AI Debt: The Greatest Risk Transfer Since 2008

Each dollar an AI lab (like OpenAI and Anthropic) commits toward compute ends up as someone else’s backlog.

Neoclouds, hyperscalers, and even memory chip companies…

They all have turned your attention away from their core businesses into future promises and touts of “record backlogs.”

However, there is one problem:

These backlogs only turn into profit if AI labs can fulfill them.

Anthropic’s recent S-1 prospectus filing drives a significant wedge between the amounts they have committed and how much revenue the company actually generates.

Right now, they owe $11 for every $1 in revenue they bring in.

And,

That revenue growth is slowing from 147% last quarter to 45% this quarter, with the root cause being something most people have not been able to push back on so far:

So what does Wall Street do with products they bloated their balance sheets with and can’t find a buyer for?

They dump it onto unsuspecting customers/investors.

Like pension funds.

Click here to see how Goldman Sachs is already dumping their AI exposure.

In today’s analysis, we’ll go over a few things:

And more importantly,

How all of this risk chain ends up affecting a list of companies, stack by stack, starting with the ones breaking right now.

Let’s get into it,

AI Debt: The great risk transfer.

Let’s begin with a lay of the land, and understand exactly how (and where) each AI investment dollar flows.

The two largest participants used to be:

This is when you saw partnerships like Softbank show up to help NVIDIA finance its GPU-backed loans for other companies to be nearly forced into buying their chips.

Then venture capital would fund, along with NVIDIA, the AI labs like OpenAI and Anthropic.

Which allowed AI labs to commit to trillions of dollars in compute obligations and justify the “demand” for these loans and chips.

However,

Venture capital funds, along with private credit, have dried up as the cost of capital coupled with rising insolvency risks have shown up in the sector.

Their solution?

Tap into public funds, to where the largest participants are now:

That creates a whole different problem, because it opens the floodgates for these new funding sources to become the “hedge” if these investments don’t perform.

In 2025 and 2026, AI-related venture capital investments rose from 65% to 83% respectively.

Meaning a lot of these funds have a concentrated bet which is not only underperforming, but on the edge of being nearly written off as impaired.

Like I said above, the fact that new funding is opened from public sources (savings, pensions, insurance) allows for this risk to be transferred freely.

Just as it was in South Korea, where people began to place life insurance money into AI-related investments.

We all know what happened there, or at least, what is happening.

Click here to see how and why South Korea’s AI trade crashed the market.

I believe that handoff is already happening.

And it will probably boost a lot of retirement accounts and pension plans in the short-term, but if one thing goes wrong then average people will be the ones bailing Wall Street out as they always do.

As of June 2026, AI-related debt issuance is now larger than equity being invested from venture capital funds.

$570 billion compared to $427 billion.

In other words, as the flow chart explains above:

All done through debt.

Your 401(k) may now have some private credit exposure without you even knowing it, and the private credit vehicles being bought are acting as a credit line to the venture capital funds losing billions in their AI bets.

There’s nothing wrong with this idea if you ask me.

If average investors can have access to private equity/credit, and participate in a fair share of another section of economic growth, then so be it.

My only problem is that the only private placements are being executed toward AI loans and venture capital (the riskiest vehicles in the market right now.)

Which leads me to a very important assessment…

Investors should make it out okay, and even with some profits, if the quality of the revenue is adequate compared to the debt terms.

It’s not.

As covered in my previous analysis, which you can check out here.

Roughly half of all U.S. large tech firms are choosing to shift their AI provider to the platform that makes most economic sense.

Meaning, the one that can deliver the same commodity for cheaper.

That’s why you see over 50% of all companies in the U.S. choosing to go outside of Anthropic and OpenAI, despite them being the most popular and fast adopted.

In a nutshell,

80% of all AI spend for Anthropic and OpenAI comes from the top 1% of firms, like Microsoft and Oracle.

The other 20% comes from everyone else, like small to medium-sized entrepreneurs spending an average of $7,500/month on credits.

Even if they 10x the amount of people spending that much on the platform, it would still fall short of how much one big tech firm can spend (hundreds of billions.)

Were this any other type of business, with that much debt relying on its revenue quality, and I would be the first to tell you to liquidate.

At the current rate, the pensions, savings, and insurance money funding this debt is not going to recoup their principal.

Add on top of that the level of customer concentration, and you really got a problem here.

