Wall Street wants growth this quarter, not clean power in 2032. That single fact explains why Microsoft, Amazon, and Meta now chase old reactors instead of greenfield renewables. Industry consensus still points toward new microgrids and clean-sheet builds. Reality, however, points somewhere else entirely. Hyperscalers are buying legacy generation, restarting shuttered reactors, and locking down existing nuclear output. This piece explains why physics and finance both favor old iron over new construction right now. Every dollar of capital expenditure now competes against a hard physical constraint. Chips, land, and cooling infrastructure can scale relatively quickly with enough money. Power generation, unfortunately, follows an entirely different, much slower calendar. Understanding that mismatch explains almost every strange deal announced over the past two years.
The Consensus View Everyone Repeats
Ask most analysts how AI infrastructure gets powered, and you’ll hear the same story. New renewable microgrids, paired with battery storage, will meet exploding demand. Clean-sheet solar fields and geothermal wells will scale alongside compute clusters. This narrative isn’t wrong exactly, just incomplete. Nearly all major operators have committed publicly to renewable power purchase agreements. Wind, solar, and hydroelectric contracts remain a genuine, growing part of every hyperscaler’s portfolio. Yet timing gets glossed over in this telling constantly. New generation projects, however clean, take years to permit, finance, and build. Meanwhile, AI training clusters need gigawatts within quarters, not years.
What Wall Street Actually Demands
Hyperscaler capital spending tells its own urgent story here. Combined capex is forecast to exceed $600 billion in 2026, up 36% from the prior year. Roughly $450 billion of that sum ties directly to AI infrastructure buildout. Investors expect that capital to generate returns on a predictable schedule. Every quarter without deployed compute capacity represents lost competitive ground against rivals. Consequently, power procurement decisions get judged on speed first, purity second. Credit markets have noticed this pressure too. Rising power, land, and construction costs have pushed hyperscalers toward heavier bond issuance and project-level financing. By late 2025, this shift had already begun worrying credit investors and economists.
The Scale Problem Nobody Anticipated
Standard data centre campuses once drew between 200 MW and 500 MW of power. Next-generation AI campuses now target 1 GW to 5 GW each. That range equals the output of an entire nuclear power plant, per campus. Goldman Sachs Research quantified this trajectory clearly in May 2026. US data centre power demand reached 31 GW in 2025, and should hit 41 GW in 2026. That figure will nearly double again, to 66 GW, by 2027. S&P Global’s separate October 2025 outlook broadly agrees with this pace. US data centre capacity should reach 75.8 GW in 2026, crossing 100 GW by 2028. Annual capacity additions must reach 36.3 GW in 2027 alone, versus just 6.4 GW added in 2024.
Why New Gas Plants Can’t Move Fast Enough
Gas turbines sit at the centre of new power plant construction, and only three manufacturers build them at scale. GE Vernova, Siemens Energy, and Mitsubishi Heavy Industries together dominate this global market entirely. Each now reports backlogs stretching roughly five years out. GE Vernova alone expects its combined gas turbine backlog to reach 110 GW by the end of 2026. Annual manufacturing capacity, meanwhile, sits at roughly 10 GW. Simple division shows a full decade of committed output already booked. Pricing reflects this scarcity sharply too. New order pricing in the first half of 2026 ran 10 to 20 percentage points higher per kilowatt than the previous quarter. That increase outpaced general inflation considerably. Anyone planning new gas generation before 2030 should verify supplier queue position first. Interconnection status alone tells an incomplete story. Without a confirmed turbine slot, a signed interconnection agreement means very little in practice.
Coal Plants Get a Second Look Too
Gas turbine scarcity has produced one more unexpected consequence. Utilities planning coal retirements now hesitate before finalising shutdown dates. Watch for announcements delaying retirements specifically because replacement turbines aren’t available yet. This dynamic shifted sharply during 2025 and 2026 specifically. New federal policies expedited gas infrastructure permitting right as AI demand shocked the market. Together, these forces made keeping old coal plants running temporarily cheaper than replacing them quickly. Environmental groups understandably view this reversal with real concern. Extending fossil fuel operations, even temporarily, locks in emissions that climate commitments were meant to avoid. Yet grid operators facing genuine reliability gaps have limited alternatives available today.
