Data Percent Power
America's Data Center Hunger — and One Overdue Sequel
The United States now consumes nearly forty percent of global data center electricity, a figure that becomes concrete when set against the International Energy Agency's projection that worldwide data center power consumption could double by 2030. The primary growth driver is AI inference workloads: every large-language-model query, every AI-generated image, every document analysis consumes meaningful electricity, and the volume of those workloads is rising sharply.
The grid consequences are already visible geographically. Northern Virginia hosts what is estimated to be the world's largest concentration of data centers, and some counties there have effectively halted new data center approvals because local grid capacity cannot support additional load without transmission upgrades that take years to complete. Major operators — Microsoft, Google, Amazon, and others — have committed to hundreds of gigawatts of new renewable capacity through power purchase agreements, but renewable intermittency means the gap between clean-energy commitments and reliable baseload power is, in many cases, being filled with natural gas, creating tension with the sustainability pledges those same companies have made publicly.
On a lighter note, author Harlan Coben has broken his publicly stated rule against sequels to agree to a continuation of Netflix's Myron Bolitar series. Coben's adaptations have been among Netflix's most reliable international performers, and the reversal suggests the platform's internal data showed audience demand strong enough to make a creative case that overrode a longstanding authorial commitment — a small signal about how streaming platforms are weighing franchise continuity against anthology structure.
Stepping back from the day's events, the most honest assessment of the AI productivity question — whether the technology is genuinely transforming the economy or arriving more slowly than predicted — is that current measurement tools are too blunt to give a definitive answer. Economic history, including the decades-long lag before electrification showed up in productivity statistics, suggests transformative technologies can be invisible to standard metrics until they suddenly are not. The signal to watch: sector-specific productivity data from the Bureau of Labor Statistics for professional services — law, consulting, finance, software development — over the next four to six quarters. A divergence from historical trend in those sectors would be strong evidence that real gains are running ahead of the measurement. And if Anthropic's IPO, should it occur, prices above a hundred billion dollars despite today's eleven percent market share figure, that too would tell a story — that sophisticated institutional investors are pricing long-term trajectory, not current reality.
Forty-Five Illegal Diesel Generators Near Two Schools: The Data Center Reckoning Arrives in New Jersey
Sam Altman acknowledged publicly this week that Americans 'hate data centers' — a notable admission from the chief executive of a company pushing a fifty-billion-dollar infrastructure buildout that has already surpassed its original ten-gigawatt Stargate target. The hostility, Altman implicitly conceded, is not irrational. Large-scale AI data centers consume as much power as small cities, generate significant heat and noise, require enormous quantities of water for cooling in regions where water scarcity is increasingly acute, and employ relatively few local workers — meaning communities absorb substantial infrastructure and environmental costs without the job creation that would typically justify them.
A Microsoft-backed data center in Vineland, New Jersey, has provided the week's starkest illustration of how that tension plays out in practice. The facility was found to have been operating forty-five diesel generators without the required environmental permits. The generators are located approximately one mile from two schools. Environmental groups are demanding Microsoft halt operations entirely. Diesel generators produce nitrogen oxides and particulate matter — pollutants with documented respiratory health effects, particularly in children. Running forty-five of them without permits near schools is not a technical paperwork lapse; it is a substantive public health compliance failure.
What the New Jersey case exposes is a widening gap between the pace of AI infrastructure deployment and the regulatory frameworks designed to govern it. Permitting processes exist to conduct environmental review, assess health impacts, and notify affected communities. When a company bypasses those processes, it is effectively deciding unilaterally that its timeline supersedes the requirements — a pattern visible across the technology industry for years, summarized as deploying first and seeking permission later, if at all. The implicit political favorability the current administration has shown toward AI development creates pressure on regulatory agencies to move slowly on enforcement, but forty-five unpermitted diesel generators near schools is the kind of documented failure that becomes difficult to manage away.
Altman's public acknowledgment of community hostility signals that OpenAI is aware it faces a perception and relations problem capable of generating real regulatory and political headwinds for its infrastructure ambitions. Whether the response will be substantive community engagement or cosmetic outreach remains to be seen. What is certain is that energy consumption will remain a flashpoint: if AI computing scales as industry projections suggest, the power demands of the next five years will require significant new generation capacity, major efficiency improvements, or both — and the communities hosting the infrastructure will keep demanding an accounting of why they bear the costs.