The US has seen almost zero electricity generation growth over the past two decades, even as oil and natural gas production surged, largely due to regulatory complexity and political choices that prioritized shutting down fossil fuel plants over grid reliability. Now, AI-driven data center demand is forcing a dramatic reversal, with the country needing 15–20% annual energy growth to keep up.
Secretary of Energy Chris Wright and General Matter founder Scott Nolan discuss how the Trump administration is attempting to shift the US from “energy subtraction” to “energy addition” through regulatory reform, nuclear restarts, and rapid natural gas buildout.
The stakes are framed in existential terms: if the US doesn’t lead in AI, China will, and AI dominance translates directly into military, economic, and technological supremacy.
The Regulatory Reversal
The Biden administration had plans to close an additional ~100 GW of coal plants over five years while building only 22 GW of new firm generation, a net reduction of ~88 GW that would have made AI leadership, manufacturing reshoring, and lower electricity prices impossible.
The Trump administration stopped 17 GW of coal plant closures in its first year alone, which Wright argues likely prevented hundreds of deaths during a severe January 2025 cold snap (Storm Fern) that pushed the eastern grid to the ragged edge of blackout.
The administration is reforming NEPA (National Environmental Policy Act) implementation, which had been weaponized to block projects through litigation, and the Supreme Court unanimously supported this narrower interpretation.
Nuclear: Restarts and New Builds
Three nuclear restarts are underway, adding roughly 3 GW of existing capacity back to the grid:
Palisades in Michigan (~1 GW) relicensed and expected back online by late summer 2025.
Three Mile Island (renamed Crane Clean Energy Center) expected back online in 2026.
Duane Arnold in Iowa possibly back in 2028.
Four executive orders signed in May 2024 launched a fast-track program for 11 test reactors to go critical, with several expected before July 4, 2025—the first non-lightwater reactors turned on in the US in 40 years.
These test reactors are being observed by NRC staff embedded with DOE teams, allowing companies to begin the licensing process for commercial electricity generation in parallel.
First real commercial deployments are expected around 2028–2029, with meaningful scale-up in the early 2030s.
The Fuel Problem and General Matter’s Role
The US currently enriches only about a quarter of its own uranium needs, relying on a European consortium facility in New Mexico. To supply both the US and its allies would require roughly 8x current capacity; to meet a 4x nuclear buildout target by 2050 would require over 30x.
Scott Nolan’s General Matter is building uranium enrichment capacity in the US, with its first manufacturing plant expected online around 2030.
In the interim, the DOE is providing over 20 tons of HALEU (high-assay low-enriched uranium) fuel—derived from downblended weapons-grade uranium or plutonium-uranium mixtures—as a bridge supply.
Nolan argues the opportunity is for 10–100x growth in US enrichment capacity, drawing a parallel to how SpaceX went from ~1 launch per year to over 100 and captured 90%+ of global mass-to-orbit capacity.
Natural Gas: The Dominant Near-Term Source
For the next five years, natural gas will be the primary source of new dispatchable (24/7) electricity because it is the cheapest and fastest to build.
Gas turbines are in tight supply with 5–6 year lead times for new orders, but turbines already ordered and under construction will deliver close to or over 100 GW over the next five to six years.
Beyond turbines, gas reciprocating engines (similar to large diesel generators) and fuel cells are emerging as significant new sources of electricity from natural gas, with multiple gigawatts expected within five years.
The fuel itself is not a constraint: the US has so much associated natural gas in the Permian Basin that pipeline shortages are curtailing ~2.5% of production—enough to generate 15–20 GW of electricity if it could be transported.
The cost of electricity is driven primarily by the cost of the machine (turbine, reactor, etc.), not the fuel, which is why scaling up manufacturing of these machines is critical to lowering costs.
The Grid and Peak Demand Reality
What matters for grid reliability is not average generation capacity but deliverability at peak demand moments. The US currently has hundreds of gigawatts of spare capacity on mild days, but peak summer and winter demand absorbs nearly all of it.
During the January 2025 peak in New England, wind, solar, and batteries combined provided only 2% of electricity. Nuclear, natural gas, and coal kept the lights on, with oil-burning plants as the actual number one source because pipeline constraints diverted gas to home heating.
