February 1, 2025
DeepSeek or Deep Fake?
In late January, the news that a Chinese company formed in 2023 had trained a highly competent AI model for $5 million sent data center infrastructure companies plummeting. Technology luminaries jumped on their X accounts to proclaim how robust this model appeared to be and how earth-shattering its development is to the future path of AI.
And while the companies selling the chips, cooling systems, and networking gear to those data centers inched upward in the days following the carnage, they remain well below highs set in mid-January. And so, the question remains, is the DeepSeek model a stark turning point in the march toward ubiquitous AI? Is it a mere curve in the road on the existing path? Or, is it closer to a deep fake than a DeepSeek of the truth?
Let’s tease out these three possibilities.
First of all, cutting edge technology companies such as Meta, Google, and OpenAI currently spend billions upon billions to build, train, and maintain AI models. The claim that this startup achieved all that using cheaper chips, piggy-backing techniques, and short timeframes ought to be viewed with a healthy dose of skepticism. Additionally, DeepSeek comes from a portion of the world currently locked in a philosophical and economic war with the US. A careful eye could see reasons they might want to insert a bit of chaos into our stock markets. If this take is closer to the truth, then advanced chips from Nvidia, efficient cooling systems from Vertiv, and networking gear and software from Arista will remain in high demand at least until data centers are at capacity or a true revolutionary method emerges.
The second possibility is sort of an in-between interpretation. Perhaps they did spend considerably less money and time building their model by using existing open-source AI models (e.g. Meta’s Llama). In this case, those building the vanguard models like Meta or Google still demand high-end chips and other gear. However, a very solid second tier of AI related companies will be able to operate using the open-source models on which to build themselves. It is less clear in this case if infrastructure demand would decrease slightly or remain the same as use cases for cheaper AI proliferate.
The third possibility, and least likely in our estimation, is that everything in the recent news regarding DeepSeek is true. They built and trained this model for what amounts to pennies in today’s corporate world. The model resulting from these inexpensive efforts is actually comparative to the current best available. IF this is the case, then yes, the need for infrastructure gear and for data center land and for electricity to power it all will not grow nearly as much as forecasted just a few weeks ago. HOWEVER, this result is actually the best result for the entire planet (ecologically AND economically). Instead of hammering countless trillions into building this infrastructure further, ALL companies would now have access to the best AI for literal pennies. Construction companies, financial companies, and manufacturers would have more and more use cases that fit into this new inexpensive model. The only companies that would “suffer” due to less growth than expected would be those directly involved in the build out of these massive data centers. The net result is more efficient, more profitable businesses the world over.
And so, we hold these stocks. Yes, we are perhaps a little more suspect of the overly rosy growth projections for the infrastructure companies. But, then again, we were always skeptical of trees growing infinitely to the heavens.
As unsettling as AI is to us and many others, it will progress and be used. Many companies (some not yet formed) will benefit from the technology’s advancement. We don’t know beforehand which businesses will be at the forefront, or which will lag. This is why we don’t own just one company, selling one type of product. We build concentrated, yet diversified portfolios with the strength to grow in the sun and the flexibility to bend in the storms like the one that blew in late January.
Stirling Bridge Wealth Partners, LLC is fortunate to count many of you as clients. In the good times and bad, we remain committed to providing customized investment solutions and robust financial planning wrapped in a package of exceptional service. We thank each of you for your dedication to us and for your trust.
Sincerely
Jason Born, CFA
President