If the market loses confidence in the “AI trade,” it’ll be a downward spiral for everyone involved.

In 2015, SunEdison was one of the hottest stocks on the market. It went from unknown to a $10 billion market cap, driven by billions in wind and solar energy deals and financing innovations that were the talk of Wall Street.

In 2016, the company filed for bankruptcy.

I was one of the people following SunEdison closely from the start. I never felt entirely comfortable with how SunEdison sold its never-ending ability to finance projects, but it wasn’t until 2015 that it became clear SunEdison was playing a confidence game.

If the market was confident in SunEdison’s business and the stock kept rising, the money pool was endless.

When confidence dried up, the pool dried up.

I see the same confidence game being played in AI today, and investors may be taking on risks they don’t understand clearly.

If you’re getting value from posts like this and want to get access to my market-beating Asymmetric Portfolio, become a premium subscriber below. Thank you for your support!

Here’s how SunEdison worked in simple terms.

There were two primary subsidiaries for SunEdison: TerraForm Power and TerraForm Global. Each of those companies was owned in part by SunEdison, but SunEdison also got preferred dividends from the companies. The TerraForm companies were financing arms known as yieldcos.

A yieldco utilizes a combination of debt and equity to finance renewable energy projects, which are typically backed by long-term contracts to sell electricity to a utility, thereby generating a rate of return. You can think of one of these projects like buying a bond with a known upfront cost and predictable cash flows for 20-30 years.

Over time, more debt was issued, and more equity was sold, creating a flywheel of growth that increased the dividend. This is because, if the cost of capital is low enough, each project is accretive to the business and ultimately to SunEdison.

In a very simple sense, the finances look like this example. I have laid out a project that costs $1 million and generates $162,745 in revenue for ten years (end of life is 10 years in this example), which equates to a 10% rate of return.

If this project is financed with a 50/50 combination of debt and equity, and the cost of debt is 6% and the cost of equity is 8%, or a net cost of capital of 7%, it makes sense to buy the project.

And if this project is profitable, why not buy as many projects as possible?

Any investor should sell a 7% bond to buy a 10% bond!

Seriously, sell stock. Issue debt. Buy, buy, buy as much as you can!

It’s free money!

Here’s the problem.

What happens when the cost of capital changes?

In the table below, I’ve laid out the cost of capital that could change over time. When the cost of capital is low, buying a 10% ROI project makes sense. But what happens when the cost of capital changes?

And projects are signed years in advance, but are often financed when construction is completed. There’s a timing risk in this business model, which you’ll hear about again later.

Remember, both debt and equity are publicly traded, so we can see these costs change in real time.

At a certain point, projects are no longer accretive to earnings and the dividend. And if your entire business is predicated on growth through issuing more debt and equity, you’re building a house of cards.

The projects you’ve already contracted 1, 2, 3 years out were once future dividend growth, and now they’re pulling you further underwater.

That’s exactly what SunEdison found out.

Project obligations couldn’t be met with cash from the business, and the pool of money from debt and equity investors dried up.

When a company’s stock drops low enough or when debt costs rise high enough, making big capital investments no longer makes sense.

The current AI buildout is eerily similar to SunEdison’s business model, at least for companies like CoreWeave $CRWV ( ▲ 22.64% ), Oracle $ORCL ( ▲ 6.63% ), Nebius $NBIS ( ▲ 14.56% ), IREN $IREN ( ▲ 11.51% ), and even NVIDIA $NVDA ( ▲ 3.93% ).

They’re making big capital expenditures, which will pay off via long-term contracts…as long as their cost of capital stays low.

Who cares about negative free cash flow when RPO is doubling every few months!!!

Issue debt as fast as you can if the market will allow it!

Oracle is taking the same playbook. Who cares about going from cash flow positive to cash flow negative when RPO is up and to the right?

Issue stock. Issue debt. Issue whatever someone will buy!

The music only stops when the market drops….

Or if the market gets really worried, these companies are going to default…

Now, go back and look at those remaining performance obligations and think of them more like obligations to spend money on projects that may now be underwater. Suddenly, RPO goes from a strength to a weakness.

That’s why Oracle’s stock is DOWN 25% since announcing $300 billion in added RPO in September. RPO is now a weight around Oracle’s neck.

Business models matter.

If there’s anything you take away from Asymmetric Investing, it should be those three words.

What is the business model of Coreweave, IREN, Nebius, and Oracle in AI?

They’re kind of tech companies, just like SunEdison was kind of a renewable energy company.

But their business models tell us they’re financial engineering companies.

And financial engineering works as long as confidence is high.

If confidence falters, the business model can turn upside down quickly.

Ask yourself, what happens to any of these AI companies if their cost of debt goes to 10% and the cost of debt goes to 14% tomorrow?

SunEdison found out the hard way.