The AI Bubble is Worse Than You Think

September 2025, Nvidia announces that it will invest up to $100 billion in OpenAI to help the AI startup fund a data center build-out of at least ten gigawatts, enough energy to power New York City. Only a few weeks later, OpenAI and AMD announced a strange computing deal. AMD is going to provide OpenAI with warrants to purchase up to 160 million AMD shares.

That's nearly 10% of the company at a $0. 01 per share. In exchange, OpenAI is committed to purchase six gigawatts of AMD GPUs, which represents about $90 billion in spend.

But here's the thing: $160 million of AMD shares; and you can look this up; are worth nearly $40 billion at $90 billion of revenue and 20% of margins, AMD is getting $18 billion of EBITDA back, but at a cost of $40 billion in stock. Less than one week later, they announced another deal, but this time it was with Broadcom. OpenAI and Broadcom are going to work together to develop and deploy ten gigawatts of custom AI chips over the next four years.

And you know what? The estimated deal value is $350 billion. But here's the thing.

There are more deals than just this. A deal with Oracle: $300 billion. Microsoft: $250 billion.

Amazon: $38 billion. Coreweave: $22. 4 billion.

All in all, OpenAI has committed to $1. 15 trillion of infrastructure over the next five years, prompting this viral clip: How can a company with $13 billion in revenues make $1. 4 trillion of spending commitments?

First of all, we're doing well more revenue than that. Second of all, Brad, if you want to sell your shares, I'll find you a buyer. But this is just the tip of the iceberg.

What if we were to include venture investments like those with Ambience, Harvey and Innosphere? But these venture investments were made so that these companies could actually use the money to spend more with OpenAI. Here's how it works.

OpenAI invests in startups. The startups use that money to spend with OpenAI. Based on all of this, you'd think that OpenAI is a trillion company, right?

Well, obviously no. According to Sam Altman, OpenAI will have an annualized run rate revenue of $20 billion by December 2025. Okay.

I want you to do the math here. $20 billion in revenue, yet they have $1. 1 trillion in spend commitments.

Yeah, that makes sense. So how will OpenAI pay for all of this? More importantly, Is OpenAI too big to fail?

Okay, so to get a better sense of $1. 1 trillion, we need better perspective. Annually, US companies spend around $1.

2 trillion on CapEx per year. That is every single US company combined. This means that OpenAI is committing to spend nearly 100% of US CapEx over the next five years.

But this CapEx number includes nearly every US company across industries. What if we were to look at just the companies that OpenAI's dealing with? If we look at the last 12 months of financials, Amazon, Broadcom, AMD and Nvidia, Oracle, Microsoft and Coreweave, have only combined to spend $225 billion in CapEx.

Those are giant companies. And the OpenAI deal comprises of nearly 100% of that spend. And if you're an OpenAI fanboy, which I suspect some of you are, then you'll probably just say they'll grow into that CapEx.

To which I say, let's do the math. Take a look at this chart. This shows how much OpenAI is expected to spend per year for the next five years.

The number grows from $6 billion in 2025 to $295 billion in 2030. If we take OpenAI's gross margin projections at face value, that means they will need to grow actual revenue, wait for it from $12 billion in 2025 to $983 billion in 2030. Let me illustrate how crazy this sounds.

OpenAI essentially needs to grow revenue by 85X in five years to have even the remotest chance of making the bills. But how many companies even have $983 billion? Well, the answer is zero.

Yep. OpenAI just needs to become the most profitable and the biggest company on earth to meet its obligations. But this doesn't convey the extent of the problem.

As many of you know, gross profit and profit are not the same thing. These numbers don't assume any operating costs. Any first year accounting student can tell you they don't contemplate energy costs, maintenance costs, leases, debt servicing costs, and more.

And although this is not impossible, the set up is insane. And a crazy clip came out with the OpenAI CFO, where he essentially said that they might ask for a government backstop. And so this is where we're looking for an ecosystem of banks, private equity, maybe even, governmental, like the ways governments can come to bear.

Yeah, you heard that right. A federal backstop. That way, if OpenAI can't meet those obligations, the taxpayers will be on the hook.

But this is bigger than OpenAI. There are hundreds of AI companies. AI for dating, AI for voice, AI for video.

AI for AI for example. Anthropic has raised $27 billion USD and it's currently valued at $183 billion. They've also committed to purchase $30 billion of Azure compute capacity from Microsoft, and possibly $14 billion with Amazon.

What about Meta? Firstly, Meta bought Scale AI for 14. 9 billion, which was a joke.

Let's list off everything that they're planning to do. They're projected to spend $70 billion on data centers. They went on a hiring spree and are spending $1 billion in wages for the top AI talent.

And there's already reports that people either turned down those pay packages or are leaving to go back to their previous jobs or starting their own companies. We could go on forever, but the point is, the spending is just insane. To make all of this math work.

Bain estimates that $2 trillion of new revenue is needed. That is 7% of US GDP. But the sheer numbers aren't the whole picture.

