Who Pays for America's Grid Upgrade? The Microsoft-Trump Fight Misses the Real Story

Your electricity bill climbed 7.4% over the past year, from 16.8 to 18 cents per kilowatt-hour. President Trump blamed data centers. Microsoft promised to pay its own way. And neither addressed the actual problem.

On Monday, Trump warned on Truth Social that he "never wants Americans to pay higher electricity bills because of data centers." Microsoft President Brad Smith responded Tuesday with commitments to cover both power consumption and grid infrastructure upgrades—no longer socializing connection costs to residential customers.

The exchange sounds decisive. Tech giants will finally pay their fair share. Case closed.

Except transmission and distribution expenses have nearly doubled since 2014 while generation costs stayed flat or declined, according to EIA data. Over 70% of U.S. transmission infrastructure is past the midpoint of its 50-year life expectancy. We're not paying for new capacity—we're paying to replace equipment that should have been upgraded decades ago. And that bill is coming due regardless of whether a single data center ever gets built.

Investor Chamath Palihapitiya put it bluntly on the All-In Podcast: data centers are "only part of the problem."

"We have a whole new way of living that is drawing more and more electrons," he said. "For the last 20 years, [utilities have] been underbuilding. So as they catch up, even if you have the data centers that are willing to pay their fair share, rates will still go up."

That's the uncomfortable truth. Microsoft can pay every penny of its infrastructure costs. So can Google, Amazon, and Meta. Your bill still climbs—because we're playing catch-up on two decades of deferred investment.

The Microsoft-Trump spat captures a real issue about cost allocation. But it's a sideshow to the main event: America is entering the first sustained electricity demand growth period since air conditioning caught on in the 1960s, and our infrastructure was already crumbling before AI showed up.

First up, the TL;DR

If you only have two minutes, read this.

President Trump warned Monday that he "never wants Americans to pay higher electricity bills because of data centers," putting Microsoft and the tech industry on notice. One day later, Microsoft President Brad Smith responded with commitments to cover both power consumption and grid infrastructure upgrades.

The controversy centers on who pays for connecting massive data centers to the grid. Union of Concerned Scientists found $4.3 billion in grid connection costs in just seven states during 2024, expenses that utilities often spread across all customers under outdated rules designed for smaller industrial loads.

But here's where it gets complicated: transmission and distribution expenses have nearly doubled since 2014, according to EIA data, while generation costs stayed flat or declined. The real driver? According to analysis in the Journal of Public and International Affairs at Princeton, over 70% of U.S. transmission infrastructure is past the midpoint of its 50-year life expectancy, and we're now paying to replace it.

Add in tariffs of 25-50% on steel, aluminum, and transformers, extreme weather upgrades, and permitting delays averaging 6+ years, and you've got a perfect storm of rising costs—regardless of data centers.

Here's where the research splits:

Lawrence Berkeley Lab found that states with higher electricity demand growth between 2019-2024 generally saw smaller price increases. In northern Virginia, large data center customers cover 9% of transmission costs, keeping residential rates below the national average. Mississippi reports data center revenue funded grid modernization without raising household rates.

The problem emerges when data centers connect under old rules that socialize infrastructure costs. That's what Microsoft's new policy addresses.

Microsoft's commitments:

• Pay utility rates covering full power costs plus grid infrastructure upgrades
• Stop requesting property tax breaks
• Reduce water consumption and invest in local job training
• Avoid having connection costs spread to other customers

Why this matters: You're about to see a decade-long buildout of AI infrastructure—OpenAI's plans alone require seven gigawatts, Trump's $500 billion Stargate Project multiplies that, and U.S. data center electricity use is projected to hit 6.7-12% of total demand by 2028. The question is who pays for the grid upgrades these data. centers require.

Microsoft's commitment, if enforced industry-wide over the next 6-12 months, could establish a new standard where tech companies cover their infrastructure costs. But the core challenge remains: 90% of transmission spending goes to replacing aging equipment, not building new capacity. Your electricity bill reflects decades of deferred maintenance, regardless of AI.

For deeper analysis on grid economics and infrastructure, check out Matt Estes' Explaining the Grid and Doug Lewin's Texas Energy and Power Newsletter.

