Having spent the past decade focused on energy and new mobility as an investment banker, and from the research for my forthcoming book, The Displacement Machine, I’ve watched the collision between AI’s energy demands and U.S. energy policy recently unfold in slow motion. This article is my attempt to lay out the contradiction as clearly as I can, because the window for correction is now measurable in days, not years. In 99 days, the last federal tax credits for new wind and solar projects effectively expire. Meanwhile, Washington is scrambling to manage a data centre energy crisis it helped create. Here’s what you need to know:

KEY TAKEAWAYS:

• The OBBBA’s July 4, 2026 deadline gives wind and solar developers 99 days to begin construction or lose tax credits entirely
• Trump’s Ratepayer Protection Pledge (March 4) acknowledges the crisis but is voluntary and fuel-agnostic
• The bipartisan GRID Act and Warren-Hawley push for mandatory reporting signal real political urgency
• China’s 15th Five-Year Plan targets 120 GW of new wind annually; it installed 315 GW of solar in 2025 alone
• PJM capacity auctions have hit price caps three years running; data centers drove $21.3 billion in costs across the last three auctions
• The Dallas Fed finds data centers will add 0.05-0.13 percentage points to PCE inflation through 2030
• Efficiency gains (Jevons Paradox) will not offset growing demand – the IEA still forecasts a doubling of data centre consumption by 2030

On July 4th, 2026, ninety-nine days from now, the last window for American wind and solar developers to begin construction and qualify for federal tax credits will slam shut. After that date, projects that miss the deadline must be fully operational by December 31st, 2027, or lose access to credits entirely. There is no phase-down. No partial reprieve. The economic scaffolding beneath America’s clean energy buildout will simply disappear.

This deadline, set by the One Big Beautiful Bill Act signed exactly one year earlier, arrives at a moment of cascading urgency. On March 4th, President Trump convened the CEOs of seven major technology companies at the White House to sign a “Ratepayer Protection Pledge” – an acknowledgment that data centers are driving electricity costs beyond what American households can bear. On March 24th, the Energy Information Administration launched its first-ever pilot surveys to track data centre energy consumption, across Texas, Northern Virginia, and Washington state. Two days later, Senators Warren and Hawley, in a bipartisan letter that would have seemed implausible a year ago, demanded the EIA establish mandatory annual reporting for data centers, with a response deadline of April 9th. And in February, the bipartisan GRID Act was introduced, which would require data centers consuming more than 20 megawatts to generate their own power entirely off-grid.

The message from Washington is unmistakable: the political class has recognized that AI’s electricity appetite is a crisis. What it has not reconciled is that the same government dismantled the primary tool for solving it.

The Scale of the Problem

The artificial-intelligence boom has created an insatiable appetite for electricity. Data centers consumed 415 terawatt-hours globally in 2024 and are on track to devour 945 TWh by 2030, according to the International Energy Agency – roughly equivalent to Japan’s entire power consumption. In America, they will drive nearly half of all electricity demand growth this decade. The Dallas Federal Reserve published research on March 5th showing that data centers alone will increase headline PCE inflation by 0.05 percentage points in 2026, rising to 0.13 by 2030 – a drag on consumer spending that falls hardest on lower-income households, for whom electricity constitutes a larger share of expenses.

The country that wins the race to power AI cheaply and abundantly will enjoy a structural advantage in the most consequential technology competition since the space race. By this logic, the United States should be scrambling to build clean, cheap generation capacity. Instead, the OBBBA methodically dismantled the incentive structure that had begun to deliver exactly that.

The timing could hardly be worse. Utilities requested $31 billion in rate increases in 2025, more than double the prior year. Goldman Sachs projects another 6% rise through 2027. PJM Interconnection’s capacity auctions, the mechanism by which future power supply is secured for 65 million Americans across 13 states, have become a barometer of crisis. The 2026-2027 auction cleared at $329 per megawatt-day, hitting the FERC-approved cap. The 2027-2028 auction reached $333, with the grid operator falling 6,625 megawatts short of its reserve margin target. Data centers were responsible for 40% of the $16.4 billion in costs from the most recent auction. Across the last three auctions, they have driven $21.3 billion in capacity charges – nearly half the total.

The Pledge and Its Limits

The Ratepayer Protection Pledge is the administration’s answer. Under its terms, hyperscalers commit to building, buying, or bringing all new generation capacity needed to power their data centers. They agree to fund transmission upgrades, negotiate separate rate structures with utilities, and pay for reserved capacity whether they use the electricity or not. The pledge even extends to colocation arrangements, preventing companies from circumventing obligations by leasing space from third-party providers.

As policy instincts go, it is reasonable. Those who create the demand should bear the cost. But a voluntary pledge signed by seven CEOs is not an energy strategy. It does not specify what kind of power these companies must build, nor does it address the timeline mismatch between data centre construction, which takes 18 to 24 months, and new generation capacity, which takes five to ten years. It sidesteps the deeper question entirely: if these companies must generate their own power, and the tax credits that made renewables and storage economically attractive are expiring in ninety-nine days, what exactly will they build?

The GRID Act, introduced by Senators Hawley and Blumenthal, goes further, requiring data centers over 20 megawatts to source power entirely off-grid, with a ten-year transition for existing facilities. But it too assumes that generation capacity will materialize. The legislation mandates the destination without funding the journey.

The Rival That Never Rests

China does not share this confusion. In March 2026, Beijing released its 15th Five-Year Plan, which targets 120 gigawatts of new wind capacity annually through 2030, more than 100 gigawatts of offshore wind by decade’s end, and 100 gigawatts of pumped hydro storage. Solar additions, though retreating from 2025’s record 315-gigawatt installation, are still projected at 180 to 240 gigawatts in 2026.

