PJM Is 6 GW Short by 2027.
65 Million People Are in the Blast Zone.

The Grid Nobody Talks About

Somewhere in Valley Forge, Pennsylvania, a control room monitors the heartbeat of the largest power grid in North America. Most Americans have never heard of Pennsylvania-New Jersey-Maryland Interconnection (PJM) — yet it keeps the lights on for 65 million people across 13 states and Washington DC. From the suburbs of Chicago to the skyscrapers of Northern Virginia, from the steel mills of Ohio to the government buildings of the National Capital Region, PJM dispatches approximately 180 GW of generation capacity across a territory that stretches from Illinois to New Jersey.

PJM is not a utility. It does not own a single power plant, transmission tower, or distribution wire. It is a Regional Transmission Organization (RTO) — essentially the air traffic controller for electrons. PJM coordinates the dispatch of generation, manages the wholesale electricity market, plans transmission expansion, and administers the capacity market that is supposed to guarantee that enough power plants exist to meet future demand. Over 1,000 generation owners, transmission owners, and load-serving entities participate in PJM's market. In 2023, PJM facilitated over $40 billion in wholesale electricity transactions.

If PJM fails, it is not a regional inconvenience. It is a national emergency. The territory PJM covers includes the financial infrastructure of New York and New Jersey, the federal government in Washington DC, the largest concentration of data centers on Earth in Northern Virginia, major military installations, critical manufacturing corridors, and transportation hubs that connect the Eastern Seaboard to the Midwest. A sustained grid failure in PJM territory would cascade across interconnected systems and affect the entire Eastern Interconnection — which covers everything east of the Rockies.

The grid was designed for a world of stable baseload demand — predictable load growth of 1-2% per year, supplied by a fleet of large coal, nuclear, and gas plants that ran continuously. That world is gone. In its place: an unprecedented surge in demand driven by data center construction, electrification of transportation and buildings, industrial reshoring, and the AI revolution — all colliding with the accelerating retirement of the thermal generation fleet that kept the system reliable for decades.

Understanding what is happening inside PJM is not optional for anyone working in data center infrastructure, energy policy, or critical facilities engineering. The decisions being made right now — about capacity markets, interconnection queues, plant retirements, and behind-the-meter generation — will determine whether the Eastern United States has enough electricity to power the AI revolution, or whether it stumbles into rolling blackouts within the next three to five years.

The 6 GW Gap — Anatomy of a Shortfall

In January 2024, PJM released a load forecast that sent shockwaves through the energy industry. After more than a decade of essentially flat demand — where load growth was so minimal that most planning models assumed it would continue indefinitely — PJM projected demand growth of approximately 40 GW by 2039. That is not a gentle uptick. That is the equivalent of adding the entire generating capacity of a mid-sized European country onto a grid that had been planning for zero growth.

The immediate concern is not the 2039 horizon. It is the near-term window of 2027-2028, when multiple stress factors converge. PJM's reserve margin — the percentage of excess supply above the forecasted peak demand — has been steadily declining. The target reserve margin is 14.7%, the level PJM considers necessary to maintain reliability with a loss-of-load expectation of one event in ten years. By the mid-2020s delivery years, the actual reserve margin under stress scenarios approaches or breaches that threshold.

The arithmetic is brutally simple. On the supply side, approximately 40 GW of thermal generation — primarily coal, but also aging natural gas plants — face retirement risk by 2030 due to a combination of EPA regulations, unfavorable economics, and age-related mechanical decline. On the demand side, data centers, electrification, and industrial load are driving growth rates that PJM has not seen in a generation. The gap between what is retiring and what is being built to replace it is the 6 GW shortfall that has regulators, grid operators, and capacity market participants deeply concerned.

The NERC Long-Term Reliability Assessment has flagged the PJM region as facing "elevated risk" for resource adequacy. NERC — the North American Electric Reliability Corporation — is the regulatory authority responsible for reliability standards across the continent. When NERC identifies a region as elevated risk, it means the organization's own models show that the area may not have sufficient resources to meet demand under plausible stress scenarios. This is not a fringe analysis. This is the official reliability watchdog raising a red flag.

