The promise of artificial intelligence has never been more ambitious. Mark Zuckerberg recently predicted that within five years, billions of people will each have a personal AI agent handling their finances, health, and relationships. But the infrastructure required to deliver that promise is already reshaping the lives of people who will likely never use an AI agent. In Georgia, families are losing their homes to make way for power lines built to serve data centers. Across the country, electricity bills are climbing in communities that have no connection to the technology they are subsidizing.
Georgia Power is using eminent domain to seize more than 330 properties across Coweta and Fayette counties to build a 35-mile transmission line for its Project Wansley expansion. Between 20 and 30 homes will be demolished to make way for the power corridor. Roughly 80 percent of the new generating capacity the utility is adding will serve data centers, according to reporting from Fortune. The $16 billion grid expansion is one of the largest in Georgia's history, driven almost entirely by the electricity demands of artificial intelligence.
The human price of a power line
For the families in the transmission line's path, the developer's compensation offers have been deeply inadequate. Residents told Fortune that Georgia Power's offers ran roughly $100,000 below their properties' market value. The utility has said condemnation represents less than one percent of its total land acquisitions, but for the homeowners, the loss is total. A 500-kilovolt power line will cut through their properties, permanently changing the landscape and the value of what remains.
Eminent domain is a legal process that allows governments and utilities to seize private property for public use, provided compensation is paid. The justification in this case is that reliable electricity for data centers is a public benefit. Yet the residents losing their homes are not the ones who will reap the rewards of AI-powered personal assistants. They are bearing a disproportionate share of the cost for a technology that is still in its early stages and may never deliver on its grandest promises.
The project is part of a broader trend. Data centers have become the new factories of the digital age, consuming enormous amounts of power and water. Hyperscale facilities for companies like Meta, Google, and Amazon can each draw as much electricity as a mid-sized city. The rapid expansion of AI training and inference has accelerated this demand, pushing utilities to build new transmission lines and power plants at a pace not seen in decades.
The AI spending spree and Meta's gamble
Meta's own balance sheet reveals the scale of the bet. The company posted revenue of nearly $61 billion in the second quarter of 2026, up 28 percent year over year, but its free cash flow collapsed 91 percent to $784 million as capital expenditure surged. Meta has guided investors toward total spending of $130 billion to $145 billion this year, most of it on AI infrastructure. That is a staggering sum, equivalent to the annual GDP of several small nations.
The spending spree is not unique to Meta. Microsoft, Amazon, and Alphabet have all announced massive AI infrastructure investments, collectively planning to spend hundreds of billions of dollars in the coming years. These companies are racing to build data centers, acquire AI chips, and develop the next generation of large language models. Yet the revenue streams from AI are still uncertain. While some AI products have seen rapid adoption, many are offered at a loss or as free add-ons to existing services.
Meta's track record with speculative bets should give investors pause. The company's Reality Labs division, which builds the metaverse, has lost roughly $88 billion since 2021 on a bet that has yet to produce a mass-market product. The metaverse promised a new era of social interaction, immersive work, and digital ownership. Instead, it has generated billions of dollars in losses and a line of VR headsets with modest adoption. Investors are understandably wary of repeating that experience with AI.
Meta's stock fell nearly ten percent after the Q2 results, a sign that investors are more moved by collapsing cash flow than by the promise of an agent in every pocket. The market is beginning to question whether the enormous capital expenditures are justified by the potential returns. If AI agents fail to generate significant revenue, the $145 billion spending will become a colossal sunk cost.
Zuckerberg's vision is that AI agents will book appointments, manage household budgets, and negotiate on behalf of their users, a layer of intelligence accessible to everyone with a phone. Meta already runs AI agents for more than a million business users on WhatsApp and Messenger. But those agents are limited to specific tasks like answering customer queries or scheduling simple appointments. They are a far cry from the fully autonomous assistants that have been promised.
The widening gap between vision and reality
The harder question is whether the human costs of building AI infrastructure can be justified by a product that does not yet exist at scale. No personal AI agent currently on the market can reliably manage a household budget, let alone the full spectrum of daily life that Zuckerberg described. The gap between the vision and the present capability is wide. For all the money being spent on AI, the practical, everyday benefits for ordinary people remain largely speculative.
The paradox is that to deliver an AI agent that could help families save money on their electricity bills, utilities must first spend tens of billions of dollars on power infrastructure that will raise those same bills. The short-term pain is real and distributed unevenly, while the long-term payoff is uncertain and could be captured by the very companies making the investments.
Data centers are also straining electricity grids in regions that have nothing to do with AI. In the PJM grid area, which runs from New Jersey to Illinois, capacity charges jumped more than tenfold in a single auction cycle, driven largely by data center demand. Capacity charges are the fees that utilities pay to ensure enough electricity is available to meet future demand, and those costs are passed on to consumers. Industrial electricity prices rose 31 percent in Pennsylvania and 26 percent in Ohio in one year, against a seven percent national average. That is a crushing burden for factory towns already struggling with inflation and high energy costs.
Community resistance: a growing backlash
The costs are not confined to one utility's service area. Data center protests have spread to 42 US states, with more than 140 rallies in a single day of action coordinated by a coalition that cuts across political lines. Community pressure has blocked or delayed tens of billions of dollars in planned construction. Residents cite concerns about noise, water use, environmental impact, and the fact that data centers create few jobs once they are built. The jobs argument is particularly weak: a modern data center operates with a small staff of engineers and security personnel, nowhere near the employment levels of traditional factories.
The backlash has been bipartisan. In some areas, conservative residents object to the expansion of utility monopolies and the use of eminent domain. In others, liberal residents focus on climate change and the carbon footprint of power-hungry data centers. The coalition opposing data center construction has found common ground in local control and property rights. This has led to new zoning laws, moratoriums, and legal challenges in states as diverse as Texas, Virginia, and California.
Virginia's Loudoun County, nicknamed Data Center Alley, has become a cautionary tale. The county hosts hundreds of data centers along the internet backbone, and they now consume more electricity than many cities. Residents have complained about the constant hum of cooling systems, increased traffic during construction, and the visual blight of massive warehouses. Local officials have begun to question whether the tax revenue is worth the strain on infrastructure and quality of life.
The power demands are not limited to the United States. In Europe, Ireland has placed restrictions on new data centers near Dublin because they now account for almost 20 percent of the country's electricity consumption. The Netherlands, Singapore, and Uruguay have also imposed limits or moratoriums on data center construction. The global nature of the AI boom means that the consequences are being felt worldwide, from the water-scarce deserts of Arizona to the temperate landscapes of Northern Europe.
Even as communities push back, the industry continues to argue that AI will solve the very problems it creates. Companies say AI can optimize power grids, reduce energy waste, and help develop new clean energy technologies. Zuckerberg's pitch is that the pain is temporary and the payoff transformative. He imagines a world where AI agents free people from drudgery, making them more productive, healthier, and better connected. But that vision requires a leap of faith, and the evidence so far suggests that the leap is getting more expensive.
The families in Coweta County facing demolition orders do not have the luxury of waiting for the future to arrive. Their present is defined by a 500-kilovolt power line cutting through their property, a right-of-way that will uproot their lives for the sake of data centers training models that might, eventually, produce something useful for ordinary people. Whether that trade-off is worth making is a question the industry has not yet been forced to answer. The enormous sums being spent, the rising power bills, and the bulldozers in suburban neighborhoods are all part of the real cost of AI agents — a cost that is being paid today, long before the promised benefits materialize.