Interconnection Queues Compound the Delay
Even secured turbines don’t guarantee timely power delivery. PJM, the largest US grid operator, illustrates this problem starkly. Projects reaching operation in 2025 spent an average of eight years waiting in queue. That timeline has worsened steadily since 2008, when average wait times sat under two years. PJM’s own reform efforts have reduced backlog somewhat, yet roughly 46 GW of projects still await final study completion. Processing time for new agreements now runs one to two years at best. Compounding this further, PJM’s capacity auction prices have spiked dramatically. Prices jumped from roughly $29 per MW-day for one delivery year to $270 per MW-day the next. In certain zones, prices cleared even higher, near $466 per MW-day.
Existing Nuclear Solves Both Problems at Once
Against that backdrop, existing nuclear plants suddenly look extraordinarily attractive. These facilities already hold grid interconnection rights, transmission access, and operating permits. Restarting one skips the queue entirely, rather than joining it. Nuclear also delivers something gas and renewables both struggle to match consistently. It runs continuously, at high capacity factors, regardless of weather or fuel price swings. Constellation’s fleet posted a 92.3% capacity factor in the first quarter of 2026 alone. Proponents of restart projects frame this clearly. New nuclear construction remains at least half a decade away, often longer. Restarting a recently shuttered plant, by contrast, can happen within two to three years.
Three Mile Island Becomes Crane Clean Energy Center
Microsoft’s September 2024 deal with Constellation set the template for this whole strategy. The 20-year power purchase agreement covers Three Mile Island Unit 1, retired in 2019 for economic reasons. Renamed the Crane Clean Energy Center, the plant targets restart by 2027 or 2028. Financially, the numbers behind this restart are enormous. The facility adds roughly 835 MW of carbon-free capacity to the grid. Constellation expects it to generate over $3 billion in state and federal taxes over its lifetime. Federal support followed quickly behind private capital too. The Department of Energy issued a $1 billion loan to Constellation in November 2025. That loan marked the first project financing closed under the current administration’s nuclear expansion push. Regulatory momentum has only accelerated since then. FERC granted Constellation a waiver in April 2026, transferring interconnection rights from another facility. That waiver helps the Crane unit reach full deliverability status faster.
Palisades Breaks New Ground as a Restart Precedent
Michigan’s Palisades plant offers a second, equally instructive case study. Entergy sold the facility to Holtec International in 2022 purely for decommissioning work. Rising clean energy demand reversed that plan entirely within two years. Holtec secured a $1.5 billion DOE loan guarantee to fund the reversal. The Nuclear Regulatory Commission approved transitioning the plant back to operating status in July 2025. This marked the first time any US facility made that specific transition. Even so, restarts aren’t instant or risk-free processes. Palisades faced repeated delays tied to equipment inspections and safety exemptions. Critics, including former plant staff, questioned whether the NRC moved too quickly. Despite friction, the strategic logic remains intact regardless. Holtec now plans two small modular reactors adjacent to Palisades itself. Restarting existing infrastructure, evidently, doubles as a springboard for future nuclear expansion.
The Susquehanna Deal Shows a Different Structure
Amazon’s approach with Talen Energy took a notably different legal path. Rather than restarting anything, Amazon bought a data centre campus adjacent to Talen’s operating Susquehanna nuclear plant. The original 2024 deal proposed a behind-the-meter connection for direct power supply. Regulators initially blocked this exact structure, however. FERC rejected the amended interconnection agreement in November 2024. Commissioners worried the arrangement would shift transmission costs onto other ratepayers unfairly. Talen and Amazon adapted quickly rather than abandoning the plan. Their June 2025 agreement restructured the deal into a front-of-the-meter arrangement instead. Talen now acts as Amazon’s licensed retail electricity provider directly. The revised contract’s scale remains genuinely striking regardless of structure. It covers 1,920 MW of capacity, ramping fully by 2032 at the latest. Talen expects roughly $18 billion in revenue over the agreement’s full life.