But the most sobering data point is not a forecast — it is something that already happened. During Winter Storm Elliott in December 2022, PJM experienced 46 GW of forced outages at peak demand. Forty-six gigawatts. That is more than a quarter of PJM's installed capacity going offline simultaneously during a period of extreme cold and high demand. PJM came within a razor-thin margin of ordering rolling blackouts across its entire footprint — something that has never happened in the organization's history.

Elliott exposed a fundamental vulnerability: the grid's reliance on natural gas generation that failed when gas supply was constrained by the same cold weather driving electricity demand. Gas plants accounted for a disproportionate share of the forced outages because pipeline capacity was prioritized for heating, gas prices spiked, and some plants lacked firm fuel supply contracts. The storm demonstrated that PJM's capacity market was procuring megawatts that did not actually show up when they were needed most.

The 6 GW shortfall projection is not theoretical. It is the mathematical consequence of watching supply leave the system faster than replacement supply can interconnect, while demand accelerates in a direction nobody planned for even five years ago. Elliott proved that even the existing system — before the shortfall materializes — is more fragile than anyone assumed.

Data Centers Are Eating the Grid

No single driver of PJM's capacity crisis is more concentrated, more aggressive, or more geographically specific than the explosion of data center construction. Data centers now account for 30-40% of PJM's projected load growth — a share that would have been inconceivable a decade ago. And the epicenter of this demand is a stretch of Northern Virginia that the industry calls Data Center Alley.

Ashburn, Virginia, and the surrounding Loudoun County corridor represent the single largest concentration of data centers on Earth. More than 300 data centers occupy over 25 million square feet of whitespace in this region. Ashburn is the physical nexus of the modern internet: it hosts the majority of the world's internet exchange traffic, the primary cloud regions for AWS, Microsoft Azure, and Google Cloud, and now the GPU-dense AI training clusters that are reshaping the industry's power demands. Every major hyperscaler, colocation provider, and AI company either operates here or is actively trying to build here.

The utility serving most of this territory is Dominion Energy Virginia, a subsidiary of Dominion Energy. Dominion's numbers tell the story more starkly than any industry report. In 2023, Dominion projected data center load demand at approximately 3 GW. By its latest filings, that projection has ballooned to over 13 GW by 2038 — a fourfold increase in just a few years of forecast revisions. The growth is not slowing. Each quarter brings new upward adjustments as hyperscalers announce larger campuses, higher power densities, and more aggressive deployment timelines.

The broader PJM interconnection queue tells a similar story of overwhelming demand colliding with inadequate infrastructure. As of early 2024, the PJM queue contained over 250 GW of pending generation and storage projects — primarily solar, wind, and battery storage. That sounds encouraging until you examine the completion rate: fewer than 5% of projects that enter the PJM queue ultimately reach commercial operation. The average time from queue entry to interconnection is 4-5 years, and that timeline has been growing as the queue itself has become congested with speculative applications.

The interconnection queue bottleneck is one of the most underappreciated structural problems in American energy policy. Building a power plant or a large-scale solar farm requires interconnection studies, transmission upgrades, environmental permitting, land acquisition, equipment procurement, and construction — a process that routinely takes 5-10 years from conception to energization. Data centers, by contrast, can be designed, built, and commissioned in 18-24 months. The temporal mismatch is staggering: the demand arrives in years, but the supply infrastructure takes a decade.

This creates a paradox that defines the current crisis: data centers need power now, but the grid infrastructure to provide that power cannot be built for years. The hyperscalers are not going to wait. They are building campuses today, signing leases today, ordering GPU clusters today. When those facilities come online and need power, they will either get it from the existing grid — potentially at the expense of other users — or they will find creative workarounds like behind-the-meter generation, direct nuclear co-location, or relocating to regions with more available capacity.

Every megawatt that a data center consumes from the PJM grid is a megawatt that is no longer available for homes, hospitals, factories, and other critical loads — unless new generation is built to match. And right now, new generation is not keeping pace. The market signal is clear: Northern Virginia is approaching a physical limit on how many data centers the existing grid can support. The question is whether the industry will respect that limit or continue building until the grid itself forces the answer.

