Way Up at the Top of This Wobbly Market Is One, Huge AI Trade
Let’s look at the concentration risks investors shouldn’t be ignoring.
You've reached your free article limit
You've read 0 of 1 free Pro articles.
The major indexes were up between 2% and 2.7% in the third quarter. But 200% of the gains in the S&P 500 came from just four stocks: Apple (AAPL), Microsoft (MSFT), Meta Platforms (META) and Nvidia Corp. (NVDA). The “S&P 496” was down for the quarter. Such market concentration is quite unusual and feels like a potential sign of a top in the overall market.
The majority of the stocks in the S&P 500 are down at least 20% from their 52-week highs. This means most investors that did not have exposure to the four tech behemoths listed in the opening paragraph, probably posted a loss in their portfolio during the third quarter. This is simply one of the most top-heavy markets that I can recall in four decades of investing.
The AI trade is not only the straw that is stirring the drink but has become one of the only games in town. Without the massive amount of investment going toward the largest infrastructure buildout in U.S. history, the economy would likely be flat-lining at best. And I felt much better around this effort when it was being funded out of the cash on the balance sheets of the hyperscalers as well as their free cash flows
Those times, however, have been in the rear view mirror for many quarters now. The five major hyperscalers have seen their free cash flow plummet and go negative in many cases. The buildout is now funded in conjunction with a very large amount of equity and debt issuance. And unfortunately, not always in a straightforward way. The five hyperscalers have amassed a combined $3 trillion in off balance sheet liabilities in addition to the over $1.3 trillion in debt on their balance sheets.
And more of the AI ecosystem is using what I would call sketchy accounting maneuvers and is relying on a huge amount of circular financing. Depreciation schedules that were mostly kept at three years for chips in decades past have been pushed out to six years for graphics processing units. Nvidia is helping to backstop over $100 billion for a massive AI data center complex in Ohio being constructed for OpenAI. Broadcom (AVGO) has greatly expanded its presence in what used to be called vendor financing back during the internet boom. Amazon (AMZN) is now trying to wrap up and unload some $8 billion of Nvidia GPUs to investors via a special purpose vehicle, or SPV.
And this huge AI bet relies on two things that make this investor nervous. First, can the electrical grid provide the electricity needed to operate the huge data centers going up throughout the country on a timely schedule. That could be a topic for an entire column. The bigger question is whether OpenAI and Anthropic live up to their commitments. OpenAI plans to spend at least $600 billion and up to $750 billion through 2030. Anthropic’s S1 listed $518 billion in commitments. Both companies are bleeding cash, need to raise massive amounts of new capital, and are seeing much cheaper Chinese and other AI models take token utilization market share after both companies moved to usage based pricing this spring.
Anthropic is trying to execute a huge IPO in November, valuing it at $2 trillion. To me, that seems a tough ask, given the challenges listed above. This is especially true as Anthropic is listing a company risk of potentially unleashing AI that could extinguish humanity. Maybe I am just nitpicking.
That said, a market at these valuations, extreme concentration levels, and largely dependent on two huge AI labs that are bleeding cash with questionable moats is making this investor increasingly anxious. As such, I have recently taken my allocation in short-term Treasuries and cash up to 35% from 25% for most of 2026 within my portfolio.
At the time of publication, Jensen was long AMZN
