I Modeled Oracle's AI Delay (It's Worse Than You Think)

The internet bubble of 2000 marked a very important inflection for the 10-year bond yield in the U.S.

It reset the benchmark in what yield level can be actually considered restrictive or “dangerous” for other asset classes like equities.

That level is now 5.0%.

As of September 2026, we have printed four closes above 5% on these bonds, the first time since July of 2007 as a lead up to the credit crisis.

Like back then, this is not an inflation issue despite what talking heads say about Iran and oil supply disruptions…

In fact, over 60% of the move in the 10-year this year can be explained by term premiums alone.

Forget Fed rate hikes and inflation, it’s mostly priced in already.

Markets are now focused on the fiscal and economic uncertainty circling the S&P.

Click here to understand how AI earnings and debt are driving term premiums (and bond yields.)

I believe bonds will stay above 5% for longer, and could even go past 6% unless the market is allowed to cleanse itself.

This week, Oracle may have fired the first shot to initiate that cleansing panic.

“Force Majeure” was effected on a New Mexico data center, delaying projects for everyone else involved while also raising concerns for further projects to go into development.

It just so happened that OpenAI hacked into an Australian government entity in the same week, which triggered more regulatory delays.

At 3% or 4% bond yields, the gap hurts.

At 5% though, it starts changing the entire return profile of all AI investments.

The problem now compounds into something much bigger:

In other words,

Demand can remain intact while the investment still fails.

This is not a “sell everything” newsletter, but rather the initial compass you can carry with you as we enter the final stages of the market cycle.

From this research, I will point you in the direction of some beneficiaries in this delay, and who stands to lose the most (should you want to take a bearish bet.)

Let’s get into it.

I modeled Oracle’s delay (it gets much worse.)

Let’s keep it simple here.

The long-term average trigger yield for the 10-year (according to Invesco) is set at 4.72%, and by trigger I mean the level which - if crossed - begins to break the stock market.

At the current yield of ~5.11%, we will see a 12-month moving average cross above 4.72% by February 2027 or even sooner if my thesis on term premiums is right.

Now let’s test this “rule” stated by Invesco, and see what actually happens to the S&P 500 after this 4.72% level is crossed:

The data very clearly shows there are two probable outcomes for the stock market from here:

We are currently headed to a path that most resembles a 12-month moving average crossing 4.72% and the 10-year yield remaining above it as well.

Meaning,

The next decade of S&P returns are projected to be around 4.6%, which adjusted for inflation would be a mediocre ~1.0%.

Yes, that historically has made bonds and emerging markets a much more attractive investment until S&P 500 valuations come back to competitive levels.

This is why the Yen and Japanese equities have been rallying so much.

Here’s the final point on this mediocre decade expectation on the stock market.

In every single occasion, the market has returned 0% to negative 5% returns in the following decade when CAPE multiples cross above 40.0x (we are now at 40.6x.)

Combine this with the 4.72% rule from Invesco, and we’ve got ourselves a real problem in the stock market.

In rising bond yield environments, investment logic is very clear as to which areas of the market should perform better/worse than others.

Let’s begin with the one who should suffer the most:

This includes the ARKK ETF as a benchmark, which itself is overweight a lot of companies that have become correlated to what happens in the AI trade.

Where the returns should have (historically) been closer to negative 6%, they are currently at ~8%.

The same thing can be said about semiconductors, where history commands a 0% to negative 1% return range, but currently sit with unprofitable tech at ~8% instead.

Now let’s pause because this breakdown is probably the MOST IMPORTANT aspect to the AI trade:

From NVIDIA’s earnings, you now understand the company is no longer a chipmaker but rather a lending platform to all other companies borrowing to expand on their AI projects.

Simply put,

Chipmakers and neoclouds are now real estate and lending companies.

Digging further into that idea, this is where it begins to prove real rather than just a thesis.

