How the AI era is reshaping America’s electric grid — and what your community needs to know
We’ve been here before — many times.
This guide does not tell you what to think or how to vote. It provides factual information so you can form your own informed opinion.
We are entering another period of transformation.
We don’t know exactly what it will look like when it arrives — we never do. The people who watched the first locomotive move didn’t see the railroad city coming. The families who watched Edison’s first lights flicker on Pearl Street didn’t see the electric century coming. The engineers who built the interstate highway system didn’t fully see the suburban economy it would create.
What we do know, from every one of those moments, is this: transformation runs on infrastructure. And infrastructure has to change before the transformation can fully arrive.
That is where we are right now. Society is changing again. The tools are different — artificial intelligence, data centers, new demands on the electric grid — but the story is the same one humanity has been living for centuries. Progress arrives. Infrastructure scrambles to keep up. Communities make decisions, some wise and some not, about what they want the new era to look like in their backyards.
This guide is not about being afraid of what’s coming. It’s not about stopping it. It’s about understanding it well enough to make good decisions — for yourself and for your community.
We’ve been here before. We’ll be here again. The question is never whether change comes. It’s whether we’re ready for it.
| AT A GLANCE |
| 945 TWh Global data-center electricity projected by 2030 — roughly double today (IEA)* |
| 2–3× U.S. data-center demand by 2028 vs. today (Dept. of Energy)* |
| 224 GW New peak electric demand NERC projects over the next 10 years — a 69% increase over prior forecasts* |
| 40% Of AI data centers Gartner projects will face power restrictions by 2027* |
| Electricity availability — not chips — is now the primary constraint on AI infrastructure growth |
| Communities that negotiate well can capture jobs, grid upgrades, and tax revenue. Those that don’t may bear costs without benefits. |
What’s Actually Being Built
The AI economy isn’t just software. Behind every AI tool you use — every search result, every chatbot response, every medical diagnosis assisted by algorithm — sits a physical computer in a physical building drawing real electricity from the same grid that powers your home.
Right now, in communities across America, decisions are being made about where to build that infrastructure. Those decisions will shape electric rates, tax revenues, job markets, and quality of life for a generation. AI infrastructure is not one thing — it is a stack of six interdependent layers. Here is what each layer is, in plain terms.
The Six Layers of AI Infrastructure
Layer 1 · Chips — The Foundation
Every AI system runs on specialized chips that are extraordinarily difficult to manufacture. NVIDIA designs the dominant AI chips; TSMC in Taiwan makes them — along with chips for AMD, Apple, Google, and Amazon. One company, in one country, produces most of the world’s advanced AI silicon. TSMC’s revenue from AI computing rose from 51% to 61% of its total business in three years.* The U.S. government has responded by pushing domestic chip manufacturing — TSMC has committed up to $165 billion toward U.S. operations.*
Layer 2 · Data Centers — The Building
A data center is a large building full of computers running constantly. There are roughly 4,000–4,400 in the U.S. today. The newest — hyperscale campuses — cover 100,000+ square feet and draw 50+ megawatts of power. In Florida: a $2.6 billion campus was approved in Polk County; “Project Tango” in Palm Beach County was voted down in July 2026; 12+ counties have enacted moratoriums. Florida’s SB 484 (effective July 1, 2026) preserves local zoning authority and bars utilities from shifting infrastructure costs to existing ratepayers.
Layer 3 · Power and the Electric Grid — The Governing Constraint
This is the layer that most directly affects you. The IEA projects global data-center electricity use will reach 945 terawatt-hours by 2030 — roughly double today, growing 15% annually.* The DOE projects U.S. data-center demand could double or triple by 2028.* NERC’s 2025 reliability assessment found peak demand will grow 224 GW over ten years — a 69% increase over its prior forecast — and warned that much of North America faces risk of failing to meet power needs during extreme weather.* The system, NERC noted, is “changing faster than the infrastructure needed to support it.”
