From Constrained to Nuclear-Powered, Energy Realities of EVs, AI, and America’s 2025 Grid

In Brief

• Surging AI workloads and EV adoption are accelerating U.S. electricity demand, revealing critical infrastructure constraints and forcing a strategic reassessment of grid capacity.
• Recommissioning dormant nuclear assets and advancing small modular reactors (SMRs) provide the most immediate and scalable path to reliable, carbon-free baseload power.
• A cohesive Revive–Rebuild–Reshape framework integrates nuclear generation, intelligent grid orchestration, and demand flexibility to ensure energy resilience and digital continuity.

Escalation in Twelve Months – Scorecard and Pain Point

Twelve months ago, in our article, “Power Crisis – Demands of Electric Vehicles and AI,” we signaled that America’s power system might crack under the twin surges of AI compute and electric-vehicle adoption. Since then, the data-center curve has steepened sharply. An updated Electric Power Research Institute study now projects that U.S. facilities, driven by generative-AI clusters, could draw 8%-10% of all domestic electricity by 2030, more than double their 2023 share.  Globally, the International Energy Agency’s 2025 outlook projects data-center demand to exceed 945 TWh by the end of the decade, roughly equivalent to Japan’s current annual consumption, with AI responsible for most of that growth.

Uneven Electric-Vehicle Demand – But It Remains Substantial

California, the bellwether market, crossed the two-million zero-emission-vehicle threshold in late 2024 and is still targeting 12.5 million EVs on the road by 2035. Nationwide, slower uptake in several heartland states has trimmed the most aggressive load scenarios; yet, the Department of Energy’s late-2024 revision still lifts the 2028 U.S. electricity forecast by ≈ 325-580 TWh once data-center, electrification-of-process-heat, and crypto-mining loads are included.

At the same time, infrastructure slack is vanishing. A National Infrastructure Advisory Council report indicates that large-power-transformer construction lead times range from 80 to 210 weeks, with unit prices up to four times pre-pandemic levels, resulting in delays that hinder new generation and storage projects before they can be interconnected. In Texas, ERCOT now forecasts that peak demand could increase by approximately 75% by 2030, driven primarily by data center growth, prompting legislation to regulate large, flexible loads.  Similar stress surfaced on the West Coast, where the California ISO’s 2024 heatwave analysis showed reserve margin compression despite a record 6.3 GW of new capacity being added the previous year.

Scorecard. Of the three structural pressures we highlighted last year – runaway demand, aging assets, and supply-chain fragility – two have worsened decisively, while the third (EV load) has merely decelerated. Aggregate U.S. electricity requirements for 2030 are now 600-800 TWh higher than planners assumed a year ago, despite the tightening of the ability to connect new resources. With headroom gone, the question is no longer whether the grid can stretch, but what breaks first. The following section examines the systemic constraints that make incremental stretching untenable.

Why the Grid Can’t Stretch Any Further

The U.S. transmission system was built for a twentieth-century load profile and a far simpler generation mix. Five converging constraints now make incremental “stretch” strategies ineffective:

Recent field incidents highlight the increasing complexity of the system. In northern Virginia’s “Data-Center Alley,” large AI campuses are distorting local power quality; sensor data collected for a Bloomberg investigation shows degraded harmonics affecting nearly 3.7 million nearby residents. On the demand side, NYISO’s 2025 Power Trends report projects a 1.6 – 4 GW capacity gap for New York City early in the next decade as thermal retirements outpace new connections.

With slack erased and failure modes already breaching operating envelopes, the grid’s most urgent need is for firm, carbon-free capacity that can be deployed on compressed timelines and capable of riding existing corridors without exacerbating congestion.

The Nuclear Renaissance – Big-Tech Demand Turns Dormant Reactors Back On

America’s debate over “next-generation” nuclear skipped an unexpected step this year. Instead of waiting for advanced designs, hyperscale power buyers are reviving reactors that were already built, licensed, and mothballed. Microsoft, Amazon, and Holtec have each signed deals that convert long-idle assets into near-term, 24/7 carbon-free supply.  This is an acceleration no one in the industry had modelled twelve months ago.

