We All Want AI — But Who Wants the Infrastructure That Powers It?

The contradiction gets sharper when you look at what's happening across the country. Communities want economic development, high-paying technology jobs, stronger tax bases, better digital services, and continued American leadership in AI. At the same time, many of the projects required to support those goals face rising opposition. Data centers hit zoning resistance. Transmission projects take years to permit. New substations draw scrutiny. Gas generation raises emissions concerns. Solar and wind face land-use opposition. Battery storage prompts safety questions. Even advanced nuclear, which could eventually deliver enormous carbon-free baseload power, faces steep regulatory and development hurdles.

Individually, many of these concerns are reasonable. Collectively, they create a hard question: if we don't want data centers, power plants, transmission lines, substations, renewable projects, battery storage, or pipelines built near us, where exactly is the electricity for the AI economy supposed to come from? Electricity can't be generated, moved, and consumed without physical infrastructure. Every megawatt eventually has to come from somewhere.

This isn't only an AI problem. The U.S. is simultaneously electrifying transportation, expanding domestic manufacturing, building semiconductor fabs, electrifying industrial processes, and trying to reduce fossil-fuel dependence. AI data centers are landing on top of those trends. What makes AI different is the speed and concentration of the new demand: a hyperscale campus can request hundreds of megawatts — and increasingly more than a gigawatt — at a single location, creating electricity requirements comparable to a small city.

That scale changes the conversation. A utility used to adding incremental residential and commercial load can suddenly receive an interconnection request for 500 MW, 1 GW, or several gigawatts from one development. Serving it may require new generation, transmission, substations, transformers, and switchgear — assets that take years to plan, permit, procure, and build. The developer, meanwhile, wants computing capacity online much sooner. That mismatch between the speed of digital development and the speed of energy infrastructure is becoming one of the defining challenges of the AI boom.

Part of the problem may be psychological. Digital services have spent decades becoming invisible. You tap an icon and a movie streams. You upload thousands of photos and they simply stay available. You ask an AI model a complicated question and get an answer in seconds. Because the physical infrastructure is deliberately hidden from the user experience, it's easy to think of digital technology as almost weightless.

But every digital service has a physical supply chain: steel, concrete, copper, transformers, generators, cooling equipment, servers, networking hardware, batteries, substations, transmission capacity, and enormous amounts of electricity. AI adds another layer, because the specialized GPUs used to train and run advanced models can consume far more power than traditional computing.

The energy infrastructure behind that computing is equally physical. A Power Purchase Agreement may look like a financial contract on a spreadsheet, but the electricity behind it comes from an actual generation facility. A 500 MW solar project needs thousands of acres. A wind project needs turbines, roads, transmission, and interconnection. A gas plant needs turbines, pipelines, fuel supply, and air permits. A utility-scale BESS needs containers, inverters, transformers, land, interconnection, and sophisticated controls. Transmission projects can stretch hundreds of miles across multiple jurisdictions.

None of this means we should wave through every proposed project just because AI needs electricity. Communities deserve a meaningful voice on land use, environmental quality, water, electricity prices, and local infrastructure. But those decisions should acknowledge the tradeoffs. Rejecting one form of infrastructure generally means relying more heavily on another — or accepting slower growth and tighter electricity supply.

Maybe the biggest disconnect between the technology and energy industries is their fundamentally different clocks. Software updates in weeks. Computing hardware improves dramatically in a few years. AI models advance even faster. Electrical infrastructure operates on an entirely different timeline.

A major transmission line can take many years of planning, permitting, environmental review, land acquisition, engineering, procurement, and construction. Large generation faces similar challenges. Even equipment that once drew little public attention — transformers, switchgear, gas turbines, high-voltage gear — has become a constraint as utilities and developers compete for limited manufacturing capacity.

