When investing, your capital is at risk. The value of investments can go down as well as up, and you may get back less than you put in. The content of this article is for information purposes only and does not constitute personal advice or a financial promotion.
9–5 Investor Summary
What’s happening
By 2026, more of the market’s excitement about AI has shifted away from software and platforms toward the physical side: electricity, grid access, interconnection, and data centre capacity.
Why it matters
Chips are no longer the only bottleneck in AI. Power delivery is becoming just as important. This shift is changing where pricing power, backlog visibility, and investor focus are found.
What the market is missing
This isn’t just about energy or technology. It’s really a story about how capital is allocated. Infrastructure owners are already making money from scarcity, while many platforms are still working to prove the value of their AI investments.
Key risk to watch
This idea could lose strength if AI becomes much more power-efficient sooner than expected, or if a lot of new power generation and capacity come online before demand catches up.
Investor lens
Pay attention. The key question isn’t who has the best model, but who controls the bottleneck.
The New AI Trade Is Not Purely Digital
During most of the AI boom, the market expected the main winners to be companies building models, selling chips, and offering cloud services. That made sense and explains much of what happened from 2023 to 2025. But by 2026, it’s clear that AI isn’t just about software. It’s also about physical infrastructure, and those who control scarce resources tend to benefit most. Today, those scarce resources include power, substations, interconnection rights, land ready for development, cooling systems, and long-term contracted capacity.
This is the simplest way to bring the three drafts together into one clear idea. Data centre energy and infrastructure stocks are doing better because the market is now valuing what’s hardest to provide, not just what’s easiest to sell. There are plenty of algorithms, but not enough permitted megawatts.
Modern Data Centres: Electrical Trends, Risks, and NEC® 2026 Implications – IAEI Magazine
Why Power Has Become the Binding Constraint
The International Energy Agency now projects that global electricity consumption by data centres will more than double by 2030 to roughly 945 TWh, with usage rising about 15% a year from 2024 to 2030. In the United States, the IEA says data centres account for nearly half of electricity demand growth through 2030. Reuters, citing the EIA, reported this week that U.S. electricity consumption is expected to hit record highs again in both 2026 and 2027, with AI and crypto data centres among the main drivers.
This matters because power grids weren’t built to handle thousands of GPU-intensive facilities running nonstop, such as factories. Uptime Institute’s 2025 survey shows the industry is facing higher costs, tighter power limits, supply chain delays, and the challenge of meeting AI’s power needs. Earnings calls can’t ignore this part of the AI story. Even if a model can scale in the cloud, a data centre still needs to be built, connected, cooled, permitted, financed, and powered reliably.
In markets, bottlenecks work like toll booths. When everyone has to go through the same narrow spot, whoever owns that spot usually benefits.
Why Big Tech Looks Different in This Phase
This doesn’t mean Big Tech is struggling to operate. It just means expectations are higher. Investors no longer reward big AI spending just because it’s large. They want proof that big infrastructure investments will turn into steady, high returns. BlackRock’s 2026 survey found that, among 732 EMEA-based institutional clients, only about one in five saw the biggest U.S. tech companies as the best AI opportunity. More than half preferred companies supplying power to data centres, and 37% chose infrastructure as their top AI investment.
That is a subtle but important shift. When the market starts favouring firms supplying energy for the AI buildout over those buying it, the narrative has changed. Power is revenue for the utility, the generator, or the infrastructure owner. Power is a cost for the hyperscaler. Capex is optional in theory, but competitively mandatory in practice. That is not a pleasant combination for valuation multiples.
Key comparison table
Primary-source grounding for the table (selected):
Apple FY2024–FY2025 net sales and operating income are presented in its Form 10‑K tables.
Microsoft FY2025 revenue/operating income and FY2024 “over $245B revenue / over $109B operating income” are stated in its annual report narrative and results commentary; figures are rounded where phrased as “over.”
Alphabet FY2024–FY2025 revenue and operating income totals are reported in its Form 10‑K segment and consolidated tables.
Amazon's FY2024–FY2025 consolidated net sales and operating income are reported in its Form 10‑K.
Equinix FY2023–FY2025 revenues, income from operations, and Adjusted EBITDA are reported in its Form 10‑K.
Digital Realty FY2024–FY2025 total operating revenues and operating income are in its FY2025 results release tables.
NextEra FY2024–FY2025 revenue is shown via an accessible third‑party compilation (Macrotrends); the SEC filing itself could not be reliably parsed in this environment, so operating margin is left as a data gap.