The following companies have moved higher along with rising yields:

Historically, and logically, all of these - with the exception of Apollo - should suffer negative effects from rising yields.

Proving they are now behaving like banks or property companies.

When you transform your balance sheets to carry data center properties, loans, and equity circular financing, this tends to be the result…

In other words,

This is no longer a theory, the market is beginning to treat these companies for what they have become.

The Jupiter data center delays, and further impacts on regulation from the continued OpenAI hacks, will likely cause a 2-2.5 year delay on future projects.

In terms of the financing, here’s exactly how it plays out:

Specifically,

This delay will result in a ~5-6% haircut for the project’s net return.

For end users and providers, this can be closer to 25-30% when you throw in the leverage levels that are present.

Simply put,

At 5% yields and rising, there is little to no incentive left to keep investing in these data centers considering how many headwinds are emerging to throw the IRR significantly lower.

Think of it like this:

The longer it takes to complete a fully operational data center, the longer it takes to monetize on a heavily leveraged project, and the more interest expense begins to build up as a result.

With NVIDIA and hyperscalers at the forefront of all this financing (now acting like banks), their earnings and EBITDA margins will take a massive hit on accruing interest.

Now if we visualize the situation:

Roughly 53% of the added costs to this Oracle data center, and those already in development, will come from accruing interest expenses.

All of which will be paid by NVIDIA, neoclouds, hyperscalers, and AI labs.

One of the reasons why credit default swaps (CDS) are spiking for these companies can be traced back to this development.

What these CDS products measure is simple:

As the price of these instruments increase, the market’s vote is clearly leaning to a higher chance of default.

Why wouldn’t it?

S&P 500 valuations pointing to a mediocre decade (no more equity financing.)
10-year bond yields crossing the 4.72% trigger.
Data center delays wiping out IRR on projects and spiking interest costs.

To finish everything off, I’d like to point you to the potential beneficiaries of the next wave in the AI trade.

It won’t be AI supply chain names, as their returns do not match the rapidly rising risks in the economics.

It won’t be chipmakers, since delays on compute demand will create a terrible oversupply scenario, which I pointed out to you in this post.

It will likely be cybersecurity names, who are now in charge of making sure all existing AI models operate within the baseline of safety.

The OpenAI hacks on Hugging Face were buried by the NVIDIA buyout…

You can’t hide a hack on government entities like the Australian health system.

So if you still have any ounce of faith in the AI trade, I would recommend you start looking into cybersecurity and software.

The “bottlenecks” on the supply chain are now effectively dead, companies that make chips are moving into lending and real estate, and insiders have dumped them all quarter long.

In fact, you can see the market is already betting on such an outcome.

Forward P/Es are the market’s way of assigning future expectations on company earnings.

And right now,

Cybersecurity names like OKTA and PANW have seen their multiples go up this year.

Meanwhile,

Chipmakers like NVIDIA and SNDK see their multiples fall.

I believe everything we’ve covered today is enough evidence to justify the market’s choice in this matter.

We are now in a stage where the timing is just as uncertain as the risks involved with AI.

I’ve given you the SMH, EWY, and several other trades in this topic free in this newsletter.

Today,

I cannot - in good conscience - point you to a clear bet, as it feels like the last whipsaw/shakeout is happening around AI, so giving you a one-directional bet would be doing you a disservice.

There are other directional bets given to you in this section of previous newsletter, like trucking, diesel, and even software.

But for this one, I would either sit back and watch it all unfold…

Or,

I would give Offside Premium a try, and see just how we’re deciding to play these AI outcomes in a responsible way.

That spread is now up over 25% since we opened it.

And the best part? It pays no matter which way the AI trade ends up playing.

Extreme markets call for different strategies.

Join Offside Premium Now.

Meanwhile, here is the latest from Goldman Sachs, covering the issue we talked about at length today.

Are rising yields truly a risk for the stock market?

Until next time,

OFFSIDE RESEARCH

Against the Tape, Ahead of the Curve.