The Auction Shock — $14.7 Billion Wake-Up Call

In July 2024, PJM held its Base Residual Auction (BRA) for the 2025/2026 delivery year — the annual capacity market clearing event where generation resources commit to be available during future peak demand periods in exchange for capacity payments. The results sent a shockwave through the energy industry. The clearing price for most of PJM's territory came in at $269.92 per MW-day. The previous auction had cleared at $28.92 per MW-day. That is an 800% increase in the price the market assigns to keeping a megawatt of generation capacity available.

The total procurement cost of the auction jumped from approximately $2.2 billion to $14.7 billion. In one auction cycle, the cost of ensuring resource adequacy for PJM's territory increased by $12.5 billion. These are not abstract market numbers. They flow directly through to electricity bills for every residential, commercial, and industrial customer in PJM's footprint. The estimated impact on residential customers is $2-5 per month. For large commercial and industrial users — including data centers — the increases are proportionally larger, potentially adding millions to annual operating costs.

The auction price spike was driven by the convergence of every factor discussed in this article. Tighter supply due to plant retirements. Higher demand forecasts driven by data centers and electrification. Updated risk modeling that reflects the lessons of Winter Storm Elliott, where PJM tightened its performance requirements and accreditation rules for generation resources. The market was doing exactly what it was designed to do: signaling that supply is scarce relative to demand and that more investment in generation is needed.

But the signal has a problem: it arrives too late. The capacity auction signals that more generation is needed, but the lead time to build new generation — even natural gas plants, which are among the fastest to construct — is 3-5 years. For renewables with storage, it is longer. For nuclear, it is a decade or more. The auction price tells the market that supply is insufficient, but it cannot accelerate the physical timeline of construction, permitting, and interconnection. The price spike is a fire alarm going off in a building where the fire exits are years away from being built.

For data center operators, the auction results carry a specific warning. Capacity costs are a pass-through component of electricity rates. Higher auction clearing prices mean higher electricity costs for every data center connected to the PJM grid. Hyperscalers with long-term power purchase agreements (PPAs) may be partially insulated, but colocation providers and smaller operators face direct exposure. The days of cheap, abundant electricity in PJM territory — a key competitive advantage that made Northern Virginia the data center capital of the world — may be coming to an end.

The political fallout from the auction has been immediate. State regulators, consumer advocates, and legislators are scrutinizing whether the capacity market is functioning as intended or whether structural flaws are driving prices higher than necessary. Some argue the price spike reflects genuine scarcity and is a necessary investment signal. Others contend that market design issues — including the treatment of renewable resources, the accreditation methodology for intermittent generation, and the barriers to entry in the interconnection queue — are artificially constraining supply and inflating prices.

Regardless of the debate over market design, the fundamental physics has not changed: PJM needs more power than it has, the gap is widening, and the market just priced that reality at $14.7 billion per year. This is the market screaming that supply cannot keep up with demand.

40 GW of Retirements — The Coal Cliff

The supply side of PJM's crisis is dominated by a single, irreversible trend: the retirement of thermal generation at a pace that the system was never designed to absorb. Approximately 40 GW of coal and aging natural gas generation face retirement risk by 2030. This is not a projection based on optimistic environmental modeling. It is the consequence of economics, regulation, and mechanical reality converging on a fleet of power plants that are, in many cases, 40 to 60 years old.

Coal plants — once the backbone of PJM's generation fleet — are closing under a cascade of EPA regulations. The Mercury and Air Toxics Standards (MATS) imposed costly pollution controls that many older plants could not justify economically. The Good Neighbor Plan, designed to reduce cross-state air pollution, added further compliance burdens. Proposed carbon emission rules for existing power plants would make continued operation of unabated coal generation effectively impossible by the end of the decade. Each regulation alone might be manageable. Stacked together, they create a compliance cost that exceeds the market revenue these plants can earn.