Below, we'll dive into the facts we found from combing through these excellent deep dives to try to explain what's really going on with energy prices and AI.

How "Paying Your Own Way" Actually Works: The Stargate Example

So, OpenAI recently announced Stargate Community, a blueprint for exactly the kind of cost allocation Microsoft promised. And the scale reveals why this matters: OpenAI is already "well beyond halfway" to 10 gigawatts of AI infrastructure capacity, with the first site in Abilene, Texas already training frontier AI systems.

Ten gigawatts. For context, that's roughly equivalent to ten large nuclear power plants, or enough to power 7.5 million homes. And they hit the halfway mark in one year; from announcement in January 2025 to operational sites in January 2026.

The Stargate model shows three different approaches to keeping local rates stable:

• Wisconsin (Port Washington): Oracle and Vantage are working with WEC Energy Group to develop new solar and battery storage. The developers underwrite 100% of power infrastructure investment through a dedicated electricity rate from WEC—designed specifically to protect existing customers from price increases. The partners commit a minimum $175 million for local infrastructure upgrades and water restoration projects.
• Michigan (Saline Township): Oracle and Related Digital are working with DTE Energy using existing grid resources, augmented by new battery storage financed entirely by the project. The structure prevents impact on DTE's existing customers' energy supply or rates, while the project contributes its share to fixed costs of maintaining and improving the grid—potentially lowering per-unit costs for existing customers.
• Texas (Milam County): Partner SB Energy plans to fund and build new generation and storage to supply the majority of power needed—essentially creating dedicated supply that doesn't draw from the existing grid.

The commitment to "pay your own way" is actually an emerging industry standard driven by local pushback, regulatory pressure, and the recognition that socializing infrastructure costs to residential customers creates political opposition that blocks projects entirely.

Microsoft specifically referenced working with Wyoming utility Black Hills Energy to negotiate a larger power contract that brought on renewable energy without affecting existing customer rates. Wyoming lawmakers passed legislation allowing customized rates for large loads—exactly the kind of regulatory innovation that lets utilities serve hyperscale customers without cross-subsidizing from residential ratepayers.

But here's the key insight: these commitments only address new infrastructure built specifically for data centers. They don't fix the aging grid that everyone uses. They don't solve permitting delays. And they don't eliminate tariffs on transformers and transmission equipment.

If every AI company in America adopted Stargate's model tomorrow (paying for 100% of incremental infrastructure, bringing dedicated generation, funding grid upgrades), your electricity bill would still be rising 5-7% annually. That's because the bulk of transmission spending goes to replacing equipment installed when Eisenhower was president, not accommodating new load. Let's talk about that...

The Data Center Paradox: When More Demand Can Lower Prices

Here's what makes this confusing: large, consistent electricity loads can actually reduce per-unit costs for other customers.

Grid utilization averages around 50% most of the year. Electric infrastructure—transmission lines, substations, distribution networks—costs the same whether it's carrying 50% capacity or 80% capacity. Those fixed costs get divided by total kilowatt-hours consumed. More consumption means each kilowatt-hour carries a smaller share of fixed infrastructure costs.

Lawrence Berkeley National Laboratory examined state-level data from 2019-2024 and found that states with higher electricity demand growth generally experienced smaller retail price increases. In some cases, prices declined.

Northern Virginia's Data Center Alley provides a concrete example: large data center customers cover roughly 9% of transmission costs, helping keep residential transmission rates below the national average. Mississippi reports that revenue from data center loads funded grid modernization without raising household rates.

Texas, which has the highest concentration of data centers, saw electricity prices increase about 4% year-over-year—below the national average. California, with the third-most data centers, saw prices increase roughly 1%.

So what's the controversy about?

When the Rules Break: The $4.3 Billion Grid Connection Problem

The trouble emerges when utilities need to build new infrastructure to accommodate data centers, and old rules determine who pays for it.

The Union of Concerned Scientists analyzed grid connection costs across seven states in the PJM Interconnection (which covers 13 mid-Atlantic states): Illinois, Maryland, New Jersey, Ohio, Pennsylvania, Virginia, and West Virginia. They found $4.3 billion in costs during 2024 alone for connecting data centers to the transmission system—expenses that utilities passed to all customers under decades-old rules.

