grade because the guarantee only kicks in when the data center is built.

Which brings me to the latest AI funding shocker. Today 6 bankers/financers (actually, mostly private equity companies with an uncomfortable degree of control over insurance company assets) have committed to fund Nvidia’s $500B, quite possibly psilocybin induced, funding goals.

These “funders” include Apollo Global, Blackstone, BlackRock Brookfield Asset Management, Goldman Sachs, and KKR. Let’s be clear. Most of these are PE sponsors and DO NOT have the balance sheet for this. They are operating with OPM – Other People’s Money - mostly pension and insurance money.

It’s not clear how much of this is new commitments vs just rolled up existing commitments but given the numbers and players involved it’s important to see how the risk is evaluated and mitigated (eg credit ratings, guarantees), structured (public corporates, infrastructure bonds, CMBS, etc.) and distributed (public vs private, captive insurance and pension money, etc.).

Other than Goldman, I don’t believe any of these funders have the balance sheet to take down the debt. It is especially true with Brookfield, KKR and Apollo that they directly control the insurance balance sheets that can be used to absorb this debt.

I have a suggestion for the 6 leaders of the above organizations: I saw you on CNBC couch the opportunity as existential and vital for national economic security – why don’t you pool partner personal net worth to back the debt? Why don’t you commit to limit the amount of leverage used? Why not commit to reducing fees by 50% to help the project viability? Of course none of this will happen.

The insurance regulator, NAIC, is already scrutinizing insurance company data center investments. This deal should be a 4-alarm fire for the NAIC. Will they take up the challenge.

Let me help.

Dear NAIC: Please immediately call Apollo, KKR and Brookfield and ask three simple questions:

1. What will be the structure of the debt? Will it be public or private?

2. How much do you expect will be placed on insurance balance sheets?

3. Who will rate it?

4. What is your specific, defensible forecast that sufficient revenue materializes and that the data centers can get built on time and on budget?

Reuters reported this week that bank lenders are increasingly interrogating the underwriting assumptions behind US data center financings — a signal that capital market discipline may be arriving in a sector that, until recently, was financed without due consideration whether there would be sufficient revenue to justify the cost.

Bloomberg recently reported that utilities were increasingly requiring bank guarantees to protect utilities from data center developers requiring building out massive energy infrastructure only to face a default from the developer. This could force banks into ever greater data center related exposure in order to protect existing investments. Or it could be one of the many ways the data center build out could be slowed or scuttled.

A Tech Times article addressed increasing banker concerns over exposure to community backlash against data centers. While the current administration is free to have a “Make Air Pollution Great Again” policy, it turns out that people in the communities who have to live with the data centers are not obliged to passively accept it.

Finally, the recent spate of AI data center related Synthetic Risk Transfers (SRTs) signals the desperation of banks to reduce their exposure to datacenter and AI debt. TD Bank, SMBC, SocGen, BBVA, ING and Fifth Third are among those who announced multibillion dollar infrastructure related SRT deals. Details of the counterparty (often Private Equity related like Blackstone), attachment/detachment points and the structure of the SRTs are subjects for future research.