Natural gas plants face a different but related calculus. Many of the gas plants in PJM's fleet were built during the combined-cycle construction boom of the early 2000s. These plants are now 20-25 years old, approaching the point where major capital expenditures are required for turbine overhauls, heat recovery steam generator repairs, and control system upgrades. For plants that have not been earning strong capacity or energy market revenues, the economics of refurbishment do not pencil. It is cheaper for the owner to retire the plant and redeploy the capital elsewhere than to spend $50-100 million on extending the life of a 25-year-old facility.

The critical nuance that is often lost in discussions about the energy transition is that not all megawatts are created equal. PJM uses a metric called Effective Load Carrying Capability (ELCC) to determine how much reliable capacity a generation resource actually contributes to system adequacy. A 1,000 MW coal or nuclear plant that can run 24/7 in all weather conditions receives a capacity accreditation close to its full nameplate rating. A 1,000 MW solar farm, which produces power only during daylight hours and at reduced output during cloudy conditions, receives an ELCC value that is a fraction of its nameplate — typically 30-50% for solar in PJM's territory, depending on the specific ELCC methodology applied.

This means that replacing 1 GW of retiring coal with 1 GW of nameplate solar capacity does not yield 1 GW of reliable replacement capacity. Depending on the ELCC values, it might yield 300-500 MW of capacity credit. To replace the full capacity contribution of a retiring coal plant, PJM may need 2-3x the nameplate capacity in solar, plus battery storage to cover the hours when the sun is not shining. The math works — but it requires substantially more investment, more land, more interconnection capacity, and more time than a like-for-like replacement.

And here is the structural bind: the PJM interconnection queue has 250+ GW of renewables and storage pending, but fewer than 5% of those projects ultimately get built. The queue is clogged with speculative applications, many of which lack firm financing, site control, or viable interconnection paths. PJM has been reforming its queue process — moving from a first-come, first-served model to a clustered study approach — but the reforms take years to implement and years more to work through the backlog of existing applications. Meanwhile, the retirements are happening on a fixed schedule driven by regulation and economics.

The result is a grid that is getting thinner every year. Not weaker in absolute terms — the lights still work today — but thinner in the sense that the margin between adequate supply and shortage is narrowing. Every winter storm, every summer heat wave, every unexpected plant trip tests a system that has less slack than it used to. Winter Storm Elliott was a preview of what happens when the margin runs out. The 40 GW of pending retirements will make that margin thinner still, while the demand from data centers and electrification will make the demand peaks higher. The question is not whether the system will be stressed. The question is whether it will be stressed past the breaking point.

The Nuclear Lifeline

In the search for generation that can match the demand profile of data centers — 24/7 operation, zero carbon, high reliability, and massive scale — one technology stands alone: nuclear power. And the market is responding accordingly. The two most significant power procurement deals of 2024-2025 both involve nuclear plants in PJM territory, both are driven by data center demand, and both represent a fundamental rethinking of how big tech companies secure their energy supply.

The Microsoft-Constellation deal is remarkable for its sheer ambition. Restarting a nuclear reactor that has been shut down for nearly a decade requires relicensing with the Nuclear Regulatory Commission, extensive equipment inspection and refurbishment, fuel procurement, staffing of a full reactor operations team, and integration with PJM's transmission system. Constellation estimates the restart will cost approximately $1.6 billion and take until 2028 to complete. The 20-year PPA with Microsoft provides the revenue certainty that makes this investment viable — a reactor restart that market economics alone could not justify becomes feasible when a creditworthy offtaker commits to two decades of power purchases at a premium price.

The Amazon-Talen deal, however, triggered a more contentious regulatory response. Amazon's original plan involved an Interconnection Service Agreement (ISA) modification that would have allowed the Susquehanna data center campus to draw power directly from the nuclear plant behind the meter — effectively bypassing PJM's transmission system and capacity market for a portion of the plant's output. In November 2024, FERC rejected this arrangement, citing concerns about cost-shifting to remaining PJM ratepayers.

The FERC rejection exposed a fundamental tension in the data center nuclear strategy. When a large load like a data center co-locates with a generator behind the meter, that load no longer pays for the shared transmission system that all other PJM customers use. The generator's output that was previously available to serve the broader grid is now dedicated to a single customer. The remaining ratepayers must absorb a larger share of transmission costs while having access to less generation capacity. FERC determined that this arrangement constituted an unjust and unreasonable cost shift.