These aren't generation costs. They're not even system-wide infrastructure upgrades. They're direct connection costs: new transformers, transmission lines, and substation equipment specifically built to serve individual data centers. Many individual connections cost $25-100 million.

The mechanism is subtle but consequential. Utilities add these projects to their "locally planned" transmission needs. Those local plans automatically flow into the Regional Transmission Expansion Plan without additional regulatory review. Once in the regional plan, costs get bundled into transmission rates that all customers pay.

The rules made sense when the largest industrial customer might draw 5-20 megawatts. Modern hyperscale data centers draw 100-500 MW—as much as small cities. Some proposed data center campuses have power demands matching the total generation capacity of small states.

The infrastructure was designed for one scale. We're operating at another. Nobody updated the cost allocation rules.

That's what Microsoft's commitment addresses: paying for infrastructure upgrades their facilities require rather than having those costs spread across all ratepayers. If enforced industry-wide, it establishes precedent for how large loads fund their grid connections.

But even if every data center in America paid its full share tomorrow, your electricity bill would keep rising. Because the real driver isn't new connections—it's replacing what we already have.

The Decades We Ignored

Annual U.S. expenditures on transmission hit record highs, but the Princeton analysis found that 90% of that spending goes to replacing aging equipment. Construction of new high-voltage transmission has slowed to what grid analysts call a "trickle."

The math is brutal. A Brattle Group analysis estimates new and upgraded transmission equipment will cost $760 billion to $1.4 trillion over the next 25 years. The scale reflects decades of deferred investment and mounting delays on projects that are approved.

Some transmission projects take 6+ years for permits. Major lines have been stuck in regulatory limbo for over a decade. One effort to bring Canadian hydropower to New England—started in 2011—went through a statewide referendum in Maine, followed by a court case that overturned the referendum results. Fifteen years later, the project remains incomplete.

Meanwhile, 1,650 GW of solar and wind projects in advanced development stages await grid connections. That's actual generation capacity sitting idle because transmission planning and approval moves too slowly. Transmission congestion alone added $11.5 billion to customer bills last year.

And it's getting worse. Supply chains for transformers, transmission towers, and substations have few onshore manufacturers. Lead times stretch for months or years. Recent extreme weather events damage equipment faster than utilities can replace it.

This creates what economists call deadweight loss: costs imposed not by physical scarcity but by regulatory and logistical barriers. We're not paying market rates for electricity—we're paying a "policy premium" on every kilowatt-hour.

The Policy Premium: Tariffs, Permits, and the Cost of Delay

Grid infrastructure requires steel, copper, aluminum, power electronics, batteries, transformers, inverters, and miles of conductor. Steel and aluminum tariffs hit 25-50%. Grid-scale batteries face total duty stacks approaching 65%. The Trump administration announced additional tariffs on medium and heavy trucks, raising logistics costs across energy supply chains.

Every tariff adds cost without adding capacity.

The permitting burden compounds this. Projects wait years for approvals that don't meaningfully improve safety or environmental outcomes—they just create uncertainty, which raises financing costs and deters new entrants. Interconnection queues—where developers apply to connect new generation to the grid—are drowning in applications. Many are speculative, but the process can't filter them quickly, so real projects wait behind phantom ones.

Extreme weather adds another layer. Utilities must harden infrastructure against increasingly severe storms, wildfires, and temperature swings. California utilities are relocating portions of their distribution system underground to reduce wildfire risk—necessary for safety, expensive for ratepayers.

The nationwide effect: rising fixed costs of transmission capital have outweighed increases in variable generation costs. You're not paying more because electricity is harder to generate. You're paying more because delivering it requires replacing and upgrading infrastructure that's 50+ years old, during a period when supply chains are constrained, tariffs are high, and permitting takes years.

Data centers are a contributing factor in specific regions. But they're not the factor nationally.

The Demand Wave

Let's put some data in perspective: U.S. electricity demand is projected to rise roughly 25% by 2030. And data centers consumed about 4.4% of total U.S. electricity in 2023, projected to hit 6.7-12% by 2028.

That's significant. But it's one piece of a larger electrification wave: electric vehicles, heat pumps, domestic manufacturing reshoring, industrial facilities returning to the U.S. We're in the first sustained demand growth period since the 1960s.