Layer 4 · Cooling — The Hidden Cost
Traditional data centers use evaporative cooling towers that consume significant water. Modern AI facilities are moving to liquid cooling — circulating coolant directly to chips — which can cut water use up to 90%.* For high-density AI workloads, liquid cooling is no longer optional. Water-use plans are a legitimate and negotiable permit condition communities can require.
Layer 5 · Networks — The Connective Tissue
Data centers connect through vast fiber-optic networks. An estimated $13 billion in subsea cable projects comes online between 2025–2027 — nearly double the prior three-year total.* Tech giants now own the backbone: Google leads with 33 cables, Meta has 16, Microsoft 6, Amazon 4.* Florida’s peninsula geography puts it at a natural junction for Caribbean and Latin American cable routes — a real infrastructure asset.
Layer 6 · Who Controls Each Layer
The AI infrastructure stack is highly concentrated. A handful of companies dominate every layer.
| LAYER | WHO DOMINATES IT |
| Chips | NVIDIA (design), TSMC (manufacturing), ASML (equipment) — three companies, largely in Asia |
| Data Centers | Amazon (AWS), Microsoft (Azure), Google Cloud, Meta — the four hyperscalers control most cloud capacity |
| Power | Utilities and independent power producers — but demand is outpacing supply across most regions |
| Cooling | Vertiv, Schneider Electric, Modine — specialized equipment companies scaling rapidly |
| Networks | Google (33 cables), Meta (16), plus dark fiber owners in key corridors |
| AI Services (top layer) | OpenAI, Google, Anthropic, Amazon, Microsoft — the companies building AI products on top of this infrastructure |
A Grid Built for a Different Era
Before you can understand what AI is doing to the electric grid, you need to understand what the grid already is: aging infrastructure designed and built for a world that consumed far less electricity than we do today — and far less than we are about to demand.
On September 4, 1882, Thomas Edison threw a switch at 257 Pearl Street in lower Manhattan. His generator lit 400 lamps for 82 customers in a one-quarter-square-mile area. That was the entire grid. Within a generation, electricity had become the foundation of modern life. Within two generations, the grid covered a continent. Today we are asking that same aging system to carry a load its builders never imagined.
Consider how U.S. electricity use has grown since the grid was born:
| ERA | U.S. ELECTRICITY CONSUMPTION — AND WHAT WAS DRIVING IT |
| 1970s ~1,400 TWh/year* | The grid as we know it was largely designed and built in this era. Demand was driven by household appliances, industrial manufacturing, and commercial lighting. Computers existed but were rare, room-sized machines. There was no internet. |
| 2000 ~3,800 TWh/year* | Demand more than doubled as the internet arrived, personal computers spread to every home and office, and the digital economy was born. The grid strained to keep up — the Northeast blackout of 2003, affecting 55 million people, was a direct result of aging infrastructure meeting modern load. |
| 2020 ~3,800 TWh/year* | Consumption leveled off despite massive growth in data and digital services, because efficiency gains in appliances, lighting, and industrial equipment offset new demand. Cloud computing, streaming video, and smartphones all expanded on roughly the same grid footprint. |
| 2035 (projected) ~5,000–6,000 TWh/year* | AI changes the efficiency equation. Unlike prior digital growth, AI workloads are computationally dense and power-intensive by design. The DOE projects U.S. data-center demand alone could double or triple by 2028. NERC projects 224 GW of new peak demand over the next decade — more than the total generating capacity of many countries. |
The Efficiency Question — and Why AI Is Different
U.S. electricity consumption stayed nearly flat from 2000 to 2020 despite explosive digital growth. The reason: LED bulbs, tighter appliance standards, smarter industrial motors, and dramatically improved data center efficiency held demand in check. Most of those gains are already captured.
AI breaks the pattern. Each new chip generation does more work per watt — but lower cost per computation drives more computation. This is the Jevons Paradox: fuel-efficient cars led Americans to drive more miles, not fewer. More efficient AI chips mean companies run more of them. And unlike prior digital growth, AI workloads are computationally dense by design — there is no efficiency shortcut.