Regulatory Guardrails Are Still Being Written
FERC’s rejection of the original Amazon-Talen structure revealed a deeper regulatory gap. Rules for co-located loads paired with existing generation simply didn’t exist yet. Regulators have spent the following eighteen months building that framework from scratch. PJM now faces a December 2026 Show Cause Order from FERC directly. That order requires clear rules for co-located load and behind-the-meter generation. Three new transmission service options must emerge from this process. Utility opposition adds further complication to these negotiations. Exelon and American Electric Power challenged the original Amazon expansion request directly. They argued it could shift up to $140 million yearly onto other customers. Given this friction, front-of-the-meter deals like Talen’s revised structure look increasingly durable. They avoid the ratepayer cost-shifting concerns that sank the original approach. Expect more hyperscaler deals to follow this exact template going forward.
Coal and Gas Retirements Are Getting Postponed
Nuclear isn’t the only legacy asset benefiting from this dynamic. Aging coal and gas plants scheduled for retirement now face delayed shutdown timelines instead. Operators cite persistent grid tightness and equipment procurement challenges as justification. Federal policy has explicitly encouraged this reversal too. Extensions for fossil generation facilities planning retirement have already increased both costs and emissions somewhat. Critics argue this approach merely delays an inevitable transition rather than solving underlying scarcity. PJM’s most recent capacity auction results illustrate this shift numerically. The cleared resource mix included 43% natural gas, 21% nuclear, and 20% coal. That coal share would have looked unthinkable in climate-focused industry forecasts from just three years ago.
Vistra and Constellation Race to Consolidate Legacy Assets
Beyond individual restarts, entire portfolios of legacy generation are changing hands rapidly. Constellation completed a $22 billion acquisition of Calpine Corporation in January 2026. That deal added 23 GW of natural gas, geothermal, and solar capacity instantly. This single transaction transformed Constellation’s entire business model overnight. Once a nuclear-focused operator, it became the largest competitive power generator in the country. Revenue jumped 64% year-over-year in the first quarter following the deal’s close. Vistra pursued a similar consolidation strategy through separate channels. It finalised acquiring roughly 2.6 GW of natural gas power from Lotus Infrastructure Partners in October. That deal added seven gas plants across key American markets to Vistra’s fleet. Both companies now sign long-term contracts directly with hyperscalers as their core growth strategy. Vistra’s agreements include a 20-year deal for its Comanche Peak nuclear plant. These consolidations effectively turn legacy generation portfolios into AI infrastructure bonds.
Meta Chooses Uprates Over New Construction
Meta’s approach reveals a subtler version of the same buyout logic. Rather than restarting shuttered plants, Meta extended life on reactors already running. Its June 2025 deal with Constellation covers 1.1 GW from an Illinois facility, the Clinton plant, for twenty years. Meta later expanded this strategy dramatically further. Three additional deals, signed with Vistra, TerraPower, and Oklo, cover up to 6.6 GW combined. That makes Meta the largest single corporate nuclear buyer in American history. Crucially, Vistra’s agreements center on power uprates, not new plants. Existing Ohio reactors will gain 433 MW of combined output increases through 2034. Stacey Doré, Vistra’s chief strategy officer, called this the largest corporate-backed nuclear uprate programme to date. Even Meta’s next-generation reactor bets lean on this same bridging logic. Jefferies analysts noted hyperscalers increasingly favour advanced reactor designs over conventional new-builds. Yet even those projects won’t deliver power until the early 2030s at the earliest.