Transmission is a major bottleneck: building new power lines is extremely slow and expensive. The DOE is pushing “reconductoring”—replacing old conductors with higher-capacity ones on existing lines—to move more electrons without new permits.
FERC recently moved to reduce regulatory burden on natural gas plant and pipeline permitting.
Data Center Economics and Scale
A traditional 1 GW data center costs ~$10 billion ($10/watt); advanced AI data centers cost $40–60/watt. Harnessing even 100 GW would require trillions of dollars in capital.
A single gigawatt of compute can generate ~$25 billion in revenue for an AI company like Anthropic, creating enormous incentive to find power wherever it exists.
Over the past decade, companies (first in Bitcoin mining, then AI) have been finding isolated pockets of cheap energy supply, but most of those opportunities are now exhausted, meaning all new power must be built from scratch.
Winning Over Communities
Public opposition to data centers and AI is significant and growing, driven by fears of rising electricity prices and water usage, mirroring earlier anti-fracking campaigns.
The Trump administration’s “Ratepayer Protection Pledge” has all major hyperscalers committing to pay above-market rates for power, fund transmission upgrades, and in many cases build 1.2 GW of generation for every 1 GW they consume—adding net capacity to local grids.
Multi-year rate freezes (in nominal dollars, meaning real declines) are becoming standard in data center deals. States with fast demand growth like Texas and Nebraska have seen declining electricity prices, while states that have shrunk generation like California and New York have the fastest-rising prices.
Water concerns are being addressed through closed-loop cooling systems that are filled once and don’t consume ongoing water.
Nolan suggests data centers should aim to be net energy producers (building 1.1x their needs), pitching communities on lower rates rather than neutrality.
Behind-the-Meter and Grid Interconnection
Many data centers will generate power on-site (“behind the meter”), but both Wright and Nolan argue that grid interconnection is still economically superior even for self-generating facilities.
Being connected allows data centers to curtail their own load during grid peak stress and sell power back at high-value moments—potentially just a few hours per year—earning significant revenue while helping stabilize rates.
Nuclear plants paired with data centers may operate more independently, but gas turbine setups benefit enormously from grid connection for reliability without massive redundant infrastructure.
Long-Term Scaling: 2030–2050
The long-term vision involves factory-built nuclear reactors, following the pattern that everything built in factories gets cheaper and faster while stick-built construction gets more expensive.
The Vogtle 3 and 4 plants in Georgia were expensive precisely because the US hadn’t built a reactor in decades and had to stand up a supply chain for just two units. Scale is the path to lower costs.
Nolan draws a direct parallel to SpaceX: the first Falcon 1 launch was years of work, but by year 20 the company was doing 100+ launches per year. The early stage of any exponential curve looks linear.
General Matter is planning multiple enrichment facilities, building core-and-shell structures in advance of equipment installation, and acting as its own EPC (engineering, procurement, construction) firm—following the Tesla Gigafactory model of owning the entire project to avoid cost and schedule overruns.
Wright notes that even Democrats and Democratic governors are increasingly supporting nuclear and natural gas expansion, suggesting the political consensus is shifting away from the “energy subtraction” mindset.
The China Comparison
China has tripled its electricity grid since 2010 while the US has stayed flat, but China’s growth has been almost entirely coal-dependent, and China is now by far the world’s largest importer of oil, natural gas, and coal—making it structurally energy-dependent in ways the US is not.
The US’s problem is bureaucratic and fixable; China’s is a fundamental resource dependency.
Wright argues that if the US doesn’t serve the exponential growth in AI energy demand, China will—and will do so in a way that is less safe, less clean, and less aligned with democratic values.
Advice for Founders and Entrepreneurs
Wright encourages entrepreneurs building physical things in America to engage with the DOE, which has land, labs, and a mandate to permit projects quickly. The administration wants to eliminate the uncertainty that has driven manufacturing overseas for 30 years.
Nolan’s formula for deciding what to work on: find a critically important problem that won’t get solved unless someone new tackles it, and where your skills overlap with the need. If all three boxes are checked, it becomes a moral imperative.
Both emphasize that the next few decades will be about the physical world—energy, manufacturing, construction—after decades dominated by software, and that the low-hanging fruit in the physical world is enormous.
The key mindset shift: build first, demand will follow. Just as SpaceX built launch capacity before the satellite market existed, energy companies need to scale production ahead of confirmed demand curves.
