AI Gets Crushed

So you’re telling me that a Chinese hedge fund only needed 2 months to create an AI model that doesn’t use Nvidia’s best chips, only costs $5 million, and is every bit as good as the AI offerings from Microsoft, Google, and OpenAI…??

And you’re also telling me this $5 million AI model is responsible for a $1 trillion stock price wipeout for the Nasdaq?

I don’t buy it. If I believed a Chinese hedge fund could build a world-beating AI model for $5 million, then I’d probably think The Professor really built a radio out of a coconut and one of Gilligan’s shoelaces.

There are some shenanigans going on here.

The U.S. went to some pretty extraordinary lengths to keep Nvidia’s top-performing chips out of China’s hands. And Nvidia has beaten earnings and revenue estimates for two years running…

So what’s more likely:

This AI model (called DeepSeek) is exactly what THE CHINESE say it is, or
There’s a massive backdoor for supposedly “banned” US technology into China

I’ll take #2 for the win, Chuck.

Here’s a little nugget from last Friday’s New York Times pre-market newsletter:

Alexandr Wang of the A.I. training giant Scale AI…[said] at the World Economic Forum that Chinese companies had way more high-end chips than U.S. controls allowed. DeepSeek, he said, probably has about 50,000 Nvidia advanced H100 processors, “which they obviously can’t talk about.”

Funny we haven’t heard much from this guy so far today…

Seriously, though, would it really come as a shock that China has plenty of Nvidia chips? Would it come as a shock that China’s government is helping offset costs for DeepSeek?

And, all the DeepSeek press is coming from a HEDGE FUND! Would it come as a shock that this fund had massive downside positions on Nvidia before it started releasing all this information?

The Valuation Problem

The flip side of all this is that It was bound to happen. Something was bound to come along and poke a big hole in the AI investment thesis.

It is inevitable that any emerging technology trend will suffer from over-investment. That’s true for individual investors and the companies themselves. Once you start pulling on a thread, the whole thing starts to unravel.

In my Predictions for 2025 I wrote:

Investor sentiment is what drives P/E ratios (how much investors are willing to pay for earnings). Coming into 2024, I was convinced that investors were underestimating the power of the Mag 7, that investor sentiment would improve and that P/E ratios would move higher.

The opposite is true for 2025.

2024 was a year of nonstop upside surprises. And valuation metrics like P/E ratios will always rise when the story (and the earnings) keeps getting better. It’s a confidence game – confidence that earnings will keep beating expectations.

Now, here we are: not even a full month into 2025, and confidence is being sorely tested. Even if (when?) the truth behind DeepSeek is revealed, a little more skepticism from investors is likely. P/E ratios have likely peaked for the foreseeable future.

It’s still worth asking what happens when Microsoft (MSFT) and Meta (META) release earnings reports that beat expectations later this week – which they will…

Winners and Losers

Facebook parent company Meta said it will spend $65 billion on AI this year, and Microsoft said it will spend $80 billion…

Those are astronomical numbers. But what if they aren’t necessary? What if DeepSeek really has found a better way – a way that yes, still uses Nvidia’s top H100 processers, just not as many?

Well, that would obviously be bad for Nvidia and AMD and the other chip companies. But it is very good for Meta, Microsoft, and Google because it suggests they could get the same bang for less buck. It’s not like Microsoft will be dropping the price for its AI-enhanced Office software…

And if there’s any indication that the hyperscalers (Microsoft, Meta, Google, and Amazon) can scale back CAPEX, those savings go right to the bottom line.

But here’s the thing: American companies are not going to use AI that is hosted by Chinese companies. They will use American companies. They will use Google Cloud, Microsoft Azure, and Amazon Web Services, powered by Nvidia chips.

These data centers still have electricity demands.

It may be a rough road for small modular nuclear reactor (SMR) stocks, like NuScale and Oklo. The actual deployment of these companies’ nuclear tech is still years away. That’s a long time to wait, and I find it easy to imagine that these stocks will not be able to hold their current valuations while investors wait.

But there’s one electricity source that won’t take years to deploy: natural gas. Natural gas is firing turbines that make electricity for AI data centers right now. And there are over 200 new natural gas plants are being built in America.

Natural gas is the only no-brainer AI investment you can make today. Here’s what to do.

Cheers,

Briton Ryle
Chief Investment Strategist
Outsider Club

X/Twitter: https://twitter.com/BritonRyle

Latest Articles