The debate over behind-the-meter nuclear co-location is far from settled. Proponents argue that data centers bringing new demand should be allowed to secure dedicated generation, especially zero-carbon generation, without being forced to rely on a grid that may not have sufficient capacity. Opponents counter that PJM's grid is a shared resource, and allowing large loads to cherry-pick the best generators undermines the economics and reliability of the system for everyone else. Both arguments have merit, and the resolution will likely require new regulatory frameworks that balance grid reliability, cost allocation, and the legitimate need for firm, clean power to support AI infrastructure.

Looking further ahead, Small Modular Reactors (SMRs) represent a potential long-term solution for dedicated data center power. Companies like NuScale, X-energy, and Kairos Power are developing reactor designs in the 50-300 MW range that could be sited directly at or near data center campuses. However, the first commercial SMR deployments are not expected until 2030 at the earliest, and more likely 2032-2035 for widespread availability. For the immediate crisis window of 2027-2028, SMRs arrive too late.

Nuclear power is the only zero-carbon baseload technology that operates at utility scale, 24 hours a day, 7 days a week, 365 days a year, in all weather conditions. It is exactly what data centers need. But restarting shuttered reactors takes years, building new ones takes a decade, and the regulatory framework for integrating nuclear with data center load is being invented in real time. The nuclear lifeline is real — but it may not arrive in time to close the 6 GW gap that is opening right now.

The Political Battlefield

The PJM capacity crisis is not merely a technical problem. It is a political war fought on multiple fronts, with utilities, data center operators, environmentalists, grid operators, federal regulators, state commissions, and local communities all pulling in different directions. The fundamental tension is brutally simple: everyone wants reliable power, nobody wants to pay for it, and nobody wants to live next to the infrastructure that provides it.

Utilities like Dominion Energy and American Electric Power sit on one side of the battlefield. They want data centers to pay for the grid upgrades their massive load growth demands. When a single hyperscale campus requires 300 MW of dedicated transmission capacity, the cost of substation upgrades, new transmission lines, and interconnection infrastructure can exceed $500 million. Utilities argue that these costs should fall on the customers creating the demand, not be socialized across millions of residential ratepayers. Data center operators counter that they already pay billions in property taxes and energy bills, and that forcing them to fund grid infrastructure creates a competitive disadvantage against regions with faster, cheaper interconnection processes.

Environmentalists and reliability hawks represent another axis of conflict. The Sierra Club and allied environmental organizations oppose any new natural gas generation, arguing that building gas plants to power AI workloads locks in decades of carbon emissions and undermines climate targets. Grid operators and reliability planners at PJM and NERC counter that renewables alone cannot fill the capacity gap in time. Solar produces nothing at night. Wind is intermittent. Battery storage at the scale needed — tens of gigawatts — does not yet exist. The uncomfortable reality is that maintaining grid reliability through the 2027-2030 transition period almost certainly requires some new gas generation as a bridge, regardless of its climate implications.

State regulators add another layer of complexity. The Virginia State Corporation Commission is scrutinizing Dominion Energy's data center load projections, questioning whether the utility is inflating future demand to justify massive capital spending programs that earn regulated returns for shareholders. Virginia hosts the largest concentration of data centers in the world — over 300 facilities in Loudoun County alone — and the SCC's decisions directly impact whether new generation and transmission projects get approved and at what cost. Meanwhile, the Ohio Public Utilities Commission is debating coal plant subsidies, with some commissioners arguing that keeping aging coal plants online is necessary for reliability even as those plants become increasingly uneconomic.

Data center tax incentives have become a flashpoint for local opposition. Loudoun County, Virginia, which generates over $600 million annually in data center property tax revenue, now faces organized resident opposition to new facilities. Citizens complain about noise from backup generators and cooling systems, visual blight from massive windowless structures, water consumption in drought-prone regions, and the conversion of farmland and residential areas into industrial zones. The tax revenue argument that once silenced opposition is losing its potency as residents calculate that data centers generate far fewer local jobs per dollar of investment than almost any other commercial development.

