Never a Dull Moment

There’s always something happening in the financial markets and yesterday was no different. I was actually sitting down to write a bit about the numerous questions I’ve received regarding what to do investment-wise related to the Trump Administration. Then the market opened, popular stocks started getting hammered, and the news flow increased dramatically. This event seemed more pressing so we’ll get to investment implications related to President Trump next week.

As you’ve likely heard, some parts of the stock market fell dramatically yesterday while other parts of the market did okay. The issue was a surge of anxiety over the weekend related to reports of a Chinese company ramping up an AI program that costs a small fraction of what the technology currently costs to produce here at home. Sound familiar? The program is known as DeepSeek and reportedly costs about 20x-40x less to produce than equivalent domestic AI models from companies like OpenAI. If that’s true, the burgeoning industry’s economics would have to be rethought and the tens of billions, even hundreds of billions of dollars of corporate investment into data centers and other AI infrastructure could be at risk. This rethinking was happening in real time yesterday and led most investors in AI stocks to indiscriminately hit the sell button.

Ultimately, big AI-related names saw major losses. Nvidia, the most popular AI stock, ended yesterday down about 16% but had been down more throughout the market session. Nvidia’s share price had grown so much lately that as of last Friday it was the largest public company in the world, with a market capitalization (total value of publicly available shares) of roughly $3.5 trillion, which is frankly a little ridiculous. With yesterday’s drop the company’s market cap shrunk to less than $3 trillion and it’s now the third largest public company. This drop was the largest ever for a company in terms of dollar value lost in a day. Yikes. But it wasn’t just Nvidia. Other AI companies suffered major losses, especially those expecting to make money from the infrastructure investments mentioned above.

Share prices for AI-related companies have been rising a lot lately and this has created risky concentrations in the Tech and Communication Services sectors within market indexes. The two sectors make up roughly 44% of the S&P 500, the typical market benchmark, well above historical norms. This helps lift overall performance when these sectors are doing well, but the opposite is also true and, unfortunately, this concentration risk was on full display yesterday.

Larger, more diversified companies in the AI mix like Microsoft and Google fared better, down from about 2% to a little over 4%, respectively. Interestingly, most of the stock market actually did fine yesterday. Within the S&P 500, more than two stocks advanced for every stock that declined. Positive performance from so many wasn’t enough to counter the AI drop, however, and the S&P ended the day down about 1.5%. The tech-heavy NASDAQ ended down by about 3% while the Dow Jones Industrial Average actually ticked up on the day. Bonds were up as well.

Okay, so is all this stuff about AI going to be our next existential crisis? Time will tell a bunch of things, including how true these reports out of China are and what it all means for an industry that’s potentially as revolutionary as the Internet. Along these lines I wanted to pass along some snippets of analysis that my research partners at Bespoke Investment group sent out late yesterday. Check this out for some quick context in a fast-moving situation.

Beyond that, it’s important to note that this is why diversification and asset allocation are important. They’re not always fun or the sexiest investment concept but practicing both consistently helps when trying to monitor and limit exposure to risky parts of the market.

From Bespoke…

… The shock to the market tied to AI over the DeepSeek model’s cost have been profound. One of the big stories of the past year has been surging investment in data centers and the systems that feed them, especially power. From a macro perspective, that underlying demand (along with ongoing investment fueled by fiscal policy over the past several years that is still playing out) has been a huge stabilizer for the economy, along with a relatively wide deficit.

One of the fears from the Fed’s tightening cycle is that interest-rate sensitive sectors would crash, with sharp drops in demand for durable goods and slowdowns in investment, especially related to housing. But as the data today shows and as we’ve seen from homebuilder reports recently, the housing construction sector looks very strong relative to recent history despite those high rates. For housing, the Fed’s tightening was a manageable blow because demographic and long-run factors including underbuilding since 2007 have created a structural supply/demand mismatch. That’s also why home prices moved so little in response to interest rates.

The AI investment craze is a similar supply-demand mismatch… hyperscalers are ploughing unheard of sums into capex as they rush to either build a dominant model or service those who are. That’s created a handoff of sorts from consumer spending (which the Fed has successfully cooled to sustainable levels) fueling the overly hot economy of 2021-2023 to investment even as rates have soared.

Of course […] all of that AI investment needs to earn a return. For much of the past two years, skeptics of the craze have focused on the demand side of the equation: is there actually willingness to pay enough to cover all the investment costs over time, as well as a profit margin for the entities doing the investing? The physical construction costs, GPUs, the power to run them, and the cooling needed have generated impressive equity returns.

But what if increased supply of models rather than not enough demand for them was the problem with the massive investments being undertaken by Oracle (ORCL), Meta (META), Google (GOOGL), Amazon (AMZN), and a host of others? Silicon Valley VCs have been focused on the possibilities of generative AI that might even lead to general AI - a true artificial intelligence that most of us would recognize from science fiction. But a world where AI models are very cheap to train and use in specific and more tailored applications would look very different. The 1990s vision of cyberspace was fueled by Robert Gibson’s Neuromancer but turned out looking more like an American mall that was ubiquitous at the time, accessed by the masses and part of everyday life but not a complete shift in our experience of the physical world.

A version of an AI future that DeepSeek points to is much more like the evolution of cloud computing or basic digital analysis tools like the ubiquitous Excel, email, or chat functions: productivity enhancers that don’t significantly displace labor.

If that’s correct and we have no edge in how the future plays out, there is still enormous profit potential in AI through dramatically higher productivity. But the actual compute necessary to support that deployment environment and the costs associated with it could make $500bn of investment in US data centers look excessive.

As with any other rapid technological advance, it’s also deflationary. Higher productivity per worker means lower labor intensity for a given level of output and ultimately less inflation…especially if the upfront costs of data center investment don’t need to be as high. It’s not hard to imagine a world in the late 2020s or early 2030s where excess compute capacity makes it cheap to deploy a wide range of AI tools which are ultimately controlled and managed by human workers who could do much better quality work much faster, driving rapid increases in standards of living amidst low inflation. But that’s probably not a world where ploughing billions into [Nvidia’s chips] and nuclear reactors [part of the energy infrastructure needed to power AI] creates a high return.

Have questions? Ask us. We can help.