For the past three years, artificial intelligence has been sold as one of the greatest productivity revolutions in modern history. Business leaders promised that AI would eliminate repetitive work, increase efficiency, and dramatically reduce operating costs. Investors responded by pouring hundreds of billions of dollars into AI companies, while executives across nearly every industry rushed to integrate chatbots, AI coding assistants, image generators, and automated customer service into their daily operations.
The economic argument seemed almost too good to question. Why continue paying employees to perform tasks that software could complete in seconds? From writing marketing copy and generating software code to answering customer questions and summarizing documents, AI appeared capable of replacing entire categories of work for a fraction of the cost.
That promise became one of the driving forces behind the largest technology investment boom since the early days of the internet.
Yet only a few years later, cracks are beginning to appear in that narrative. Companies that once viewed AI as an inexpensive replacement for human labor are now discovering that running these systems at scale can be extraordinarily expensive. Enterprise AI budgets are ballooning, finance departments are questioning whether the promised savings are materializing, and some organizations are beginning to limit employee access to premium AI tools after unexpectedly large bills.
The irony is difficult to ignore. While workers were told they were too expensive, businesses are now learning that artificial intelligence comes with a price tag of its own.
At False Solutions, we’ve spent the past year examining artificial intelligence from a different perspective than most technology publications. While much of the public conversation focused on what AI could do, we asked a different question: What does AI actually cost, and who ultimately pays for it?
In The New Data Center Gold Rush, we explored the rapid expansion of AI infrastructure as technology companies raced to secure land, electricity, and water for the next generation of hyperscale data centers. In Why Are Data Centers Becoming the New Power Plants?, we examined how utilities across the country are planning new generating capacity specifically to satisfy projected AI demand. More recently, Big Tech’s Data Center Boom Is Being Subsidized by You looked at the public incentives, tax breaks, and utility investments helping finance this unprecedented buildout.
Taken together, those stories challenged one of AI’s biggest selling points. The technology appeared inexpensive because many of its true costs were simply being shifted somewhere else. Instead of paying workers, companies increasingly depended on enormous investments in electricity, computing hardware, cooling systems, transmission infrastructure, and public subsidies that rarely appeared in discussions about AI’s affordability.
Those hidden costs never disappeared. They were simply absorbed by utilities, taxpayers, local governments, and communities asked to host increasingly larger data centers.
One of the biggest misconceptions surrounding artificial intelligence is that it exists only in the digital world. To most users, AI feels almost weightless. A question is typed into a chat window, and within seconds a polished response appears on the screen. The process feels almost magical.
In reality, every prompt travels through an enormous industrial network that most users never see. Behind every conversation is a vast collection of servers packed with high performance graphics processors, cooling equipment, backup generators, transformers, substations, and transmission lines that operate continuously to keep AI systems online.
That physical infrastructure requires staggering amounts of electricity. It also requires significant volumes of water for cooling in many locations, large quantities of construction materials, and increasingly scarce computer chips that remain in high demand around the world.
This is why communities from California to Virginia have become increasingly skeptical of massive new data center projects. Residents are not objecting to artificial intelligence itself. They are questioning whether their neighborhoods should absorb the environmental and economic costs associated with supporting an industry whose benefits often flow elsewhere.
The digital economy, it turns out, depends on a very physical foundation.
For years, technology companies promoted AI as a way to reduce payroll expenses. That calculation often looked compelling when executives compared the monthly subscription cost of an AI tool to the annual salary of an employee.
The comparison, however, overlooked what happens when thousands or even tens of thousands of employees begin relying on AI every day.
Unlike traditional software, many AI platforms charge based on usage. Every prompt consumes computing resources, and those computing resources translate directly into operating costs. A handful of employees occasionally using AI may represent only a modest expense. An entire enterprise relying on AI for coding, document review, customer service, research, and content generation can generate billions of tokens each month, resulting in unexpectedly large invoices.
Recent reporting suggests that several major companies have already begun rethinking unrestricted AI use after watching costs climb far faster than anticipated. Finance departments are increasingly asking the same question communities have been asking for months: Is the return on investment really there?
The central irony of the AI boom is that companies hoped to reduce labor costs, but many are now replacing payroll expenses with infrastructure expenses. Instead of paying additional employees, businesses are paying cloud providers, AI vendors, chip manufacturers, data center operators, electric utilities, and energy developers.
The money did not disappear. It simply changed destinations.
That distinction matters because labor costs generally stay inside an economy in visible ways. Workers pay rent, buy food, support local businesses, raise families, and contribute to their communities. Infrastructure costs, especially in the AI economy, often flow upward into some of the largest and most powerful corporations in the world.
When a company replaces a customer service team with an AI platform, the savings may look attractive on a spreadsheet. But the broader economy loses wages, while the energy system absorbs new demand, and communities are asked to accept the pollution, water use, and land impacts associated with the infrastructure behind that platform.
This is not simply automation. It is a transfer of cost and power.
Even as businesses wrestle with rising AI expenses, the public continues absorbing many of the industry’s hidden costs. Ratepayers may be asked to fund grid upgrades needed to serve massive data centers. Local governments may offer tax breaks to attract facilities that promise economic development but deliver relatively few permanent jobs. Communities already overburdened by industrial pollution may face new backup generators, new substations, and new gas-fired generation built to satisfy projected electricity demand.
Freshwater resources are also becoming part of the fight. In regions already facing drought, extreme heat, or competing industrial demand, the idea of using large volumes of water to cool data centers raises basic questions of fairness. Who gets priority when water becomes scarce? Residents, farms, ecosystems, or the next wave of AI infrastructure?
These are not anti-technology questions. They are public interest questions.
If artificial intelligence is going to become part of everyday life, then the infrastructure supporting it must be planned with transparency, accountability, and environmental limits. Communities should not be expected to sacrifice clean air, water security, or affordable utility bills simply because technology companies insist that the AI race cannot wait.
False Solutions is not opposed to artificial intelligence, and we are not opposed to data centers. AI has real potential to support scientific discovery, improve healthcare, strengthen climate modeling, expand language access, and help people work more efficiently. Used responsibly, it can be a powerful tool.
But powerful tools still need rules.
The question is not whether society should use AI. The question is whether we build the infrastructure behind AI in a way that protects workers, communities, ratepayers, water resources, and the climate.
That means prioritizing energy efficiency before new generation, requiring data centers to use clean electricity rather than driving new fossil fuel demand, protecting water supplies, limiting public subsidies unless there are clear public benefits, and ensuring communities have a real voice before projects are approved.
It also means rejecting the idea that innovation automatically justifies harm. The same companies promising an intelligent future should be expected to make intelligent choices about where and how they build.
The first chapter of the AI revolution was built on the promise that artificial intelligence would be cheaper than people. Today, that assumption is facing its first serious test.
Companies are scrutinizing AI budgets more closely. Investors are asking whether the spending can produce real returns. Communities are pushing back against data centers that consume vast amounts of electricity and water. Regulators are beginning to ask tougher questions about who ultimately pays for the infrastructure required to support this new digital economy.
The conversation is evolving because the cost of AI was never just the monthly subscription. It was always the power plant, the transmission line, the cooling water, the public subsidy, and the neighborhood asked to host another warehouse-sized data center.
As we have argued throughout our AI series, the challenge is not artificial intelligence itself. The challenge is building an AI economy that does not simply shift its financial, environmental, and social costs onto everyone else.
That is not anti-technology.
It is simply asking whether the future we are building is as intelligent as the machines we are creating.