The political battlefield has no clean resolution. Data centers need power that does not yet exist. Utilities need cost recovery that ratepayers resist. Environmentalists need emissions reductions that conflict with reliability requirements. Regulators need to balance competing mandates with incomplete information and political pressure from all sides. The only certainty is that the decisions made in the next 24 months by FERC, state commissions, and PJM's board will determine whether 65 million people keep the lights on or experience the first major reliability crisis of the AI era.

What Other Grids Can Teach PJM

PJM does not operate in isolation. Other major grid operators across North America and internationally face similar pressures from data center load growth, plant retirements, and the energy transition. Examining how these grids have responded — and where they have succeeded or failed — offers critical lessons for PJM's path forward.

ERCOT, the Texas grid operator, faces aggressive data center load growth but operates under a fundamentally different market design. Texas uses an energy-only market with no capacity payments, relying instead on scarcity pricing during peak demand to incentivize new generation investment. This design has attracted massive solar and battery deployment: Texas added over 10 GW of utility-scale solar and 5 GW of battery storage between 2022 and 2025. However, ERCOT's energy-only design also produced Winter Storm Uri in February 2021, when extreme cold caused cascading generator failures, load shedding for 4.5 million customers, 246 deaths, and an estimated $195 billion in economic damage. The lesson for PJM: market design matters enormously, and the capacity market PJM uses provides a reliability backstop that ERCOT lacked.

CAISO, California's grid operator, faces summer stress events annually but has pursued the most aggressive battery storage buildout in North America. California now has over 12 GW of grid-scale battery capacity, enough to shift significant solar generation into evening peak hours. CAISO's experience demonstrates that battery storage at scale is technically viable and operationally reliable — but California's mild winters and massive solar resource are advantages PJM does not share. PJM's winter peak vulnerability requires resources that can sustain output for days, not the 4-hour duration typical of current lithium-ion installations.

MISO, the Midcontinent grid operator, has also been flagged by NERC for capacity adequacy concerns. MISO's 2024 capacity auction in its southern zone cleared at the price cap, signaling potential shortfalls ahead. Like PJM, MISO faces coal retirements and growing demand, though its data center load growth is less concentrated than PJM's.

Internationally, the approaches vary dramatically. Singapore imposed a moratorium on new data center construction from 2019 to 2022 to protect grid stability, lifting it only after implementing strict efficiency requirements: new facilities must achieve a PUE below 1.3 and source a significant share of their energy from renewables. The European Union has proposed mandatory reporting requirements for data center energy consumption and is considering green power purchase mandates for new facilities. Ireland, where data centers now consume approximately 20% of national electricity, has implemented planning restrictions in the Dublin region to prevent further grid strain.

The international experience reveals a spectrum of policy responses, from market-based incentives to outright moratoria. The common thread is that no grid anywhere in the world has solved the problem of accommodating exponential data center growth without either constraining that growth or massively expanding generation and transmission capacity. PJM has the least time of any major grid to find its answer: the 6 GW gap opens in 2027, and no policy intervention or market reform can build power plants that fast.

What Happens If PJM Fails

The analysis to this point has been clinical: gigawatts of capacity, reserve margins in percentages, auction clearing prices in dollars per megawatt-day. But the consequences of a PJM capacity shortfall are measured in human terms that no spreadsheet can capture. If PJM cannot maintain adequate reserves during a peak demand event, the result is controlled load shedding — the industry term for rolling blackouts that affect millions of people simultaneously.

Consider the scenario: a polar vortex event in January 2028, with temperatures dropping below 0°F across PJM's mid-Atlantic footprint. Heating demand surges. Gas pipelines, already constrained by high demand for building heating, cannot deliver sufficient fuel to gas-fired power plants. Wind generation drops as turbines reach cold-weather cutoff thresholds. Solar output is minimal during the short winter days. PJM dispatches every available generator, calls on demand response resources, and requests emergency energy imports from neighboring grids. It is not enough. The reserve margin drops below zero. PJM orders utilities to implement rolling blackouts to prevent an uncontrolled grid collapse.