This is just the nature of these deals. Real quick. Over 96% of you aren't subscribed to the channel, and I know that it takes a lot to hit that subscribe button, but please do.

It helps the channel a lot and we'd really appreciate it. Now back to the video. And although Sam Altman and Jensen Huang aren't finance nerds by background, you'd be surprised to learn that their recent deals would even make the shadiest of financers proud.

So what do I mean by this? Well, the spending commitments being made by OpenAI and others are all interconnected and super complicated. Let's give an example with pretty graphics to explain.

Microsoft has invested over $13 billion into OpenAI's for-profit arm for equity and a revenue share, but much of this $13 billion is not cash, but instead it is Azure Cloud Computing credits, meaning Microsoft is investing in OpenAI so that OpenAI can buy Microsoft products. But it gets better because this deal means OpenAI is using a ton of Azure capacity. Microsoft needs to expand and improve its Azure data centers.

So Microsoft takes this book revenue from OpenAI and places massive orders with Nvidia for millions of GPUs so that it can handle the commitments to OpenAI. So essentially, Microsoft is investing in OpenAI. So OpenAI can buy from Microsoft.

So Microsoft can buy from Nvidia. And now that Nvidia has a bunch of money coming in, a new loop starts to emerge. Nvidia takes their money from Microsoft and others and makes a commitment, you're not going to believe this, to invest up to $100 billion in OpenAI.

OpenAI then commits to using that investment to purchase a huge volume of Nvidia chips. And with that leftover money, OpenAI can pay off others like Microsoft. So Nvidia pays OpenAI, so OpenAI can pay Nvidia and also pay Microsoft so that Microsoft can invest in OpenAI so that OpenAI can buy it from Microsoft, so that Microsoft can buy from Nvidia, so that Nvidia can invest in OpenAI and so on and so on and so on.

There are dozens of investment structures just like this. And the problem is, these ouroboros- like dealings, are impossible to track and measure because the revenue of one company is the cost of another company is the investment of another company. This money flows from entity to entity, making it difficult to understand what is real, what is news and what is garbage.

Layer on the fact that many of these deals have a ton of complex contingencies and financial instruments, then it becomes impossible to assess the quality of cash flows and that, in our view, is a problem. How do you value these companies when it's unclear what's real and what's fake? And if all these companies are interconnected, what if one of them fails?

The question on everyone's mind is: what will the ripple effect be? Many people are concerned that the AI bubble looks a lot like the 2008 great financial crisis and the 90s tech bubble. But is it true?

All right. Firstly, let's look at the great financial crisis. The GFC led to nearly 9 million people losing their jobs and household net worth falling by $11 trillion.

The cause was subprime and alt-A mortgages. These low quality mortgages represent approximately 20% of all US mortgages, or $2. 2 trillion.

And this is when the house of cards began to collapse. Over $500 billion of those loans began to default, which led to a systemic loss of over $10 trillion. But there lies the question how did $2.

2 trillion worth of loans, of which only less than half defaulted, lead to over $10 trillion in losses? Securitization, leverage and derivatives. These types of tactics increase market exposure to these crappy garbage assets by nearly 15 times.

Which is why we're saying the AI situation is slightly different. OpenAI's obligations are not being securitized and spun off into exotic derivatives, multiplying that global exposure. What's more is that unlike many of the institutions in the GFC, the big tech players have a lot of cash and very little debt.

So if something were to go wrong, they have the cash flow and balance sheet to handle it. But what about the tech bubble? Well, this one is a bit more interesting.

The dotcom bubble was brewing for years and everyone got excited by the prospect of the internet. You could finally communicate with people all over the world, you could buy stuff online, etc. etc.

From 1995 to its peak in March 2000. The Nasdaq surged by 572%. But in ‘99, the fed started raising interest rates and investors started asking about profits, only to find none.

Panic set in, and by October 2002, $5 trillion in market value was wiped out. During this time, the Nasdaq 100 was trading at a forward P/E ratio of 60x. Today, that number is 30x, which is 50% cheaper than before.

And not to mention that the internet bubble busting didn't destroy every single company. Think about Apple and Amazon. There were companies that persisted and survived throughout.

The bad businesses failed, and the good ones, they survive. What's more is that very few of these companies were profitable. Whereas today, all the major tech companies are super profitable.

That said, yes, the AI bubble doesn't have the debt and derivatives of the GFC. It doesn't have the same froth of the dotcom bubble, but it does have trillions of inter-company dealings, questionable earnings, relatively high valuations and an underlying economy that cannot sustain low interest rates. And way too much AI slop, if we're being honest.

As the great Charlie Munger famously said, if you mix raisins with turds, well, you still got turds. There are some raisins in here. Great companies with strong cash flows and real business models.

But around all those lovely raisins, there are a ton of less than lovely turds. Inter-company deals that create fake revenue and cash flow. Crazy amounts of spend that would require OpenAI to become the biggest company in human history.

Discussions of government backstops to reduce the risk. Turd, turd, turd, and raisins still equals turd. But I guess some people like that.

Thank you for watching. Subscribe to 2&20 and check out our free discord in the description where we actually hang out in.