As AI investments grow, focus on solid portfolio fundamentals
The AI bubble is on the mind of markets and investors nowadays, and the debate about the economic sustainability of the AI buildout is taking up a lot of attention with the press and pundits. To keep this in perspective, research from the Capital Group asserts spending on AI infrastructure to be about 2% of all economic activity (GDP) in the U.S. in 2026, eventually rising to as much as 3.6% over the next five years.
Two percent may not seem like a huge number, but the research indicates the underlying dollars to this number to be a whopping $650 billion in 2026. The only historical precedent for investment at this level is the building of our nation’s railroad system in the late 1800s, which at its peak comprised about 2.2% of GDP. We are living through one of the great capital investment expansions in all of U.S. history.
So, what is the AI infrastructure buildout? Well, said simply it is mostly the dreaded data centers, including the computer equipment inside, occupying an inordinate amount of public debate. It also includes the energy infrastructure (power generation and transmission lines) required to run this technology.
One of the interesting components of this discussion to me, is the fledgling level of the current AI technology in the minds and lives of most Americans. We tend to see AI as chatbots and clunky “agents” who attempt to book flights and hotel rooms, but these high-profile day-to-day uses of the technology are literally just surface level utilization of what is already occurring, and especially what is coming.
I’ve become convinced this technology will end up being nothing short of transformative in ways not even anticipated at this time, eventually merging with the physical world in ways not even yet imagined. When I look at this technology in totality, it is clearly “what comes next” economically and socially, but does that mean all the hundreds of billions of dollars going into the infrastructure it requires is sustainable?
Well, this question is a different story. The current economics (revenues and costs) of AI are not yet completely clear, even to industry experts. Data centers are absurdly costly to build and run, and the only functional publicly available business model on a hyper-scaled data center is SpaceX’s Colossus complex based near Memphis.
According to analysts’ estimates, the SpaceX center will ultimately require investment of about $50 billion, not including the power components, and $90 billion when power, water and ancillary costs are included. This is of course an absurd amount of capital investment, but what does the revenue side of the page look like?
Well, this is when the numbers get really obscene. According to public filings from SpaceX, three large customers currently lease compute (the currency of AI), generating revenue of roughly $2.1 billion per month, or $26 billion a year. It doesn’t take an MBA to do this simple math, depending on the ongoing cost structure of the center, this is a three-to-five-year payback on capital in ground. Now this conversation is starting to make sense. Huge numbers, generating huge numbers, and the technology is still in its infancy.
Bottom line for me, it’s hard to call a bubble when the economics of the conversation are still not completely understood. Yes, most of this extreme investment level is being financed by bonds and debt, yes, interest rates are rising for a multitude of reasons, making financing for these centers more expensive to service, and yes, public resistance to these facilities is for the time being rising considerably. The numbers, however, are just too huge, the business models are too new, and most investors (myself included) simply don’t have the acumen to make the “bubble” call.
In a situation like this, what is the typical investor to do? Well, when the road ahead becomes unclear, the smart thing to do is slow down and revert to cautious decision making. Most investors in AI infrastructure stocks over the past few years, mainly semi-conductors, and other computer equipment, have made outsized gains.
The index tracking semiconductor stocks is up nearly 60% in 2026 and was up 42% in 2025. Can we all agree this is not typical stock market performance? After returns like this, these investments are probably overly concentrated in most investor portfolios; they are also likely the most adored positions most of us own, and prudence requires re-evaluation.
So, I’m not going to tell any reader to sell their AI infrastructure stocks. What I am going to do is remind us all, me included, that diversification and rebalancing continue to be prudent risk management tools, and owning overly concentrated positions in any one holding or sector often does not end well. Instead of trying to call a bubble or a top in this sector, instead focus on solid portfolio fundamentals and make decisions accordingly.
The opinions voiced in this material are for general information only and are not intended to provide specific advice or recommendations for any individual. Stock investing includes risks, including fluctuating prices and loss of principal. No investment strategy can guarantee a profit or preserve against loss. Past performance is not a guarantee of future results. This material may contain forward looking statements; there are no guarantees that these outcomes will come to pass.
Marc Ruiz is a wealth advisor and partner with Oak Partners and registered representative of LPL Financial. Contact Marc at marc.ruiz@oakpartners.com. Securities offered through LPL Financial, member FINRA/SIPC.
