America is having the wrong argument about AI data centers. One side sees windowless industrial giants devouring electricity and water while shifting the infrastructure bill onto families who are already struggling. The other sees the physical foundation of the next scientific revolution and warns that every local delay is a direct gift to Beijing. Both have merit.
Data centers are not new. They already host your bank records, medical files, family photos, and municipal databases. Treating them as if they arrived with ChatGPT is historical amnesia. What has changed with artificial intelligence is not their existence, but their velocity and density.
The numbers illustrate the shift. Data centers consumed roughly 4.4% of total U.S. electricity in 2023. Updated federal estimates from the Department of Energy project that by 2030, data centers could account for 9.5% to 15.3% of total national electricity consumption. When the federal government’s own research laboratories cannot narrow a demand forecast closer than a factor of 1.6, the honest conclusion is not that the surge is exaggerated. It is that we are building faster than we can measure. You cannot govern what you cannot count.
THE END OF INTERNET PRIVACY: EUROPE HIJACKS AMERICAN TECH, AND CHINA BENEFITS
The second ledger
Every technology that ever mattered arrived looking like a bill.
Researchers at Stanford and Carnegie Mellon asked a representative sample of American adults what it would take to give up AI tools for a month. The average was $124.50, up 27% in a single year. Scaled across 115 million adult users, that’s $172 billion in annual consumer welfare.
These studies can only measure what people know they are using. As the Stanford researchers concede, their figures understate the true value because AI is increasingly embedded in services users never identified as AI. I previously made that point about exposure. It applies with equal force to benefit.
AI is everywhere. The Food and Drug Administration has authorized 1,524 AI-enabled medical devices, 76% of them in radiology. In 2015, there were six decisions made. In 2025, there were 333, and that’s just within the medical field. In finance, Visa blocked $40 billion in fraudulent transactions over a single 12-month span, saving the consumer huge pass-through costs. None of these required anyone to open a chatbot, yet they directly benefited from AI.
The retiree who has never typed a prompt still had her scan read faster, her credit card protected, her loan file processed, her grocery shelves stocked, and her utilities load-balanced, all with AI. She did not consent to any of it, and she was never told she benefited from it either. The cost arrives itemized. The benefit arrives anonymously. How do you quantify something you do not see, yet rely on its value?
The environment: The net-zero dividend
The International Energy Agency estimates that if manufacturers, utility companies, building operators, and transport fleets adopted the AI tools that already exist, the emissions avoided by 2035 would be approximately 1.4 billion tons of carbon, roughly 3 times what every data center on Earth emits, even under the agency’s most aggressive standards. Here, AI more than pays for itself.
National security
Scientific discovery can be regulated or delayed, but it rarely stays contained, especially when a rival nation possesses the technical knowledge and strategic incentive to create innovation friction.
Beijing treats computing power and data centers as strategic national infrastructure, coordinating land, energy, and subsea fiber, with a speed our federal system was never designed to match. If America constrains its physical compute buildout without a coherent strategy, we will not stop the science. We will simply export it to an adversary operating with minimal domestic constraints and possibly nefarious intent.
Yet, here is the critical reality both sides miss: Washington has already acted. Executive Order 14318, signed July 23, 2025, directs federal agencies to streamline environmental reviews, make federal land available, and expedite qualifying AI data-center projects, including facilities requiring more than 100 megawatts of new load.
While Washington accelerated the buildout, the accountability side remained voluntary. That voluntary system is expanding. On July 23, the White House announced that more than 200 additional utilities, developers, cooperatives, and states had joined the Ratepayer Protection Pledge, which it says now covers 80% of power delivered to consumers.
Meanwhile, as of April 27, individual states were considering legislation addressing developer cost, responsibility, usage, and reporting. The challenge is that tech moves at a much faster pace than governance.
Asymmetric acceleration
The remedy is not to stall the buildout and hand the future to China; it is to complete the contract responsibly. Speed was granted by executive decree; responsibility must now be codified into law.
A complete National AI Infrastructure Contract requires four binding provisions:
First, standardized resource disclosure: Developers utilizing fast-track federal permitting must report standardized, near-real-time data on electricity consumption, water usage, and backup generation to regulators, with aggregated public disclosure. A surge we cannot measure is a surge we cannot manage.
Second, enforceability: Developers must underwrite the new generation, grid interconnections, transmission upgrades, and water-recycling infrastructure their facilities require. A data center must arrive as a net contributor to the local grid, not a competitor to the families living beside it.
Third, unified security and geographic review: Ownership, supply chains, chip custody, and proximity to defense installations must be audited before construction. Efficiency and national resilience demand geographic distribution.
ANTHROPIC ARMAGEDDON: DARIO AMODEI’S DOOMSDAY FEARS CONVENIENTLY CRUSH HIS COMPETITION
Fourth, repurposing and decommissioning bonds: Developers must post a bond before construction begins. If technological change or market conditions render the facility obsolete, the bond would fund decommissioning, site restoration, or feasible repurposing. Without it, the town is left holding an empty warehouse.
Let’s remember, we are not passive observers in a world of AI. We are the architects of its infrastructure. The same ingenuity that made this acceleration possible can engineer the accountability capable of sustaining it. When faced with an unprecedented technological challenge, with huge benefits to humanity, we are the generation that stepped up and said yes.
Jacqueline Cartier is a corporate and legislative strategist focused on communications, crisis leadership, public trust, and emerging technologies that shape human behavior and decision-making. Follow her on LinkedIn.