That creates a structural problem. Technology companies make decisions based on computing demand that could arrive within two or three years, while utilities plan infrastructure meant to run for 30, 40, or 50 years. A developer asks, "How fast can you deliver 500 MW?" The utility has to ask a much larger set of questions: Where does the electricity come from? Is there transmission capacity? What happens at peak demand? Who pays for the upgrades? What if the customer doesn't consume what it forecast? What happens in an extreme-weather event? Neither side is wrong — they just operate under very different realities.

That's why we're seeing AI developers explore options that would have seemed unusual for tech companies a few years ago: signing long-term PPAs, investing in renewable generation, contracting for nuclear electricity, evaluating behind-the-meter gas, deploying BESS, and exploring fuel cells and hydrogen. In other words, some of the world's largest technology companies are gradually becoming energy companies — because access to electricity is becoming a competitive advantage in AI.

For decades, computing companies could treat electricity as an operating expense: build the facility, connect to the utility, pay the bill, and focus on servers, software, and customers. That gets much harder when a single campus needs hundreds or thousands of megawatts.

Power is becoming part of the upstream supply chain for AI. A company can have the land, permits, fiber, GPUs, financing, and customer demand for a new campus — but without sufficient electricity, none of it operates. That makes access to power potentially as important as access to semiconductors, and it pulls utilities, IPPs, renewable developers, BESS suppliers, EPCs, transmission developers, gas companies, nuclear developers, equipment manufacturers, and energy investors into the AI ecosystem. The companies that solve the electricity problem may ultimately decide where the next generation of AI infrastructure gets built.

It's also changing site selection. Developers used to prioritize cheap land, tax incentives, fiber, workforce, and proximity to customers. Those still matter, but increasingly the first question is simpler: how many megawatts can I get — and when? Cheap land with great fiber isn't worth much for a 500 MW campus if the utility can only provide 50 MW for the foreseeable future. Conversely, locations with abundant generation, transmission capacity, favorable interconnection, or the ability to build behind-the-meter generation become extremely valuable even if other characteristics are weaker. That's a big reason AI infrastructure is expanding beyond traditional data center markets — the next wave of hyperscale development may follow electricity rather than population.

Saying AI uses a lot of electricity is accurate but incomplete. Steel manufacturing uses enormous energy. Aluminum production is energy-intensive. Semiconductor fabrication requires substantial electricity and water. Transportation consumes massive quantities of energy. We accept these because we value the products and services they provide.

AI should get the same complete framework. The question shouldn't just be "how much electricity does AI consume?" It should also be "what economic and societal value are we getting from that electricity, and how do we supply it as efficiently and responsibly as possible?"

That's not a blank check. If AI companies create extraordinary new demand, they should increasingly help finance the generation, transmission, storage, and infrastructure to serve it — and utilities and regulators need to ensure existing customers aren't unfairly burdened with costs created primarily for hyperscale developments. But energy consumption itself shouldn't automatically be treated as evidence that a technology is undesirable. The more useful conversation is about how the electricity is generated, how efficiently it's used, what infrastructure is required, who pays, and what benefits the technology provides. That's where the AI conversation and the energy conversation finally become the same conversation.

If AI becomes as deeply embedded in the economy as current trends suggest, arguing about whether data centers use too much electricity accomplishes little. The harder challenge is supplying that electricity reliably, affordably, and responsibly — which requires admitting there's no single technology that solves the problem everywhere.

Renewables will clearly play a major role. Solar and wind can provide enormous quantities of relatively low-cost electricity and let technology companies support new generation through long-term PPAs. But renewable output is variable, while AI data centers run continuously — a hyperscale campus can't stop computing when the sun sets or the wind drops.

Battery storage helps bridge that gap. BESS can shift renewable electricity across hours, cut peak demand, provide fast-response power, improve reliability, and support backup systems. For short-duration work, batteries are extremely effective. But economically storing enough electricity to run a multi-gigawatt campus through several days of bad weather is a fundamentally different challenge — one where duration, not just capacity, drives the cost.

