AI Spending Is Reaching an Inflection Point

By Ellen Hazen, CFA®, Chief Market Strategist

Executive Summary

• Second quarter corporate earnings were stronger than expected
• Inflation remains stubbornly high, while the Fed’s reluctance to raise rates has pushed long-term bond yields higher
• AI spending is a significant driver of both economic and corporate earnings growth
• AI spending growth is slowing for a variety of reasons, most of which are likely to continue
• The most likely scenario is that AI spending will remain at high levels but growth is unlikely to accelerate further

AI spending is driving the equity market. What will happen next?
The question is changing. It’s no longer whether AI spending will remain elevated – it almost certainly will. The question is whether investors should continue to expect accelerating growth.

AI capital spending is a significant force in the US economy. Just the four largest hyperscalers plan to spend over $700 billion in 2026, and according to Empirical Research Partners, all AI spending is estimated to be over 35% of all US capital spending in 2026.

Estimates for both 2026 and 2027 AI spending have grown rapidly over the past 7 months: in December, the five largest were expected to spend only $467 billion. Estimates for 2027 similarly increased as shown in the table below.

But the spending increases are slowing – over the past month, the increase was only $29 billion, down from the average increase of $40 billion each month seen earlier in 2026. This is for at least three reasons.

1. 1. Some planned datacenter builds have been delayed. Delays are being caused by a myriad of factors, including component shortages, permitting delays, environmental reviews, labor shortages (particularly of construction workers and electricians), and inability to lock down power generation solutions, whether from the local grid or from so-called “behind the meter” generation that has to source, for example, natural gas turbines, currently in shortage. Of 47 large datacenter builds planned over the next 5 years, we estimate that a quarter of them (accounting for over 40% of the total capacity) are likely to be delayed by some combination of these factors.
   2. The market is imposing discipline on spenders that have not described or demonstrated a clear path to profitability. Most recently, when Meta reported that it was increasing the low end of its capital spending plans from $125 to $130 billion, yet did not lay out a path to increased revenue from this investment, investors sent its shares down by 8% – or over $100 billion in market value – in a single day.
   3. Perhaps most importantly, the era of “easy” increases are over, because the companies have run out of cash flow from operations to fund higher spending. A year ago, we were relatively sanguine about the rapidly increasing spending, because all the major spenders could easily fund the (then much lower) numbers out of existing business profits. That line was crossed a few months ago; now, if these companies wish to further increase spending, they will need to either take on debt or issue equity or both, both of which are less attractive options.

Why we expect capital expenditure (capex) to continue at the $1 trillion level for a few years:

1. 1. Hyperscalers are still getting positive ROI. The CEO of Amazon laid out the basic math on the company’s recent earnings call. The payback for an AI datacenter is under 3 years, yet the hardware and systems have a useful life of at least 5-6 years. This Amazon sees as positive return on investment. Microsoft, Google, and Amazon all reported accelerating growth in their respective web services businesses, supporting this explanation. In contrast to the reaction to Meta, Microsoft’s market value soared by $540 billion the next day, the largest one-day increase of a single company in history.
   2. Corporate adoption is accelerating and is still low. The US Census survey on AI usage, the St. Louis Fed Real Time Population Trends survey on AI usage, and data from Ramp on AI usage all point to increasing usage. Note, these are not directly comparable; the surveys ask whether companies are currently using AI, while the Ramp data measure AI metrics like subscriptions.

      

   3. To get a more forward look at companies’ intentions, we reviewed the number of AI mentions on corporate earnings calls. By this measure company adoption continues to grow. For example, in the Q2 earnings season, the average S&P 500 company management mentioned AI over ten times in prepared remarks, compared to only five times back in the Q4 2025 earnings season. The nature of comments has changed too; a couple of quarters ago, the discussion centered around experimentation and pilots, and today, it is increasingly around revenue generation.

Conclusion – spending slowing likely is to continue and the market is pricing that in

We believe the growth of AI spending has slowed and will continue to slow as a result of delays, labor and materials shortages, and simple funding constraints. What does this mean for the market, and where could we be wrong?

We believe that the accelerating – indeed, parabolic – growth of many AI-related stocks is likely behind us; from here forward, growth is likely to be solid but not the eye-popping growth of the past year. Many of the AI beneficiary companies are not particularly expensive by traditional measures; many are trading at price-to-earnings multiples of between 20x and 40x, which is’ expensive relative to history but well below the 1999 internet bubble valuations and not unusual for growth companies. Regarding those companies at very high multiples, which are implicitly assuming continued acceleration in AI spending, we would be much more cautious.

The market appears to be pricing in a gradual deceleration of growth, given most companies price-to-earnings multiples. To be clear, we are not calling for spending to decline; rather we expect that the growth rate of spending will decelerate, and the pace of positive capital spending estimate revisions will slow. Both of these dynamics are likely to make equity appreciation slower than in the first half of 2026.

We could be wrong in either direction. Growth could continue accelerating if equity and debt markets (or governments) are willing to fund increasing investment without much concern. If this happens, we could see a repeat of the rapid price appreciation of AI-exposed stocks that we saw in the first half of 2026. On the other hand, instead of merely decelerating, growth could sharply slow, causing cascading delays, skepticism, excess inventory, and likely stock weakness.

Markets’ reaction to the first wave of AI spending was defined by speed. Their reaction to the second wave is instead defined by discipline. While AI spending is likely to remain high, the anticipated deceleration requires more careful scrutiny when analyzing potential investments.