The grid wasn't designed for this. For decades, electricity demand stayed relatively flat. Efficiency gains offset population growth. Utilities planned for incremental changes, not structural shifts.

Now demand is accelerating from multiple directions simultaneously. EVs alone could add massive load as adoption scales. Heat pumps replacing natural gas furnaces shift winter heating demand to the electric grid. Domestic chip manufacturing requires enormous, consistent power.

OpenAI's data center plans require seven gigawatts—enough to power millions of homes. Trump's $500 billion Stargate Project with OpenAI, Oracle, and SoftBank multiplies that several times over. But if you add up EVs, heat pumps, and reshored manufacturing, AI might represent a third of the total demand increase.

We're fighting about data centers while missing the forest for the trees. The question isn't whether data centers should pay their fair share (they should). The question is whether America will build enough generation and transmission capacity to handle the coming demand wave from all sources—or whether we'll ration electricity through constraint thinking.

Abundance vs. Constraint: Two Visions for the Grid's Future

This gets philosophical fast.

The DOE used to have a principle about the future grid: "Users should be able to use as much energy as they want, whenever they want it, and in whatever form they want." That's abundance thinking—building enough capacity that consumption isn't a problem.

That statement no longer exists at DOE. Instead, the department funds programs for demand response, time-of-use pricing, and automated control of consumer appliances to accommodate grid constraints.

California tells EV owners not to charge between 4-9 PM due to generation ramp constraints from solar. Northern England tells homeowners not to heat houses at night in winter. Incentivized demand response programs top out around 20% participation, then decline—because people eventually reject being "grid servants."

Grid engineer Jeffrey Taft once heard a chief economist at an electric system operator say, "I see no reason why people should be allowed to use as much electricity as they want."

There's the tension. Should the grid exist to serve consumers' needs? Or should consumers adjust behavior to serve grid constraints?

Constraint thinking says: We can't build fast enough, so we'll manage demand through pricing, incentives, and automated controls. Abundance thinking says: Build more generation, more transmission, faster permitting—make electricity cheap and plentiful enough that consumption isn't a zero-sum game.

The data center debate reflects this divide. Constraint advocates see data centers as load to be managed, discouraged, or at least forced to pay premium rates. Abundance advocates see them as customers whose revenue can fund infrastructure that benefits everyone—if we build enough capacity.

Recent research from Lawrence Berkeley Lab and Energy and Environmental Economics supports the abundance view: states with higher demand growth saw smaller price increases because more consumption spread fixed costs more efficiently.

But that only works if utilities have spare capacity. Once capacity constraints emerge, new load requires new infrastructure, and costs rise. The question becomes: do we build ahead of demand, or do we ration what we have?

America historically chose abundance. Rural electrification didn't ask whether farms could "afford" to use electricity—it built capacity to make electricity accessible. The Interstate Highway System didn't ration road access—it built enough roads that traffic could grow.

But electricity infrastructure operates under different rules. Utilities are granted regulated monopolies. They recover costs plus a return for investors. They don't compete on price or service quality. And they face 6+ year permitting timelines for major projects.

That regulatory structure was designed for a stable, slow-changing system. We're no longer in that world.

What Microsoft's Commitment Actually Means

Microsoft's Tuesday announcement commits to:

• Paying utility rates high enough to cover full power costs plus necessary grid infrastructure
• Not requesting property tax breaks from communities hosting data centers
• Reducing water consumption and replenishing more than they use
• Investing in local job training and AI education programs

The company already withdrew plans for a data center in Caledonia, Wisconsin after strong local opposition, even as it continues other projects in the state.

If enforced industry-wide, Trump suggested other companies will follow, this could establish precedent that large industrial loads fund their grid connections directly rather than socializing costs to ratepayers.

That matters. It addresses the specific cost allocation problem the Union of Concerned Scientists identified. It prevents situations where rural communities see electricity bills climb to fund infrastructure serving data centers they never wanted.

On January 21, 2026, OpenAI followed with mirrored commitments regarding communities that host its Stargate data centers. Given the pushback in communities against data center construction, you can likely expect others to follow suit.