The counterargument worth knowing: AI is also the most powerful tool yet built for managing electricity use. AI-controlled building systems show savings of up to 44% in some studies.* Google cut its own data center cooling energy by roughly 40% using AI.* Smart grids using AI shift industrial loads to off-peak hours, reducing strain. The honest forecast: AI consumes far more than it saves through 2035 — but communities that invest in smart grid infrastructure alongside data center development will be better positioned than those that don’t.
The Problem: The Grid Is Already Past Its Design Life
The grid being asked to carry this new load was built for the 1970s. That is not a figure of speech — it is the engineering reality:
- 70% of transmission lines and large power transformers are more than 25 years old.*
- The average power transformer in the United States is over 40 years old — approaching or past its design lifespan.*
- Most transmission lines were engineered for electricity loads that are a fraction of what they carry today.*
- The American Society of Civil Engineers gave the U.S. electrical grid a D+ in its 2025 Infrastructure Report Card — down from a C- just four years earlier.*
- Grid reliability has been declining since the mid-2010s, as aging equipment meets growing demand and more frequent extreme weather events.*
The U.S. grid was designed for a world using 1,400 terawatt-hours per year. It now carries nearly three times that load on the same aging bones. AI wants to add hundreds of terawatt-hours more — on infrastructure that the nation’s own engineers grade as barely passing.
This is not an argument against AI infrastructure. It is an argument for being honest about what adding that load requires: not just new data centers, but serious investment in the transmission lines, substations, and power generation capacity that will make those data centers possible without degrading reliability for everyone else.
When a community approves a large data center, it is not just approving a building. It is triggering a chain of infrastructure decisions — some of which may affect the reliability and cost of electricity for every resident in the region.
Think of It Like the Interstate Highway System
“A data center is not the destination. It becomes the anchor tenant of an entirely new economic ecosystem.”
People don’t move to a city because it has a highway. They move because the highway attracts manufacturers, warehouses, hospitals, universities, startups, distribution networks, and investment capital. AI infrastructure works the same way.
A region that builds reliable AI infrastructure — electricity, data centers, fiber, skilled workers — does not just get data centers. It gets the technology companies, research institutions, and investment that cluster around them. That is what happened in Northern Virginia. It is beginning to happen in Austin and Phoenix. The communities that got infrastructure right in earlier eras — highways, railroads, ports — built lasting prosperity. The ones that didn’t are still catching up.
The Three-Stage Economic Wave
When major AI infrastructure arrives, the economic effect unfolds in three stages — and communities often see the costs before the benefits, which drives opposition.
Stage 1 · Construction (Years 1–8): Electricians, HVAC, concrete, steel, engineers, trucking, security. Thousands of workers, restaurants and hotels busy, equipment suppliers active. The most visible wave.
Stage 2 · Infrastructure Economy (Years 5–15): The center itself employs a few hundred — but it attracts cloud firms, cybersecurity companies, AI startups, research labs, and universities. Northern Virginia, Austin, and Phoenix show what this looks like a decade in.
Stage 3 · Regional Transformation (Years 10–20+): Higher household incomes, expanded tax base, better broadband, stronger utilities, venture-capital investment. The region becomes a destination for industries that depend on advanced digital infrastructure — which today is most industries.