Natural Gas Fills the Gap Until Nuclear Arrives
Here lies an important, often overlooked admission within Meta’s own strategy. Advanced reactors won’t deliver meaningful power before the 2030s under any realistic timeline. Meta has openly turned to natural gas to bridge that exact gap. Its Richland Parish campus in Louisiana runs substantially on gas today. Separately, Oracle signed multiple gas supply agreements with Energy Transfer for three US data centres. Two of those sit in Texas, with first gas flow expected by year’s end. This pattern repeats across nearly every major hyperscaler’s actual portfolio. Public announcements emphasise nuclear and renewables prominently, understandably, given their climate messaging value. Quietly, gas turbines and existing fossil capacity handle the immediate load nonetheless.
The Full Scoreboard Across Hyperscalers
Zooming out, the scale of this shift across the whole industry becomes clear. Tracking published deals shows at least 13 gigawatts of nuclear capacity contracted by four companies alone. Google, Amazon, Meta, and Microsoft each pursue slightly different combinations of restart, uprate, and SMR investment. Google’s portfolio leans toward SMR development partnerships specifically. It committed to 500 MW from Kairos Power’s molten salt reactor design. Separately, Google partnered with NextEra Energy exploring a restart of Iowa’s Duane Arnold plant. Amazon spread its bets across manufacturing capacity as much as generation itself. It led a $700 million investment round in X-energy, targeting up to twelve small reactors. Separate agreements with Dominion Energy and Energy Northwest add roughly 1.26 GW more. Across the board, one pattern holds regardless of company or reactor type chosen. Every deal either revives existing infrastructure or bridges toward future capacity with gas. Genuinely new, clean-sheet generation remains conspicuously absent from these announcements.
Small Modular Reactors Remain a 2030s Story
Given all the SMR announcements, it’s worth stating their timeline plainly. Amazon’s X-energy investment targets deploying capacity by 2039 at current pace. Meta’s Oklo partnership aims for 1.2 GW in Ohio, also by 2030. TerraPower’s Natrium design, backed by Meta, follows a similarly staggered rollout. First units, totalling 690 MW, target 2032 at the earliest. Six additional units wouldn’t arrive until 2035, assuming no delays occur. None of these projects, however promising, solve today’s actual power shortage. They represent genuine long-term diversification, not near-term supply relief. Every hyperscaler pursuing SMRs simultaneously pursues restarts and uprates precisely because SMRs alone can’t meet 2027 deadlines. Broader industry forecasts echo this same cautious timeline regardless of individual company plans. Global nuclear capacity might reach 494 GW by 2035 under current projections. Reaching that figure still depends heavily on permitting speed and manufacturing scale-up, neither guaranteed yet.
The Ratepayer Backlash Building Underneath
None of this consolidation comes free for ordinary electricity customers. PJM’s market monitor found data centres accounted for 40% of capacity costs in its latest auction. That share continues climbing as more large loads join the grid. Public pressure eventually forced a political response from the industry itself. The White House issued a Ratepayer Protection Pledge in March 2026, signed by leading hyperscalers. Signatories guaranteed their data centres wouldn’t raise household electricity costs. Analysts, however, quickly questioned the pledge’s actual force. It carries no enforcement mechanism and remains legally toothless in practice. Some signatories even described it as affirming existing commitments, not creating new obligations. Meanwhile, state regulators are pursuing their own, more binding guardrails separately. Guidelines increasingly require large-load customers to pay for upgrades triggered specifically by their own interconnection. This shifts financial risk back toward the companies actually creating new demand.
Texas Offers a Contrasting Regulatory Model
Not every state follows Virginia’s strict cost-causation approach exactly. Texas, operating its own independent ERCOT grid, takes a notably different path. JLL projects Texas will overtake Virginia as the largest data centre market by 2030. Structural advantages explain much of this shift toward Texas specifically. Land remains abundant and affordable, with fewer residential proximity conflicts nearby. ERCOT’s independent grid structure also offers generation capacity unavailable within PJM’s constrained footprint. Nearly two-thirds of all US data centre construction now occurs in these frontier markets. Texas alone accounts for 6.5 GW of the 35-plus GW currently under construction nationwide. Tennessee, Wisconsin, and Ohio round out the remaining frontier destinations. Every month of policy uncertainty in Virginia effectively hands Texas more recruitment leverage. Hyperscalers, holding roughly $700 billion in combined 2026 capital expenditure plans, negotiate from genuine strength. That scale gives them real power when choosing between competing state regulatory environments.