The impact is immediate and severe. Hospitals switch to backup generators, but those generators are designed for hours of operation, not days. Fuel deliveries depend on a transportation network that itself runs on electricity — traffic signals, fuel pumps, rail switching systems. Water treatment plants lose pumping capacity, forcing boil-water advisories for millions. Traffic signals go dark across major metropolitan areas. Cell towers exhaust their backup batteries within 8-12 hours, degrading communications networks precisely when they are needed most.

Cascading failures amplify the damage. PJM is interconnected with neighboring grid operators — MISO to the west, NYISO to the north, Duke/Southern to the south. A major disturbance in PJM can propagate across these interconnections, forcing neighboring grids to shed load to protect their own systems. The 2003 Northeast Blackout demonstrated this cascading dynamic: a software bug and untrimmed trees in Ohio triggered a cascade that blacked out 55 million people across eight states and Ontario. PJM's interconnected nature means that its reliability problems are everyone's reliability problems.

The economic cost of a major PJM outage is staggering. The U.S. Department of Energy estimates that power outages cost the American economy $150 billion annually in aggregate, with individual major events causing $1-2 billion per day in economic losses across the affected footprint. The 2021 Texas Winter Storm Uri — the most relevant analog for a PJM winter failure — caused an estimated $195 billion in total economic damage, 246 confirmed deaths (with excess mortality estimates exceeding 700), and left 4.5 million customers without power for up to 5 days in freezing conditions. PJM's footprint has 3.5 times the population of ERCOT's. Scale the Uri impact proportionally, and a comparable PJM failure would be the most expensive infrastructure disaster in American history.

Grid Reliability & DC Power Impact Analyzer

To quantify the risks discussed throughout this analysis, I built an interactive tool below. Input your assumptions about grid capacity, demand growth, and retirements, and the analyzer will calculate reserve margins, capacity surplus or deficit, blackout risk levels, and estimated residential rate impacts. The defaults reflect PJM's actual parameters based on publicly available data.

The Engineer's View

I write this analysis not as an outside observer but as someone who operates critical infrastructure that depends on the grid every second of every day. My facility runs on PJM power. My UPS systems, my generators, my cooling plants, my network infrastructure — all of it ultimately traces back to the transmission lines and generation resources that PJM coordinates. When PJM's reserve margin shrinks, my operational risk grows. This is not abstract to me.

UPS systems and backup generators buy you hours, not days. A typical data center can ride through a brief grid disturbance on battery backup and sustain operations for 24-48 hours on diesel generators, assuming fuel deliveries continue. But if PJM enters a prolonged shortage event — multiple days of extreme weather combined with insufficient generation — no amount of on-site backup makes you immune. Grid reliability is our reliability. There is no version of "resilient data center operations" that survives a fundamentally unreliable grid over the medium term. We can harden our facilities, stockpile fuel, and test our failover procedures, but we cannot generate 50 MW of continuous power from a rooftop solar array or a battery room designed for 15 minutes of ride-through.

The data center industry needs to be honest about what it is asking the grid to do. We are requesting tens of gigawatts of new capacity in a timeframe that the grid has never been asked to deliver. We are requesting this capacity while simultaneously benefiting from the retirements that make room for our preferred clean energy sources. We are requesting fast interconnection while opposing the transmission projects that make interconnection possible. We are requesting low electricity rates while our load growth drives the very capacity shortfalls that push rates higher. The cognitive dissonance is unsustainable.

The path forward requires data centers to pay their fair share for the grid infrastructure they consume. Not subsidized interconnection. Not behind-the-meter arrangements that shift transmission costs to residential ratepayers. Not tax incentive packages that exempt the industry from the financial consequences of its own growth. Fair share means contributing proportionally to generation, transmission, and distribution costs based on the load we impose and the reliability we demand. If that makes a megawatt in Virginia more expensive than a megawatt in Iowa, then the market is working correctly, and some load will migrate to regions with more abundant capacity. That is not a failure of policy. It is the grid telling us where it can and cannot absorb growth. We should listen.