AI Is a Bubble. And AI Is Real.

Is AI a bubble?

No. And yes.

AI can be a genuine technological revolution and a financial bubble at exactly the same time.

BOTH can be true.

But that doesn’t necessarily mean it’s going to pop. Maybe. This is not financial advice and if it was it would be terrible anywauy.

Nvidia's new $105 billion OpenAI guarantee is the cleanest example I've seen yet of the weirdness of the AI market. The headline version says Nvidia gave OpenAI $105 billion to buy Nvidia chips. But the reality is much stranger.

Here's the full video breakdown:

Nvidia did not give OpenAI $105 billion

Discussed at [00:00:52]

Nvidia and OpenAI are working on a new AI datacenter together. With Nvidia supposedly guaranteeing $105bn worth of the lease.

It’s a weird setup though…

Imagine your landlord refuses to rent you a house unless somebody co-signs the lease.

Normally it’s your parents.

Fairly normal. Get someone more financial stable to guarantee someone less stable. OK - so far so good.

Except in this case your guarantor is IKEA and you have promised to fill the entire house with IKEA furniture. Billy bookcases on every wall. Blue bags stuffed into every nook and cranny.

That is roughly what just happened.

OpenAI signed a 20-year lease for an enormous 8 gigawatt data centre in Ohio. SB Energy, backed by SoftBank, will build, own and operate it.

For the first 4.25 gigawatts, Nvidia is providing credit support capped at $105 billion. They are guaranteeing OpenAI’s lease payments.

AND the datacentre will be filled with Nvidia chips. The eventual plan is eight gigawatts of AI capacity, filled exclusively with Nvidia compute.

Importantly it is a guarantee. NOT a payment to OpenAI. If OpenAI pays its rent, Nvidia does not hand over a penny. If OpenAI becomes insolvent or stops paying, Nvidia covers the gap between an agreed minimum value and whatever SB Energy can recover by finding another tenant or selling the site.

Interestingly, Nvidia can also take over the lease itself. OpenAI has agreed to reimburse Nvidia for money it actually pays.

Confusing? Very…

Follow the money

Discussed at [00:06:18]

There are three companies in the main loop.

SB Energy builds and owns the land, power and data-centre shell. OpenAI leases the capacity for 20 years. Nvidia provides the compute and helps guarantee the site's value.

Nvidia's backing makes it easier for SB Energy to raise the money to build. OpenAI gets the capacity it says it needs. The completed site buys and hosts Nvidia systems. Nvidia earns the hardware revenue.

So yeah...the supplier is helping finance the facility its customer will use to buy the supplier's product.

That’s the crux right there.

And that is circular. Ruh roh.

But a circle on a diagram is not automatically a scam. One factor decides whether this becomes a productive loop or a death spiral: whether customer pay OpenAI for useful products!

OpenAI says it will fund the lease through revenue and cash flow. Oh AND money raised from investors. Those are VERY different things.

The healthy version is millions of people and businesses paying for ChatGPT, Codex and the API because the products save time or make money. That’s a good healthy business.

The fragile version is: "Please sir, just $20 billion more bro and we'll reach AGI." Reliance on outside investment.

Sure. Investor money can bridge the gap while a market grows. That’s what it is for. But it cannot replace customers forever.

Jensen's defence is better than it sounds

Discussed at [00:13:08]

Jensen Huang answered the criticism directly: "Is this circular financing? No. OpenAI will pay the lease."

That line is doing a LOT of work. But his wider argument is pretty strong. It’s worth reading in detail. It also tells us a lot about where Nvidia is headed in the future.

Nvidia says the scarce resource is becoming land with power, permits, cooling, networking and a finished building. "Land, power and shell."

It is NOT chips. Their bread and butter. It’s bigger than that. They are moving into infrastructure. Datacentres, data transfer. Maybe even power generation down the line? Wider than just the GPUs.

This is a huge vote of confidence for the future demand of AI. They are betting (big) that demand will go up and up and up. That is not certain but do remember that Nvidia has better visibility of AI compute demand than nearly anybody alive. It supplies basically every serious Western AI builder - the labs, the cloud compute centres…everyone.

If demand keeps growing, Nvidia may never pay anything under the guarantee. It has simply used its balance sheet to unlock a bottleneck, get a huge new data centre built and sell a hell of a lot more chips.

Smart deal.

If the forecasts are right.

What could break?

Discussed at [00:19:35]

First, maybe OpenAI cannot support the lease.

OpenAI is betting that usage and revenue will grow into gigantic infrastructure commitments. But Chinese models are already pushing prices down, which is why I called open source AI a price war.

Local models are also going to be an issue in a few years. When we have Claude Fable-level local models running on our own devices in a few years, do we still need expensive subscriptions and mountains of cloud compute?

Maybe yes maybe no.

Second, another tenant may only want the site at a lower price. If OpenAI cannot afford it because the whole market has softened, Anthropic (or anyone else) is unlikely to wander in and cheerfully pay the old rate. They’ll get it at fire sale prices.

Third, the building may last while the equipment inside ages fast. Cooling, networking and power designed around today's systems may need expensive changes four years from now (when construction is complete). 4 years is an eternity in AI.

And fourth...they are all exposed to the same cycle… This is the biggy.

Nvidia invests in customers and developers. Those companies buy Nvidia systems. Lenders treat Nvidia's support as confidence. If AI spending slows, several assumptions can fail at once.

That is the bit I find uncomfortable. Nvidia is no longer simply selling chips into independent demand. It is helping create the financing conditions that allow some of that demand to exist.

Too big to fail all over again.

The internet won. Lots of investors lost.

Discussed at [00:24:30]

The dot-com comparison is normally used in a painfully simple way.

"The internet survived, therefore all the spending was justified."

Nope. The internet was a genuine world-changing technology for sure. Fibre networks were genuinely useful. Demand eventually became enormous.

The internet is kinda a big deal.

BUT plenty of telecom builders, dot-com companies and investors still lost everything. National economies collapsed. The world was thrown into recession for years.

Companies laid huge amounts of capacity based on aggressive forecasts. Vendor financing helped customers buy equipment. Valuations assumed growth would arrive quickly.

Sounds familiar?

Then the market reset. Companies failed. Assets were sold cheaply. And it all came crumbling down.

Sure…the technology survived. Companies like Google and Amazon rose from the ashes and reshaped modern society.

The technology won. The financing did not always win.

Same with AI. Potentially.

We could see a financial reset. Yet we’ll just keep using AI. It’s too useful. .

If the AI bubble bursts on Thursday, I'll still be using useful AI tools on Friday.

To the Task,

Kyle