Why Communities Are Pushing Back — and What’s Actually True
Communities often see Stage 1 costs before Stage 2 and 3 benefits. That is a real timing problem, and it explains a lot of the backlash. Here are the most common objections — and an honest assessment of each.
| WHAT PEOPLE SAY | WHAT’S ACTUALLY TRUE |
| “They’re taking our electricity.” | Partly true. Large data centers draw significant power and can require utilities to build new infrastructure. The critical question is who pays for it — the developer or existing ratepayers. Florida’s SB 484 bars the latter. |
| “They use too much water.” | True for older facilities. Modern liquid-cooled designs with closed-loop systems can cut water use up to 90% compared to conventional cooling. This is a negotiable permit condition — communities can require it. |
| “There aren’t enough permanent jobs.” | Operations staffing is modest compared to manufacturing. But Stage 2 and 3 effects — tech ecosystems, higher wages, tax base — are real. The question is whether the community captures those benefits through negotiated agreements. |
| “They’re ugly and disruptive.” | Large facilities with backup generators and substations do change the character of nearby areas. Setbacks, screening, noise limits, and design standards are legitimate permit conditions. |
| “Our taxes subsidize them.” | Residents rightly question tax incentives when the public return is unclear. Transparent community benefit agreements — local hiring, education investment, infrastructure upgrades — can make the tradeoff explicit and fair. |
A “Yes, If…” Framework for Communities
The most productive framing for this debate is not “Should we allow data centers?” It is: How do we ensure our community benefits from them?
Instead of “No data centers in my backyard” — a more powerful position is: “Yes, if they make our community stronger.” These are measurable conditions communities can negotiate rather than simply accepting or rejecting projects.
Before approving any major AI infrastructure project, communities should be able to answer yes to each of the following:
- Will it strengthen the electric grid — and improve reliability for existing residents, not just serve the facility?
- Will local residents get jobs, training, and apprenticeship opportunities during both construction and operations?
- Will schools and colleges receive tangible investment or partnership commitments?
- Will roads, utilities, and infrastructure improve as part of the deal — not just be strained by it?
- Will tax revenues demonstrably support public services rather than disappear into incentive agreements?
- Is the water-use plan sustainable, documented, and enforceable?
- Are nearby neighborhoods protected from noise, light pollution, and visual impact through enforceable conditions?
- Is there a community benefit fund for direct investment in surrounding neighborhoods?
When these conditions are in place, opposition typically softens — because residents can see tangible local gains, not just corporate benefit.
What a 2035 AI Growth Corridor Looks Like
By 2035, the fastest-growing regions in America will likely be those that combined these assets deliberately, early, and on terms that benefited both investors and communities. The more of these a region can combine, the more likely it becomes a long-term AI growth corridor.
| ASSET | ECONOMIC RESULT |
| ⚡ Reliable electricity | Industrial expansion — the foundation everything else depends on |
| 🏢 AI data centers | Cloud infrastructure, anchor tenant for a growing tech ecosystem |
| 🌐 Fiber networks | Startup growth, low-latency access to compute capacity |
| 🎓 Universities | Skilled workforce, research partnerships, technology transfer |
| 🏠 Affordable housing | Talent attraction — engineers and technicians need places to live |
| ✈️ Airports | Business investment, executive access, supply chain logistics |
| 🏭 Manufacturing | Higher wages, complementary industrial base |
| 🔬 Research institutions | Innovation ecosystem, patent development, commercialization |
“The AI economy will locate where electricity, talent, capital, and community support come together. Communities that simply say ‘no’ may preserve today’s landscape but risk missing tomorrow’s investment. Communities that say ‘yes’ without conditions may bear costs without capturing enough benefits. The challenge is to negotiate AI development so it improves prosperity and quality of life — while protecting the community.”
Understand the issue. Think critically. Decide for yourself.
Sources and Notes
* Figures marked with an asterisk are externally sourced and should be verified against primary sources before publication or citation. Sources include: NERC 2025 Long-Term Reliability Assessment (January 2026); IEA Electricity 2024 report; U.S. Department of Energy data-center energy-use projections; Utility Dive (NERC coverage); Tom’s Hardware data-center cooling analysis (2025); Subsea Cables Network and CNBC subsea cable investment reporting; semiconductor supply chain analysis from VaaSBlock and IndMoney (2026); Gartner projections on AI data-center power constraints. Florida-specific data: SB 484 (2026 FL Legislature); Polk County commissioner records; Palm Beach County zoning records.