"Stargate is a partnership with communities, and we can only achieve our mission by being good neighbors," according to a press release from the company that created ChatGPT. "Across all of our Stargate Community plans, we commit to paying our own way on energy, so that our operations don’t increase your electricity prices. Every community and region has unique energy needs and grid conditions, and our commitment will be tailored to the region. Depending on the site, this can range from bringing new dedicated power and storage that the project fully funds, to adding and paying for new energy generation and transmission resources."

But it doesn't address the underlying infrastructure crisis. Even if every data center pays its full share, 90% of transmission spending still goes to replacing aging equipment. Even if AI demand disappeared tomorrow, electricity prices would keep rising because the grid is falling apart and we're finally paying to fix it.

The honest accounting: data centers are a factor in how connection costs get allocated in specific regions. They're not the factor driving national price increases. The real drivers are decades of deferred maintenance, tariffs on essential hardware, permitting delays, supply chain constraints, and accelerating demand from multiple sources.

A note on water, since it comes up constantly: modern data center cooling systems recirculate water in closed loops. The water transports heat out of the facility—it doesn't get consumed (watch our chat with Microsoft's Scott Guthrie on how their new Fairwater system operates).

Now, older evaporative cooling systems do use water (it evaporates to dissipate heat). But as Friedberg noted on All-In, modern data centers don't rely on that approach. "The water issue is really a total hoax," he said—"a subhoax of this larger affordability issue." Stated more fairly, yet still critically: "You don't actually care about the water issue"; because if you did, there's a lot more worthy villains to contend with... (especially now that we're entering an era of "water bankruptcy", but don't worry, potential solutions to this abound).

Another example: OpenAI's sites use closed-loop or low-water cooling systems. Mayor Weldon Hurt of Abilene noted the site will use in a year what Abilene consumes in a single day; about half a day's worth of municipal water consumption annually. That's the result of cooling system design innovations specific to AI workloads versus traditional data centers. You're probably better off protesting forwater restrictions in your local town than you are at these new datacenter builds.

Point is, Microsoft's commitment to replenish more water than it uses is good policy. But for most new facilities, water consumption is minimal to begin with.

The Colocation Play: When Data Centers Build Their Own Power

Here's the part most critics miss: hyperscalers don't actually want to draw from the grid.

David Sacks explained the strategy on All-In: he says Trump understood from the start that AI companies would become "the biggest power companies" because they'd stand up their own generation rather than compete for grid capacity.

The approach is called colocation, or putting a data center and power generation next to each other, or "behind the meter." Once connected, these facilities can actually sell or donate excess power back to the grid.

Why would that help you? Scale economics. Power generation has massive fixed costs that get spread across total output. More generation means each kilowatt-hour carries a smaller share of those fixed costs. Sacks argues this could actually lower residential rates over time.

The bottleneck? Regulations. FERC (the Federal Energy Regulatory Commission) has rules that make colocation harder than it needs to be. Energy Secretary Chris Wright has directed FERC to streamline behind-the-meter arrangements—but bureaucratic timelines mean those changes are still months away.

If colocation works at scale, the data center "problem" becomes a data center solution: private investment in power generation that benefits everyone connected to the grid.

What Happens Next

By 2030, U.S. electricity consumption could rise 25%. We need massive transmission investment either way. Global investment in transmission hit $140 billion in 2023 and needs to exceed $200 billion annually by the mid-2030s just to meet existing policy commitments—$250-300 billion if we achieve full emissions goals.

The U.S. share of that investment is substantial. We're looking at $760 billion to $1.4 trillion over 25 years domestically. Someone pays for that.

The fight is over who. Should tech companies building hyperscale data centers pay directly for their grid connections? Yes—and Microsoft's commitment establishes that precedent. Should EV owners pay time-of-use rates that make charging during peak hours expensive? Maybe, but that's constraint thinking.

The real question: Do we want cheap, abundant electricity that powers economic growth, AI development, manufacturing, and electrification? Or do we want rationing, demand management, and telling people when they can charge cars and heat homes?

One path requires building faster, permitting faster, accepting that electricity infrastructure is critical national infrastructure that deserves streamlined approval. The other path manages scarcity through pricing and controls.

We're currently stuck between the two. Utilities face 6-year permitting timelines while demand accelerates. Tariffs raise costs while supply chains struggle. Regulators designed for a slow-changing system now face exponential demand growth.