Natural gas offers dispatchable generation that can be built closer to large loads, which is why behind-the-meter gas has entered the conversation — though widespread dedicated gas plants raise legitimate emissions and long-term-carbon questions. Nuclear offers reliable, high-capacity-factor, low-carbon generation, but new projects historically require long timelines and heavy capital. Geothermal, fuel cells, hydrogen, and long-duration storage could eventually add options, though their economics and geographic fit vary widely.

The likely answer isn't picking one winner — it's building energy portfolios. A future campus might combine utility power, a renewable PPA, on-site or nearby generation, hundreds of megawatts of storage, sophisticated energy management, and flexible workloads. Another might sit near nuclear generation. Another might use gas as a bridge until transmission catches up. Sites with favorable geology might use advanced geothermal; others might add fuel cells or hydrogen for long-duration resilience. The energy strategy will become as customized as the data center itself.

There's another side that gets less attention. Data centers are usually framed as a problem the grid must solve. But well-designed AI campuses could eventually become resources that help the grid.

Picture a 500 MW campus with a large BESS and sophisticated controls. Under normal conditions, it draws from the grid while optimizing battery charging around prices and system conditions. During extreme grid demand, the battery discharges and cuts the facility's net grid draw. Certain non-critical workloads could be shifted in time or location, reducing demand when the system is stressed. Add dedicated generation and a microgrid combining renewables, BESS, firm generation, and a utility interconnection, and the campus can dynamically optimize where its electricity comes from — behaving like a flexible participant rather than a constant 500 MW load.

The key distinction is between computing reliability and grid consumption. A data center may need uninterrupted computing, but that doesn't mean it must draw exactly the same power from the utility every second. Batteries, generation, workload management, and thermal storage can provide flexibility behind the meter while maintaining the reliability customers expect. If the industry builds that capability at scale, the narrative shifts from "How will the grid survive AI?" to "How can AI infrastructure help strengthen the grid?"

None of this means communities should rubber-stamp every data center because AI is economically important. The contradiction shouldn't become an excuse to dismiss legitimate concerns as hypocrisy.

A community weighing a major data center has every right to ask what it means for electricity rates, water, noise, land use, transmission, tax revenue, jobs, air quality, and local development — and to understand whether infrastructure costs land on the developer or get transferred to existing customers. The same standard applies to the supporting energy infrastructure: if a project requires a new gas plant, the emissions belong in the discussion; if hundreds of acres of battery storage are proposed, safety and emergency-response planning matter; if transmission lines are needed, affected communities deserve a voice.

But those discussions should include the consequences of saying no. If a transmission project is rejected, electricity still has to reach growing loads somehow. If gas is rejected, another form of firm capacity may be needed. If renewables can't get permits, clean-energy goals get harder. If data centers are rejected outright, the investment may simply move to another community, state, or country. Every energy decision involves tradeoffs; pretending otherwise makes responsible planning harder, not easier.

There's also an enormous variable that gets less attention than generation: efficiency. Today's AI-demand forecasts rest on assumptions about how fast computing expands and how efficiently it runs — and both can change dramatically. Chipmakers keep improving performance per watt. Operators are deploying more efficient cooling, including advanced liquid cooling. Developers are making models more efficient. Workloads can be optimized to produce the same result with fewer resources. Even modest efficiency gains become enormous across gigawatts: on a hypothetical 1 GW campus, a 10% reduction in effective power is about 100 MW — the equivalent of avoiding another very large industrial load. The cheapest megawatt may ultimately be the one the data center never needs.

I understand why AI's energy footprint makes people uncomfortable. The numbers are enormous, and the speed of development is unlike anything the power industry has seen in decades. Communities watch massive data centers appear while utilities discuss new plants, transmission, substations, and rate structures. Asking questions about that isn't anti-technology — it's responsible.

But the discussion often ignores how deeply AI is already embedded in modern life. We can't simultaneously expect ever more powerful digital services, cloud computing, automation, advanced medical research, autonomous systems, and smarter manufacturing while treating the physical infrastructure behind them as something society shouldn't have to see or build. There is no invisible electricity system. Every digital interaction eventually touches physical infrastructure. Every megawatt has to be generated somewhere. That reality shouldn't frighten us — it should push us toward better planning.