Recommissioning transitions from press release to steel in the ground

• In Pennsylvania, Constellation Energy will restart the 835-MW Three Mile Island Unit 1 under a 20-year offtake agreement with Microsoft; the project is branded the Crane Clean Energy Center, promises about 3,400 jobs, and carries a public-support margin of roughly two-to-one according to a Susquehanna Polling & Research survey.
• Holtec International has closed a $1.52 billion Department of Energy loan to bring the 800-MW Palisades plant in Michigan back online by late 2025, marking the first U.S. reactor to reverse a decommissioning order.
• Amazon Web Services locked in as much as 1.92 GW of output from Talen’s Susquehanna station through 2042, anchoring a $20 billion data-center buildout in the same state.

Why are corporate boards signing up now? A restart typically costs $1,500–$2,300 per kW, about one-fifth of the cost of a new large-scale build, and delivers levelized power in the $55–$75/MWh range, well below the $149–$250/MWh band Lazard now assigns to new gas peakers even before carbon pricing. With transmission congestion charges already exceeding $20.8 billion annually, postponing firm capacity is becoming more costly than decarbonizing it.

First-wave Small Modular Reactors (SMRs) attract corporate patrons. Long-promised small modular reactors are no longer conceptual line items. Amazon placed a $500 million equity anchor with X-energy for its 320-MW Xe-100. At the same time, Google signed a master development agreement with Kairos Power for a 500-MW fleet of fluoride-salt units, each aiming for commercial operation early in the next decade. Goldman Sachs now brackets mature SMR costs in the $80–$100/MWh range, competitive with combined-cycle gas once tax credits are applied.

Obstacles remain real, not theoretical. The Nuclear Regulatory Commission must process restart amendments simultaneously with finalizing Part 53 – a new, risk-informed licensing pathway for advanced reactor designs, which strains staff capacity. Supply-chain bottlenecks persist for large nuclear-grade forgings and the high-assay low-enriched uranium fuel most SMRs require. And while polling shows 70 % of Pennsylvanians now support nuclear, activist litigation around Three Mile Island illustrates that social license can still turn on a dime.

The commercial logic is nevertheless compelling. Recommissioning plants delivers gigawatt-scale, carbon-free capacity before most new-build projects can clear their interconnection queues. At the same time, small modular reactors (SMRs) create a modular expansion path for the 2030s. Together, they transform the “nuclear renaissance” from post-conference rhetoric into contracted megawatts, laying the firm-supply foundation upon which demand-flex programs, long-duration storage, and digital grid orchestration must now build.

Beyond Restarts – Why New-Build Nuclear Finally Has Traction

Recommissioning delivers gigawatts of firm capacity this decade, yet hyperscale compute growth does not pause when the last dormant reactor switches on. To meet the next wave of demand, utilities and their corporate customers have transitioned from discussing advanced nuclear to signing construction contracts and pouring the first concrete. A pivot sharpened by a recent, highly public failure. When NuScale’s Carbon-Free Power Project collapsed under a projected US$9 billion price tag and the withdrawal of municipal buyers, the episode clarified two immutable rules. Factory repetition must replace customized fieldwork, and anchor offtakers must maintain balance sheets robust enough to absorb the risk of first-of-a-kind projects.

Nowhere is the new playbook clearer than Ontario Power Generation’s Darlington site, where provincial approval in May 2025 unlocked construction of a 300-MW GE–Hitachi BWRX-300 SMR, North America’s first advanced reactor to advance from license to excavation. The fixed-price contract, reported at roughly US$4,000–4,500 per kW, is less than half the overnight cost of the recently completed Vogtle units and will provide a real-world benchmark for subsequent builds. Momentum is crossing the border. Tennessee Valley Authority filed the United States’ first full construction-permit application for the same BWRX-300 design at its Clinch River site in April 2025, leveraging federally owned brownfield land to shave years off a green-field schedule.

