Chief Investment Strategist, The Freeport Society
There was a collective jaw drop on Monday.
Chinese AI upstart DeepSeek turned the tech world upside down with the release of its vastly cheaper competitor to ChatGPT, DeepSeek-R1.
It cost ChatGPT creator OpenAI about $80 million to build and train its model.
It cost DeepSeek about $6 million.
The Chinese model was about 13 times cheaper to create than its U.S. rival.
Ouch.
Yesterday, we looked at what this means for OpenAI, Microsoft, Meta, Google, Elon Musk’s xAI, and other major U.S. companies that’ve been plowing hundreds of millions of dollars into building their AI models.
We also looked at what it means for Nvidia, the world’s leading chipmaker.
ChatGPT and other U.S. models use Nvidia’s state-of-the-art AI chips. But DeepSeek has managed to get the same level of performance using the less powerful chips Nvidia makes to get around export restrictions to China.
Today, let’s zoom out and look at the bigger picture.
DeepSeek isn’t just a stock market story. It also exposes the shortcomings of trade protectionism and the unintended consequences that follow.
It’s also a consumer story.
Ultimately, it means better, cheaper AI is coming your way.
Hats Off to Liang Wenfeng
I’ve started a handful of businesses throughout my career. Some were successful enough to feed my family for a few years. Others failed.
But they all had one thing in common: They were financed by sweat equity.
I never had an angel investor or private equity backing. I never had a rich uncle… or Wall Street buddies to give me a loan. And unlike OpenAI, I didn’t have a mega-cap tech company in the form of Microsoft writing me unlimited checks.
I was on my own.
I started every business on a shoestring budget – the first one, a small financial publisher, with a laptop held together by electrical tape and a copy of Microsoft Excel that, ahem, may or may not have been pirated at the time.
With a ridiculously humble setup, I eked out a profit because failure simply wasn’t an option. I had little mouths to feed.
My lack of funding meant I had to get creative. I couldn’t afford legal work. So, I bluffed my way through documents and filings.
I couldn’t afford a tech guy. So, I had to figure out how to run IT on my own.
That’s why I can identify with DeepSeek.
It didn’t build a world-changing AI model on founder Liang Wenfeng’s kitchen table. He was already a wealthy hedge fund manager before he launched the company.
But if these numbers are accurate, Mr. Liang managed to build DeepSeek’s latest model for 95% less than what OpenAI spent on ChatGPT.
And he did it in a way that uses less energy and less computational power in its day-to-day operations.
He also did it without the state-of-the-art Nvidia chips his American competitors used.
The Trump and Biden administrations had policies in place to limit tech exports to China. So, Nvidia could export only its lower-performance chips to China.
So, hats off to Mr. Liang and DeepSeek. From one scrappy hustler to another, well done!
But why is this important?
Protectionism = Decay
Freeport friend and legendary newsletter publisher Bill Bonner put this in perspective in yesterday’s issue of his excellent e-letter over at Bonner Private Research…
DeepSeek is the product of $6 million, not billion (in rented GPU hours), of investment by Chinese entrepreneur, Liang Wenfeng. Despite spending $200 billion a year to dominate the AI space, the four leading U.S. tech giants — Microsoft, Amazon, Meta, and Alphabet — have let Liang steal a march on them.
Typically, a bubble attracts its own pin. By trying to contain Chinese innovation, the Trump/Biden administrations seem to have sharpened its point. And now, it heads straight for those tech oligarchs standing in the second row during Trump’s inauguration…
Protectionism and containment lead to decay and backwardness, not to growth and progress. Despite U.S. efforts to stop them, China’s chip exports have gone up for 14 consecutive months. Forced to experiment, they developed new tactics and new technology.
I’m no fan of the Chinese Communist Party (CCP). In the AI war, I’m rooting for Team America.
I don’t trust the U.S. government or any of the large Silicon Valley tech companies. But they’re models of decency compared with the CCP.
And the CCP exercises a lot of control over what DeepSeek’s models are allowed to say. (Don’t ask them about the Tiananmen Square massacre, for instance, or they’ll give you the cold shoulder.)
But I can still appreciate the grit, hustle, and chutzpah of an underdog taking on the big boys. And the success of DeepSeek’s latest model is all the more impressive given that the coders have the CCP constantly breathing down their necks.
Why DeepSeek Means More AI Is Coming Your Way
DeepSeek’s innovations will force OpenAI, Microsoft, Meta, Google, xAI, and other major U.S. AI companies to up their game.
The result?
Better quality AI apps at a better price for all of us.
The biggest single barrier to further AI advancement was the mammoth costs involved in building, training, and running these models. As I wrote in these pages in December…
Training [ChatGPT’s] GPT-5 – the company’s upcoming AI model – is expected to cost as much as $2.5 billion. That’s 17.5 times more than the cost to train GPT-4. It’s nearly 400 times more expensive than the cost to train GPT-3…
Amazon plans to invest more than $100 billion in AI-capable data centers over the next decade…
Meanwhile, Mark Zuckerberg’s Meta has budgeted between $35 billion and $40 billion for AI infrastructure in 2024… with tens of billions more on tap.
Microsoft and Google parent company Alphabet are each expected to spend in the ballpark of $50 billion on AI in 2024… and even more in 2025.
Those numbers weren’t sustainable. Something needed to change. DeepSeek’s innovation just made that happen.
All of these companies – as well as countless others – will continue to pour cash into AI development. But they’re now going to get more bang for their buck.
And that’s good news for all of us.
Oh, and does this mean that artificial general intelligence (AGI) – AI that matches human cognition and can transfer reasoning and expertise across domains – is closer to reality?
Yes it does.
So, get ready…
To life, liberty, and the pursuit of wealth,
Charles Sizemore
Chief Investment Strategist, The Freeport Society
P.S. As January comes to an end, InvestorPlace’s Luke Lango is gearing up his February Auspex portfolio.
Luke’s screener is built to help find the strongest stocks at the beginning of every month by scanning 8,000 stocks for 15 different fundamental, technical, and sentiment factors.
His strategy could help you beat the markets every month in 2025 and make you as much as 100% faster than you imagine.
And it’s already proven its ability to help him find winners like:
- 38% in a month with AnaptysBio.
- 115% in under 60 days from Aveanna Healthcare.
And even though the release of DeepSeek threw markets into turmoil, Luke’s picks for the month are holding their own.
To learn about the brand-new batch of stock recommendations Luke is going to release on February 3, simply click here.