Concerns about artificial intelligence (AI) stocks and their monster valuations aren't exactly new, but the market is now being less coy about it. Following a big multiyear run for the sector, it appears that some investors are setting a high bar for AI stocks. No longer are announcements about higher capital expenditures for AI infrastructure build-out purely driving stocks higher, as investors start to question the sustainability of it all. From Oracle (ORCL 0.25%) to Broadcom (AVGO 0.34%), the concerns about AI stocks are starting to pile up.
The returns seem less achievable than perhaps they once were
Concerns over AI stocks have been mounting, including high valuations, excessive capital expenditures, and whether the U.S. and the world truly have the resources to support all of the AI demand, considering the need for data centers, the power needed to run these data centers, and other necessary resources, such as water to provide the liquid cooling solutions for AI chips.
Image source: Getty Images.
Oracle delivered arguably the best quarterly earnings report of the year in September. The company reported over $450 billion of remaining performance obligations, due to its fast-growing AI cloud services business, which rents out data centers with clusters of graphics processing units (GPUs) to companies looking to run AI applications. Oracle has experienced strong demand from hyperscalers like Microsoft and OpenAI. The stock blasted 40% following this report.
But in recent months, shares have retraced. Media outlets in November reported that Oracle would need to raise $38 billion in debt to build the data centers necessary to meet demand. The Information also reported Oracle's AI data center business, while experiencing intense growth, is delivering a profit margin between 10% and 20%, not as high as investors might have hoped.
In Oracle's most recent earnings report for the second fiscal quarter of 2026, ended Nov. 30, the company's revenue of nearly $16.1 billion fell just shy of consensus estimates, and the stock declined. Oracle also raised its full-year capital expenditures guidance from $35 billion to $50 billion and reported negative $10 billion of free cash flow for the quarter. Following the earnings report, the price of five-year credit default swaps on Oracle's debt, which is essentially insurance that investors can buy to protect against future default, hit a record high.
Things didn't get any better for Oracle after chipmaker Broadcom reported earnings for the fourth quarter of its fiscal year 2025, ended Nov. 2. Earnings actually came in strong, with Broadcom reporting numbers above consensus estimates and solid forward guidance. However, investors once again came away feeling disappointed after Broadcom guided for its gross margin to decline next quarter. The market also felt that the company's AI product backlog was understated.
NYSE: ORCL
Key Data Points
The big question has become whether or not the hundreds of billions large companies have already spent on AI-related capex -- and the trillions they are collectively projected to spend in the coming years -- actually makes sense from a returns perspective.
The math doesn't make sense, according to IBM CEO Arvind Krishna, who said that there is "no way" the hyperscalers will see returns on all of this spending on data centers. Krishna said that roughly $80 billion is currently needed to build a 1-gigawatt data center. "Okay, that's today's number. So, if you are going to commit 20 to 30 gigawatts, that's one company, that's $1.5 trillion of capex," he said.
Perhaps it's time to slow down
I don't think investors doubt that AI can change the world, but the path is now murky. There may be limits in terms of the resources required, and committing all these resources, while taking on debt to do so, may not yield as strong returns as investors initially anticipated, at least on the desired timeline.
As most are aware, the internet boom first had to endure the dot-com bubble to reach its current state. Now's a good time for investors to review their AI holdings and carefully examine valuations and the assumptions on which they are based. If they seem too good to be true, they probably are, and it's time to think about trimming some gains.