A parallel track is unfolding in Wyoming, where TerraPower’s 345-MW Natrium reactor is repowering a retired coal station in Kemmerer. Non-nuclear site work is complete, and despite a two-year delay tied to HALEU fuel supply, the project remains on course to break ground on its nuclear island in 2025. Such coal-to-nuclear conversions preserve switchyards, cooling-water rights, and skilled labor. According to a Department of Energy study, these advantages can reduce total project lead time by up to four years and inject US$275 million in annual economic activity into host communities.

Regulation is beginning to align with this faster tempo. The Nuclear Regulatory Commission’s draft Part 53 rule, which offers a risk-informed licensing pathway for advanced reactors, entered its final comment round in February; the commission has set an unprecedented goal of issuing the rule in early 2026. While Part 53 will not eliminate supply-chain choke points, large nuclear-grade forgings and HALEU fuel remain single-vendor bottlenecks; however, it promises to compress approval timelines to something approaching the 24- to 36-month construction cycles that SMR vendors advertise.

Early cost signals are closing the competitiveness gap. Goldman Sachs Research now brackets mature SMR levelized costs in the US$80–$100 per MWh range, once federal tax credits are applied, already below new peaking gas turbines and within striking distance of combined-cycle gas turbines. With firm, carbon-free kilowatts finally under contract and in the excavation phase, planners can begin to write advanced nuclear into 2030-era load forecasts with confidence, rather than hope. That newly credible supply line must now be integrated with demand-side flexibility, long-duration storage, and digital grid intelligence, the complementary levers that will determine whether the coming decade’s energy equation balances.

Complementary Levers – Demand-Flex, Long-Duration Storage, and AI-Native Grid Intelligence

Firm nuclear capacity solves only half the problem: the other half is temporal. AI training runs and evening EV-charging peaks do not align neatly with wind, solar, or even reactor dispatch curves, so matching supply to load now depends on three interlocking capabilities: flexible demand, multi-day storage, and real-time grid analytics.

While wind and solar remain essential to decarbonization, they also expose the grid to new vulnerabilities. Spain’s national blackout in early 2025, triggered by a prolonged lull in wind generation and compounded by insufficient dispatchable backup, revealed the systemic risks of over-indexing on variable renewables. In the U.S., a flagship solar thermal plant in the Mojave drew scrutiny for contributing to extreme localized heat and devastating local wildlife populations. These incidents highlight that renewable capacity, without firm, fast-reacting backup, is insufficient in an era of machine-speed demand. They underscore the need for balanced architectures that incorporate flexibility, long-duration storage, and nuclear baseload to form a resilient core.

A first layer of flexibility is emerging directly behind the meter. Eight U.S. states, led by California and Connecticut, are recruiting electric vehicle (EV) owners into vehicle-to-grid (V2G) programs that convert idle batteries into miniature peaking resources; participants receive off-peak charging credits in exchange for sending power back during evening peak hours. Massachusetts has taken a further step, launching a two-year demonstration that will install 100 bidirectional chargers across homes, school bus depots, and municipal fleets to test the aggregate value of portable 500-kWh batteries as a virtual power plant. Early modelling by the program’s system operator suggests that just 10% participation among the state’s projected 800,000 EVs would free up roughly 400 MW of flex capacity, enough to offset an entire gas peaker on a mild summer evening.

Where flexibility alone cannot fill multi-day gaps, long-duration storage is stepping in. Form Energy broke ground on the nation’s first iron-air battery installation in Minnesota, a 10-MW/1,000-MWh unit capable of discharging for 100 hours at costs the developer projects to be below those of lithium-ion on a per-delivered-kWh basis. In parallel, the Department of Energy’s Water Office estimates that refurbishing legacy dam sites and adding new pumped-storage hydropower could unlock 35 GW of additional long-duration capacity, enough to buffer several days of nationwide renewable variability.

