The AI theme has gone through several periods of doubt and bubble fears since 2023 and, after driving the lion’s share of earnings growth and market performance this year, AI stocks have wobbled again in recent weeks. Some of the biggest semiconductor winners are down by 20 to 40% this month as of Monday’s close, as the market digests outsized gains from the last 6-12 months and unwinds some of the crowding in this high momentum cohort. Does recent volatility mean the boom has run its course? As with the doubts before it, we don’t think so, and see a buildout that is still early.
A gut check on demand
In periods like these, it pays to stay grounded in the fundamental drivers of the buildout. The demand picture is strong, driven by an explosion in token consumption from AI models. Alphabet’s platform, for instance, now processes 3.2 quadrillion tokens a month, up ~600% from a year prior, as adoption spreads and agentic workflows push usage sharply higher. This has also driven significant acceleration in the revenue run-rate of the largest AI labs.
Hyperscalers are likely to keep investing
This pace of consumption highlights a gap in the supply of compute infrastructure needed. Hyperscaler capex estimates have more than doubled over the past year, with 2027 spending now likely to surpass $1 trillion. That spending has consumed nearly all of the free cash flow these companies once returned to shareholders, and several have tapped debt and equity markets to raise additional capital. Investors have taken a more skeptical view, and the hyperscalers have underperformed the broader U.S. equity market year-to-date. Yet debt-to-EBITDA ratios remain healthy and operating cash flows are expected to inflect higher in the coming years, potentially exceeding $900 billion by 2027.
The evolving AI bottlenecks
The AI ecosystem will likely be capacity-constrained for several more years, as the demand story requires ever more infrastructure investment. That capital flows to the critical bottlenecks of the data center buildout where architectural shifts keep reshaping the bottom-up opportunity, making an active approach to investing important.
Networking demand is strong, particularly in optical equipment, as cloud giants connect ever-larger clusters of chips. In compute, traditional CPUs are a growing opportunity as agentic workloads spread. In power, grid strain and public resistance are pushing builders towards "behind the meter" energy solutions. Lastly, the shortage of high-bandwidth memory has been a key focus this year, driving pricing power and earnings.
Our investment team recently spent time on the ground with many of these companies, and insights gathered further support this view around demand, investment and the shifting bottlenecks.
What this means for portfolios
The scrutiny of the AI trade is healthy, but so far, the fundamentals are standing up to it. Demand is compounding, the strongest balance sheets are funding the build and the companies supplying AI infrastructure are thriving.
Whether the biggest spenders earn an attractive return remains to be seen while the economics of AI services are still taking shape, which could keep markets volatile. But turning defensive too early carries its own risk. An active approach, one that appreciates how these bottlenecks can evolve and spreads exposure across the layers of the value chain, can help portfolios stay invested while riding out the swings along the way.
By Stephanie Aliaga & Nicholas Cangialosi - July 22, 2026