TL;DR — Key Takeaways
-
AI can boom while data centers struggle. Rising demand for artificial intelligence doesn’t guarantee profits for every company building and financing the infrastructure.
-
Communities are questioning the AI buildout. Concerns over energy costs, water consumption, noise and local benefits are reshaping where data centers get built.
-
Today’s AI infrastructure bets face tomorrow’s technology. Smaller models, specialized chips and on-device AI could change computing demand before some planned facilities even open.
You can believe AI will change the world and still question whether that data center down the road should be built. You can expect spending on AI to grow and still wonder whether the company borrowing billions to supply it will make money.
Those positions are becoming harder to separate as the AI buildout reaches communities, utility bills and investment portfolios. Residents want to know what they are being asked to live with. Investors want to know what they are paying for. Both deserve better answers than another announcement about how many billions someone intends to spend.
That is why we put together our new Techstrong special report, The AI Buildout Meets Reality. It examines the local opposition, the construction pipeline, the financing and the technology changes that could reshape demand before some planned facilities open.
Read the full special report: The AI Buildout Meets Reality
The Pipeline Is Not the Plan
The report opens with three very different outcomes. In Loudoun County, Virginia, Amazon and local officials reached a preliminary agreement to pursue other uses for a contested site. In Arizona, Project Blue continued in unincorporated Pima County after Tucson rejected annexation. In Wisconsin, Microsoft reported its first Fairwater facility operational, with another under construction.
Download: The AI Buildout Meets Reality (PDF)
Put those stories under one headline about the data center boom and you lose much of what matters. A withdrawn proposal, a project proceeding despite local opposition and an operating facility tell us different things.
The same problem runs through the cancellation statistics. An announcement is not a financed project. A utility connection request is not an operating load. A delay is not necessarily a cancellation, and a relocation does not necessarily reduce national demand.
None of that makes the backlash imaginary. People have legitimate concerns about noise, water, electricity costs, tax incentives and who gets a say. The election has given those concerns a bigger platform. Developers still have to address them after the votes are counted.
Demand Is Growing. The Returns Need Work
Futurum’s research makes a substantial case for continued investment. In its first-half 2026 AI Platforms survey, fielded in March, 78.4% of respondents expected AI budgets to increase over the following 12 months. Just 1.2% expected decreases. Those are spending intentions, but they hardly describe enterprises walking away from AI.
The returns data is less reassuring. July results cited in the report show only 15.8% of surveyed CIOs reporting sustained ROI at scale from in-house AI and 12.6% from vendor AI.
Gartner approaches the problem from the infrastructure owner’s side. Its October analysis warns that more than $3.5 trillion of AI infrastructure spending could be exposed to impairment by 2030. It also forecasts that more than 80% of neocloud providers could cease independent operations by then.
Those claims require careful reading. Losing independence through an acquisition is not automatically a failure. Impairment exposure is not a tally of money already lost. The report examines those distinctions and the limits of comparing forecasts that measure different parts of the market.
Still, growing demand does not answer Gartner’s concern. Customers can use more AI while providers earn too little to cover financing, operations and the next equipment replacement.
As the founder of Techstrong, part of The Futurum Group, I want our readers to see the strongest case for continued investment alongside the strongest questions about its economics. Futurum’s demand outlook deserves attention. So do the obligations sitting on the balance sheets of the companies serving that demand.
The Technology Will Keep Moving
There is another complication: the infrastructure takes years to build, while the technology it houses keeps changing.
Smaller models, specialized chips and AI running on enterprise systems, phones and other devices could change where computing happens and what customers will pay for it. Local execution can make useful AI cheaper and more accessible, encouraging more use. That does not guarantee the additional work flows through the same centralized facilities investors expected to supply it.
This is part of the argument I explore in The Indispensability Trap. AI can become more indispensable while some of the infrastructure built around it becomes less profitable. Value can move toward applications, distribution and customer relationships, even as computing demand rises.
The railroad analogy is useful here. Infrastructure can transform an economy without rewarding every company that financed its construction. With AI, the equipment inside the building can also lose its competitive advantage much faster than the building itself.
Our report follows these questions through community agreements, power constraints, neocloud financing and three possible paths for the buildout. The sidebars explain how to distinguish project outcomes, financial distress and the indicators that would strengthen either the bullish or skeptical case.
I remain bullish on what AI can do. That makes me want a better account of how we are building for it, who benefits and who pays when the assumptions prove wrong.
Read The AI Buildout Meets Reality, the full Techstrong special report.
