The Unforgiving Math of Power Demand

For fourteen years, the US utility sector was boring. That was a feature, not a bug. From 2007 through 2021, despite population growth and GDP expansion, electricity generation remained effectively flat. Efficiency improvements in appliances, industrial equipment, and lighting offset the natural growth of the economy. It was a period of stagnation for power generators, who were forced to rely on rate increases rather than volume growth to pad their bottom lines.

That era is dead.

We have entered a cycle of aggressive consumption that the grid is not currently built to sustain without significant friction. In 2025, US electricity generation surged 3.0% to a record 4.5 million GWh. This follows a surge in the prior year. The driver is not subtle: it is the brute-force energy requirement of AI data centers springing up like mushrooms, compounded by the increasing market share of electric vehicles.

This is not a story about green energy transitions or corporate sustainability pledges. It is a story about physics and supply and demand. When the load curve spikes, the grid does not care about ideology; it grabs the nearest available electron. In 2025, that meant dragging coal plants back from the brink of retirement.

The Efficiency Plateau and the Demand Shock

To understand the magnitude of a 3.0% jump in a single year, you have to look at the baseline. For over a decade, the US economy essentially decoupled growth from energy intensity. We got better at using power faster than we added new demand.

That decoupling has broken. The efficiency gains from swapping incandescent bulbs for LEDs have been maxed out. Now, we are layering on massive, energy-dense loads. Data centers are not normal commercial tenants; they run 24/7 with massive cooling requirements and computational loads that fluctuate wildly.

When you combine this with the electrification of the vehicle fleet, you create a demand shock that the supply side cannot meet with “planned” capacity additions alone. The result is a scramble for reliability.

The Resurrection of Coal

The most telling data point in the 2025 EIA numbers is not the growth of solar—which is expected—but the resilience of coal.

Coal power generation jumped 13% in 2025. In a market supposedly transitioning to zero carbon, a double-digit spike in coal burn represents a collision between fantasy and reality. Coal’s share of generation rose to 16.3%, up from a record low the prior year.

Why did this happen? It comes down to the marginal cost of reliability. While coal cannot compete with combined-cycle natural gas on efficiency or price under normal conditions, and it certainly cannot compete with the zero-marginal-cost fuel of wind and solar, it possesses one critical attribute: dispatchability.

When data centers demand firm power and the wind isn’t blowing in the Midwest, utilities fire up the coal boilers. The economic logic here is stark. If the choice is between missing load requirements (brownouts) or burning expensive, dirty coal, the utility burns the coal. The 13% jump indicates that the grid was under significant stress in 2025. We are scraping the bottom of the capacity barrel.

Natural Gas: The Workhorse Holds the Line

Natural gas generation dipped 3.3% but remains the undeniable king of the US grid, accounting for 40.2% of all power generated.

The mechanics of the shale revolution fundamentally altered the US power stack starting in 2009. We have an oversupply of domestic gas, and combined-cycle gas turbines (CCGT) offer thermal efficiencies around 65%, nearly double that of the aging coal fleet.

Gas sets the price in most wholesale markets. Even with the dip in total volume, the infrastructure of the US grid is built around the assumption of cheap, abundant gas. It is the bridge, the destination, and the backup plan all rolled into one. Until battery storage scales by orders of magnitude, natural gas is the only fuel source capable of balancing the intermittency of the growing renewable portfolio at scale.

The Renewables divergence: Solar vs. Wind

The renewable data presents a split screen.

Solar is in a true exponential growth phase. Generation surged 28.0% in 2025. This includes both utility-scale farms and distributed rooftop systems. Solar’s market share is now 8.6%, and when combined with wind, variable renewables control nearly 19% of the US power supply.

The economics of solar have become undeniable in specific geographies. In California, Texas, and the Southwest, the levelized cost of energy (LCOE) for solar is incredibly low. However, this creates the “duck curve” problem, where midday pricing collapses, only to spike when the sun sets—exactly when residential demand peaks.

Wind, by contrast, is maturing. Generation rose 2.8%, a figure that barely keeps pace with total demand growth. The prime acreage in the wind corridor (Texas, Iowa, Oklahoma) is increasingly saturated, and transmission constraints are preventing new projects from connecting to the grid efficiently. You can build all the turbines you want in the Dakotas, but if you can’t move that power to a data center in Virginia, it is economically useless.

The CapEx Super-Cycle

The most critical signal for investors and strategists is the halt in retirements.

Operators had planned to retire 12.3 GW of capacity in 2025. They only retired 4.6 GW. That means nearly 8 GW of “zombie capacity”—mostly old coal and gas plants—was kept on life support because the grid operators (ISOs and RTOs) refused to let them go. The Department of Energy even issued emergency orders to extend operations.

This is a sign of a capacity crunch. The buffer is gone.

In response, the industry is entering a massive capital expenditure cycle. Developers added 53 GW of new capacity in 2025 and are planning a staggering 86 GW for 2026. If executed, that would be a record addition.

However, “planned” capacity is not “operating” capacity. Supply chain constraints for transformers, switchgear, and skilled labor are real. Interconnection queues—the line to plug into the grid—are years long.

The Economic Reality

We are moving from a period of power abundance to a period of power tightness.

For businesses, this means the era of flat energy costs is over. The cost of maintaining aging coal plants, building new gas peakers, and overbuilding solar to compensate for capacity factors will be passed through to the ratepayer.

For the data center operators—the Amazon AWS, Microsoft Azure, and Google Cloud platforms of the world—this is the primary bottleneck. Capital is abundant; power is scarce. We are already seeing these entities attempting to bypass the utility model entirely, looking to co-locate with nuclear plants or build behind-the-meter generation.

The logic is simple: If you cannot get power from the grid, you build your own. But for the rest of the economy, the 2025 data is a warning shot. The reserve margins are thin, the demand is rising, and the only short-term fix is burning more coal. Welcome to the new energy reality.