Microsoft paying for data center connections is good policy. But it's a footnote to the main story: America needs a grid upgrade that was already 20 years overdue before AI became a thing. Your electricity bill reflects that reality.

The data center debate is real. The cost allocation problem is real. But if we solve both perfectly and don't fix the underlying infrastructure crisis, permitting delays, and tariff premiums, your bill still climbs 5-7% annually for the next decade.

That's the story Trump and Microsoft aren't telling.

The Next Next Step: Could Hyperscalers Fund Your Solar Panels?

Microsoft's commitment might be step one. Chamath Palihapitiya proposed step two: a $100-200 billion tax equity fund where hyperscalers pay to install solar and battery storage on American homes.

The logic is straightforward. These companies need social license to operate—community buy-in for their data centers. What better way to earn it than eliminating electricity bills for millions of households?

The mechanism already exists. Trump's "one big beautiful bill" preserved tax advantages for exactly these kinds of clean energy investments. A hyperscaler could deploy capital, take the tax benefit, and fund residential solar installations at scale.

"Step one is you go into a local area, you tell the local residents: we'll pay for the water, we'll make sure there's minimal noise, and we'll pay our fair share—even if it means paying more than you do for electricity," Chamath said. "Step two: here's a bunch of money to fit your house out with solar, storage, next-generation heat pumps."

The result? Homeowners become energy self-sufficient. They stop caring what the grid does. And that freed-up capacity can flow to commercial and industrial users—including data centers.

David Friedberg ran the numbers:

• U.S. electricity consumption: ~4 trillion kilowatt-hours annually
• Average price: 18 cents per kWh
• Total spend: roughly $750 billion per year
• Residential share: one-third (~$250 billion)
• Commercial and industrial: two-thirds (~$500 billion)

The math gets interesting. If you increased commercial and industrial rates by 50%, you could theoretically make residential electricity free—up to a reasonable monthly cap based on home size.

That 50% increase creates a market incentive for businesses to build their own power systems, which would increase total electricity supply. Meanwhile, residential customers stop worrying about rate hikes entirely.

Is it realistic? Complicated. But the framing matters: we're not talking about managing scarcity. We're talking about building abundance and deciding how to distribute it.

The residential math, on the other hand, is surprisingly tractable. There are roughly 80 million freestanding homes in the United States. Battery storage systems—like the ones Base Power installs—cost around $15,000 per home today, potentially dropping to $10,000 at scale.

Back-of-envelope: $100 billion could cover 10% of American homes. A trillion dollars—spread over a decade—could make most of them energy independent.

Every home that goes solar-plus-storage is a home that stops drawing from the grid. That's not demand destruction—it's demand relocation. The grid capacity those homes used to consume becomes available for data centers, EV charging, manufacturing, and everything else driving the electrification wave.

Chamath's point: "We would stop worrying about the grid. And we'd actually get a two-for-one—consumers don't have that anxiety anymore, and that extra capacity can go to commercial and industrial applications, including these data centers."

The Uncomfortable Question for Utilities

Here's a scenario worth watching: What happens to electric utilities if this all works?

If homeowners install solar and storage at scale, they become their own utilities. If commercial and industrial users build behind-the-meter generation, they stop buying from the grid too. Suddenly, the utilities have massive capital investments (all the transmission lines, substations, and generation plants) serving a shrinking customer base.

Chamath flagged this as a potential long-term short (not financial advice, at least not from me!): "If even a small modicum of what we just talked about materializes... what do the utilities do? What does all this capex go to? How do they pay back their bonds? What happens to their rate base?"

It's speculative. But if the abundance path wins—more distributed generation, more self-sufficient homes and businesses—the traditional utility model faces an existential question. The companies built to deliver centralized power may find themselves with stranded assets and declining revenue.

That's a decade-out concern. But it's worth noting: the infrastructure debate isn't just about who pays for upgrades. It's about whether the entire model of centralized electricity delivery survives the transition.

For deeper analysis on grid economics, infrastructure, and the abundance vs. constraint debate, check out Matt Estes' Explaining the Grid, Doug Lewin's Texas Energy and Power Newsletter, and Saul Griffith's Energy & Stuff on electrification economics, which informed this piece.