So instead of asking whether AI is good or bad because it uses electricity, we should be asking how to build the most efficient, resilient, economically responsible, and environmentally sustainable infrastructure to support the technologies society increasingly values. That means technology companies taking greater responsibility for the infrastructure their growth requires; utilities planning more aggressively for large loads while protecting existing customers; regulators modernizing interconnection and permitting without abandoning oversight; expanding generation, transmission, and storage while continuing to invest in efficiency — and acknowledging there will be tradeoffs.

Artificial intelligence isn't weightless. Neither is the cloud. Behind every AI application is a physical network of servers, buildings, transmission lines, substations, power plants, batteries, transformers, and people keeping it all running.

The debate over AI has moved far beyond whether people like chatbots or AI-generated images. AI is rapidly embedding itself throughout the economy, often invisibly — fraud detection in finance, production optimization in manufacturing, diagnostics and research in healthcare, routing and supply chains in logistics, and integration into virtually every major software platform. Whether or not someone actively chooses to use an AI tool, artificial intelligence is increasingly part of the infrastructure of modern life.

That makes the energy conversation unavoidable. If AI keeps growing at the anticipated pace, the U.S. will need significantly more generation, transmission, substations, transformers, battery storage, and other infrastructure. Some will come from renewables, some from gas, potentially much more from nuclear; storage will manage variability and improve reliability; geothermal, fuel cells, and hydrogen may contribute as they become competitive. There won't be one universal solution.

What we can't realistically do is demand more AI while opposing nearly every form of infrastructure required to support it. We can't reject transmission lines, power plants, renewable projects, battery storage, substations, and data centers, then expect unlimited computing capacity to somehow materialize. The physical infrastructure behind our digital economy is often invisible to consumers, but it's very real.

That doesn't mean every project should be automatically approved — the opposite, really. The scale of AI development makes thoughtful planning more important than ever. Developers should be responsible for a fair share of the infrastructure costs they create. Utilities should protect existing customers from subsidizing speculative load growth. Communities should understand the economic and environmental consequences. Technology companies should keep improving efficiency while supporting cleaner, more resilient electricity.

But we also need to acknowledge the benefits of building this infrastructure. New generation increases supply. Transmission investment relieves congestion. Battery storage strengthens reliability. Modern substations and transformers replace aging equipment. Long-term PPAs finance renewables. Investments made initially to serve large loads can, structured well, contribute to a stronger, more modern grid.

That may be one of the biggest opportunities of the AI revolution. The U.S. already needs major investment in its aging electrical infrastructure, and AI is creating an enormous new economic incentive to accelerate it. If utilities, regulators, technology companies, energy developers, and communities approach this strategically, some of the hundreds of billions flowing into AI could also help finance the next generation of American energy infrastructure. The objective shouldn't be building electricity infrastructure at any cost — it should be using this moment to build better infrastructure: more generation, greater transmission capacity, smarter grids, more storage, higher efficiency, stronger reliability, and a more diversified portfolio.

We criticize AI while increasingly relying on it. The energy industry is beginning to feel the same contradiction: we want the digital economy but often resist the physical infrastructure supporting it. Eventually, those two realities have to meet. AI isn't happening in an imaginary cloud — it's happening inside physical buildings connected to physical electrical systems consuming very real megawatts.

If we want the benefits of the AI revolution, we also have to be willing to have an informed, realistic conversation about what it takes to power it.

Run the numbers behind the debate — The abstractions in this article become concrete when you size them. Our Data Center Load Calculator turns a campus into megawatts, the Utility-Scale BESS Sizing Calculator and LCOS Calculator scope the storage that firms it up, and the Hydrogen vs. Battery Storage Calculator shows where each storage technology wins as duration grows.

Written by Chris Kalowes, founder of WattThe?! — 15+ years in utility-scale battery energy storage (BESS), renewable energy, and AI infrastructure, across utilities, IPPs, EPCs, developers, and technology providers. Energy Intelligence. Simplified.