Let’s look at a cloud business that underwent a similar stage as these AI labs.

When Amazon Web Services (AWS) was getting started, even though its adoption and growth rates were far larger than what we see today, they still relied on some VC funding.

The moment monetary conditions, and the cost of capital, tightened, the flows from those funds started to slow down as well.

Which resulted in a major contraction in AWS revenue growth, as these sort of unproven startup businesses run mostly on outside investor money in the first few years.

Meaning,

The party is beginning to come to an end as the ecosystem runs out of investors, maybe why NVIDIA sought to land deals with insurance companies earlier this month and even recruited top investment banks to package more of these loans.

If the money stops, the entire AI ecosystem stops, simple as that.
It is not self-sustainable.

Knowing that equity capital is threatening the party created the need to shift into debt to keep it fueled.

When debt dried up, these people went into public funds (your savings, retirement funds, and tax dollars) to continue inflating the bubble.

Now, the debt has grown so large that some companies just can’t find ways to legally hide it from you.

On and off-balance sheet AI commitments (most of which we know will go unpaid) now add up to $3.7 trillion dollars.

Bigger than the $1.5 trillion subprime mortgage crisis of 2008.

But wait, the economy is much bigger than it was back then!

Yes it is, but:

So you tell me the difference.

With all of these pending tail risk events, it’s time to get clear on one thing.

Where is the collapse going to hit first and where can we make a profit from it all?

Neoclouds are the most exposed, as they have bet a large share of their entire market capitalizations on these backlogs and became dependent on AI lab ability to repay.

Premiums for debt are now above 10% on average, effectively junk status.

On the other end you have the hyperscalers.

These are the places where you can get as close to “bullet proof” balance sheets as possible, and why Meta’s debt - for example - trades in line with the broader credit market risk.

That creates the barbell of risk versus reward.

In the middle, you have all of the energy, infrastructure, and lateral beneficiaries of the buildout that may still continue to work.

However,

Even that space is becoming a mess, and I can say that because we hold a few AI long/short equity spreads in the Offside Portfolio.

It seems like the market just doesn’t care about incoming or promised cash flows, even if it has nothing to do with these loans directly.

The way I see it…

You’re better off protecting yourself by rotating along with the market, just follow the money.

If you missed the AI boom then you missed it, you should be here to make money in the long run rather than potentially be wiped out and no longer able to play the game.

Which is why my more “boring” picks coming out of the PMI reports have managed to outperform the S&P 500 this month along with most of the AI complex.

Even this defensive micro-cap company has trashed the market this week, up 10%+ just today!

That’s the area I am focusing my energy on, and where I would suggest you focus yours.

Coming from someone who made a killing by buying Intel in the early AI days for a 10x return, gotta know when to walk away.

Click here to see this boring multi-bagger opportunity.

I remain focused on the memory complex as one of the best ways to express this idea.

Some companies, like Micron, have seen their forward P/Es come down to multi-year lows.

Why?

Markets understand that this industry overshot its investment intensity, and that will lead to oversupply scenarios if demand falls below projections.

Judging by everything we covered today…

I think that’s a very probable scenario coming up.

Still, the timing is far from obvious (as with all bubbles) and remains one of the biggest risks to the memory trade.

So I am approaching it through a long/short options strategy, which deserves the type of attention, guidance, and hand-holding you won’t get in Free Offside newsletters.

I am already in this Micron trade, in fact I opened it in August.

Even after being wrong for nearly three months on the range, I still managed to remain in profit and will even make a decent amount by November expiry.

When you have high conviction on an idea, but are uncertain of the timing, a different strategy needs to be applied.

It’s exactly what we expose you to inside Offside Premium.

That stock that’s up 10% today?

We are also selling cash secured puts on it every month, amplifying our returns as we continue to accumulate a position.

Rotation markets are hard, and I hope you don’t get caught in the tide that tends to wash out pretend traders.

See you on the other side.

Meanwhile, here is the latest from Jordi Visser (the retail spokesperson for retail AI investments) shifting his narrative once again.

Go back to his videos from early 2026, you will notice his thesis was actually founded on true numbers and developments, to then focusing on “non-human time” and saying finance itself is changing.

Accounting is the language of finance, and 1+1 will always equal 2 in my view:

Until next time,

OFFSIDE RESEARCH

Against the Tape, Ahead of the Curve.