The numbers bear repeating: China’s single-year solar addition in 2025 exceeded the entire installed solar base of the United States. Its clean energy capacity, at 1,494 gigawatts, now surpasses its fossil-fuel capacity for the first time. It invested $818 billion in energy transition in 2024 (more than America and Europe combined) and holds 75% of all global clean-energy patents, up from 5% at the turn of the century.

China’s green juggernaut has problems of its own. Overcapacity has cratered solar-panel margins. Curtailment of renewable output remains endemic in some provinces. The new Five-Year Plan loosens carbon intensity targets rather than tightening them, and coal remains deeply embedded, with 3.17 billion tonnes consumed in 2025. But these are the growing pains of a country building too much clean energy too fast, a rather different problem from building too little of it too slowly.

Efficiency Is Not a Substitute for Abundance

Optimists point to breakthroughs in AI efficiency as a potential reprieve. In January 2025, Chinese startup DeepSeek demonstrated that frontier-quality models could be trained using a fraction of the energy and compute previously assumed. If AI can be made radically more efficient, the argument runs, perhaps the energy crunch is overstated.

The logic is seductive but flawed, for a reason economists have understood since 1865. William Stanley Jevons observed that when steam engines became more fuel-efficient, coal consumption did not fall, it soared, because efficiency made steam power economical for new applications. The same dynamic is already visible in AI. DeepSeek’s efficiency gains have lowered the cost of inference, expanding the market for AI applications and driving aggregate energy demand upward even as per-unit consumption declines. The IEA’s projections account for efficiency improvements and still forecast a doubling of data-centre demand by 2030.

Moreover, DeepSeek’s achievement carries a geopolitical sting. It was a Chinese company that demonstrated this breakthrough, using open-source methods the global AI community can now replicate. China is not merely building cheaper energy. It is also, in some domains, building more efficient AI. American firms face the worst of both worlds: rising energy costs at home and an increasingly capable rival abroad that faces neither constraint.

The Structural Contradiction

The United States now finds itself in an extraordinary position. Its technology companies lead the world in AI capabilities. Its government has declared AI dominance a national priority. Its grid operators are warning that demand will outstrip supply within years. Bipartisan coalitions in the Senate are scrambling to regulate data centre energy use. The EIA is launching emergency surveys. And the administration’s own Ratepayer Protection Pledge implicitly acknowledges that the current trajectory is unsustainable.

Yet the same legislature dismantled the primary mechanism for building the clean, cheap power generation that the entire strategy depends upon. And the deadline for the last projects to qualify is now ninety-nine days away.

Investor-owned utilities plan to spend $1.1 trillion on grid expansion between 2025 and 2029. But much of that spending is reactive, reinforcing aging transmission lines and adding peaking capacity, rather than the strategic buildout of next-generation energy systems that China is executing at scale. Brookings estimates a projected shortfall of 49 gigawatts through 2028, roughly 5% of total U.S. generation.

The Inflation Reduction Act, whatever its flaws, had catalyzed $282 billion in clean-energy investment in its first year, much of it flowing to Republican-held districts and rural communities. It was beginning to create a domestic manufacturing base for solar panels, battery cells, and wind turbines that could have reduced dependence on Chinese supply chains while simultaneously building the cheap generation capacity AI demands. The OBBBA does not merely slow this process. It reverses it. And the communities that benefited most from the IRA’s investments will be hit hardest by its rollback.

None of this is to suggest the IRA was perfectly designed. Tax credits are blunt instruments. Permitting reform (which the OBBBA does not meaningfully address) remains a greater bottleneck than financing for many projects. The case for nuclear energy, which enjoys bipartisan support through the ADVANCE Act, is strong and underexploited.

But the question is not whether the IRA was optimal. It is whether dismantling the country’s clean-energy incentive structure, at precisely the moment when energy abundance has become a prerequisite for technological competitiveness, serves the national interest. The evidence from the PJM auctions, from the Dallas Fed’s inflation research, from the bipartisan panic now visible in the Senate, suggests it does not.

The energy transition will proceed with or without American leadership. China has made its choice. In ninety-nine days, the last American developers still racing to break ground will learn whether their country has made its choice too, or whether it will continue to saw off the branch on which its AI ambitions sit.

Sources

• International Energy Agency, “Electricity 2025: Analysis and Forecast to 2027” (January 2025) -- global data centre electricity consumption (415 TWh in 2024, 945 TWh by 2030)
• Dallas Federal Reserve, “The Impact of Data Centers on the U.S. Electricity System” (March 5, 2026) -- PCE inflation estimates (0.05 to 0.13 percentage points)
• PJM Interconnection, Base Residual Auction Results (2025-26, 2026-27, 2027-28) -- capacity auction prices and data centre cost allocation
• White House, “Ratepayer Protection Pledge” signing ceremony (March 4, 2026)
• U.S. Energy Information Administration, pilot data centre energy surveys launched March 24, 2026 (Texas, Northern Virginia, Washington state)
• Senators Warren and Hawley, bipartisan letter to EIA demanding mandatory annual data centre reporting (March 26, 2026)
• Bipartisan GRID Act (introduced February 2026) -- mandatory off-grid generation for data centres >20 MW
• China National Energy Administration, 15th Five-Year Plan targets (March 2026) -- 120 GW annual wind additions
• China National Energy Administration, 2025 solar installation data -- 315 GW installed in 2025
• Congressional Budget Office / Joint Committee on Taxation, IRA clean-energy tax credit estimates ($282 billion over ten years)
• Brookings Institution, grid reliability projections -- 49 GW shortfall by 2028
• One Big Beautiful Bill Act (OBBBA), enacted 2025 -- July 4, 2026 begin-construction deadline for IRA tax credits