Making this mosaic operate like a single system requires cognition at the grid edge, which is being achieved via purpose-built silicon and digital twins. Schneider Electric has begun rolling out a “One Digital Grid” platform that enables data-center operators to throttle workloads or export on-site generation in response to ISO price signals, effectively turning hyperscale campuses from inflexible loads into dispatchable resources. At the distribution level, Utilidata’s Karman module, an NVIDIA-powered AI chip embedded in next-generation smart meters, optimizes voltage and EV-charging profiles street by street, shaving peak current by up to 12 % in early pilots.

Individually, these projects are incremental; together, they form the operating envelope that allows new nuclear megawatts to run at high capacity factors while the grid absorbs inevitably spiky AI and transport loads. With hard capacity on the way and flexible orchestration taking shape, planners can begin to balance the 2030s equation. However, only if the remaining bottlenecks—licensing, HALEU fuel, and transmission steel — are tackled with equal urgency.

Strategic Framework 2025-2035 — Revive | Rebuild | Reshape

America’s power equation now hinges on a simple hierarchy of moves: revive what is already built, rebuild with factory-made reactors sized to hyperscale growth, and reshape the grid so flexible demand and long-duration storage convert new megawatts into real capacity.

Revive – 2025-2028

Recommissioning dormant nuclear stations is the fastest way to inject firm, carbon-free power before the end of the decade. Engineering studies for Three Mile Island Unit 1, Palisades, and Susquehanna indicate average overnight costs between US$1.5 and US$2.3 million per MW, roughly one-fifth of recent green-field builds, with levelized generation in the US$55-75/MWh range. Because switch-yards, cooling loops, and containment structures already exist, most of the critical path runs through licensing amendments rather than civil works, compressing schedules by up to five years and delivering about 7 GW of capacity by 2028.

Rebuild – 2026-2035

The next tranche of firm supply will come from small modular reactors (SMRs) and advanced sodium or high-temperature designs deployed on brownfield coal sites. Ontario Power Generation’s GE-Hitachi BWRX-300 at Darlington and Tennessee Valley Authority’s Clinch River project provide the template: standardized design, factory fabrication, and a site that already has transmission rights. Early contracts indicate an overnight cost of around US$4,000-4,500/kW and a mid-range levelized cost of US$80-100/MWh once federal tax credits are applied, bringing SMRs within striking distance of combined-cycle gas even before a carbon price is factored in. A well-paced program can add 3-5 GW per year through the early 2030s, enough to match projected AI-compute growth on a rolling basis.

Reshape – 2025-2035

Firm supply alone cannot handle the daytime spikes created by evening EV charging and AI-training clusters. Flexibility must move in lock-step. Vehicle-to-grid pilots in California, Connecticut, and Massachusetts indicate that one in ten EVs enrolled in bidirectional charging could free ≈400 MW of dispatchable capacity during the 6-10 p.m. peak, comparable to a midsize gas peaker. On the storage front, Form Energy’s 100-hour iron-air battery in Minnesota, combined with the DOE’s pumped-hydro refurbishment roadmap, points to more than 35 GW of multi-day buffering potential by 2030. Orchestration comes from silicon: Schneider Electric’s “One Digital Grid” and Utilidata’s feeder-level AI chips already trim distribution peaks by up to 12 percent in field trials, turning formerly rigid loads into dynamic grid assets.

Cumulative Capacity Added or Activated (nameplate GW)

• † Effective capacity from V2G, virtual power plants, 100-hour storage, and AI-orchestrated demand.
• DOE high-case forecast adds ≈40 GW of new peak demand by 2035, driven chiefly by AI and transport electrification.

Implemented together, these tiers form a self-reinforcing stack. Revived plants stabilize the grid while modular builds ramp up; demand-side and storage programs absorb volatility, allowing reactors to run at full capacity; AI orchestration reduces the cost of every additional megawatt. The remaining barriers are execution risk, licensing bottlenecks, HALEU fuel supply, and transmission steel, which must now be prioritized for board-level risk management.

Risk and Uncertainty Lens – What Could Derail the Capacity Plan

Even the strongest three-tier program must clear a gauntlet of uncertainties before its promised gigawatts translate into reliable electrons. Five risk vectors stand out.