Broadcom printed $8.4 billion in AI revenue last quarter, up 106% year over year. That is not a forecast. It is a line item on a filed earnings report. When I see a number like that sitting next to a market narrative that says "anything but AI," I pay attention to the number.
The institutional case for energy and AI infrastructure stocks in 2026 is straightforward: the money has already been committed, and the earnings are already arriving. Investors waiting for a safer entry are pricing in a correction that the fundamentals do not support over a 12 to 24 month horizon.
Follow the Capex, Not the Commentary
Microsoft, Amazon, Google, and Meta have committed over $405 billion in AI-related capital expenditure this year, a 62% increase from 2025. That spend does not evaporate. It flows directly into the revenue lines of chipmakers, power providers, storage companies, and networking firms. Arista Networks has consensus estimates calling for 25% sales growth in 2026, with analyst earnings estimates revised upward 7% in the past 60 days alone.
Pure Storage analysts project roughly 30% annual earnings growth through 2027, with a median price target implying 45% upside from current levels. These are not meme stocks riding sentiment. They are companies whose order books are filling because hyperscalers physically cannot stop building.
A single ChatGPT query consumes about 10 times the power of a Google search. New data centers are targeting gigawatt-level loads, comparable to powering a small city. For a household, think of it this way: one of these facilities draws enough electricity to serve roughly 750,000 homes. The grid cannot absorb that demand without massive new investment, and the American Society of Civil Engineers gave U.S. energy infrastructure a D+ rating last year. That grade is not a sentiment indicator. It is a structural constraint that will take a decade to fix.
Constellation Energy, which signed 20-year clean energy supply deals with Microsoft and Meta, has gained about 195% over 2 years and trades at 29.6x forward earnings, roughly 20% below its recent highs. A 20-year contract with Microsoft is not speculative froth. It is recurring revenue with a creditworthy counterparty.
The Bubble Objection Deserves a Hearing, Then a Rebuttal
Ray Vega and others will point to Seagate's 645% one-year gain or Hut 8's 432% surge and call it late-cycle excess. Fair. Some individual names have run far ahead of near-term earnings power, and a P/E of 56 on Seagate demands perfection. I would not chase those specific tickers at these levels.
But conflating a handful of overheated names with the entire infrastructure theme is the analytical error. The power grid layer, the optical networking layer, the nuclear equipment suppliers: these companies are selling into a demand curve that U.S. power markets have not seen in 20 years. Fidelity's own portfolio managers noted that American power demand had been in gradual decline for roughly 2 decades before AI reversed the trend. That reversal is not priced into utility-adjacent stocks the way it is priced into pure-play AI software names.
GE Vernova, MYR Group, Powell Industries, and BWX Technologies all hit new all-time highs recently on grid upgrade demand. BWX, which supplies nuclear power equipment, advanced 5.9% in a single session. These are not retail-driven momentum trades. They are institutional bets on physical infrastructure that takes years to build and cannot be canceled with a software update.
I will grant the bears one thing: geopolitical uncertainty and a potential natural gas supply surge could cap energy prices and compress margins for some generators. That is a real risk. But it is a risk to a subset of the thesis, not the thesis itself. Cheaper natural gas actually accelerates data center buildouts by lowering operating costs for the hyperscalers writing those $405 billion in checks.
The market is offering you a chance to buy the picks-and-shovels layer of the largest capital expenditure cycle since the buildout of the internet itself. Broadcom's $8.4 billion quarter is not the ceiling. It is the proof of concept.