States Push Back With New Rate Classes
Regulators outside PJM’s federal process have grown restless too. As of May 2026, twenty-three states approved at least one large-load tariff. Seven more states have similar proposals pending review. Ohio moved first among major data centre markets specifically. Its 2025 tariff created a new customer class for facilities drawing 25 MW or more. Data centres must pay at least 85% of contracted capacity, whether they use it or not. Results followed almost immediately after implementation. AEP’s large-load forecast in Ohio dropped by half once the tariff took effect. Speculative interconnection requests, evidently, shrink fast once real financial commitment gets required upfront.
Virginia’s GS-5 Tariff Raises the Stakes Further
Virginia, hosting the world’s densest data centre cluster, followed with even stricter terms. Its GS-5 tariff, approved November 2025, takes effect January 2027. Large customers must sign fourteen-year contracts before connecting at all. Collateral requirements under this tariff are genuinely severe by any measure. Data centres must post $1.5 million in collateral for every contracted megawatt. That figure prices exactly the stranded-asset risk regulators previously left unquantified. Applied elsewhere, this pricing reveals genuinely staggering scale. PPL Electric’s roughly 20 GW of contracted large load, priced at Virginia’s rate, implies about $30 billion in required collateral. That number alone shows how large hidden grid risk had quietly become. Public frustration helped drive this legislative momentum considerably. Seventy-eight percent of Virginia voters blame data centres for rising electricity bills directly. That sentiment pushed lawmakers toward SB 253, shifting billions in costs toward large users.
Legacy Power Buys Insulation From This Regulatory Wave
This state-level cost-shifting adds a further, underappreciated reason for legacy power deals. A hyperscaler buying existing nuclear output through a PPA avoids much of this exposure entirely. It isn’t triggering new grid upgrades the same way a fresh interconnection request would. Front-of-the-meter arrangements, like Talen’s revised Amazon deal, reinforce this insulation further still. Existing transmission capacity, already paid for and depreciated, carries none of the collateral risk new interconnections now demand. Every dollar spent buying old capacity is, in part, a dollar spent avoiding tomorrow’s tariff exposure. More than 300 state bills addressing data centres appeared within six weeks of 2026’s legislative sessions opening. At least eighteen states now consider special rate classes for large energy users specifically. This regulatory momentum shows no sign of slowing anytime soon.
Private Capital Floods Into Adjacent Infrastructure
Beyond hyperscalers themselves, private capital keeps flowing into supporting infrastructure platforms. DayOne Data Centers raised over $2 billion in Series C financing during January 2026 alone. Sovereign wealth funds, including Indonesia’s investment authority, joined that round directly. This capital increasingly targets power-adjacent platforms specifically, not just server halls themselves. Investors recognise that whoever controls reliable generation access controls the entire value chain downstream. Land, permits, and interconnection rights have effectively become the scarcest commodity in this build-out. Leverage levels across the sector will likely keep climbing through 2028 regardless of these dynamics. Developers focused on hyperscale buildouts increasingly rely on securitised bonds and dedicated project finance vehicles. Every layer of this financing stack ultimately depends on power arriving roughly on schedule.
Why This Strategy Still Carries Real Risk
Buying legacy generation solves speed problems, yet introduces new risks entirely. Aging nuclear fleets carry rising maintenance costs as equipment approaches end-of-life eventually. Investors must weigh this against near-term revenue certainty carefully. Restart projects also face unique regulatory scrutiny unlike routine license renewals. Palisades became the first plant transitioning from decommissioning back to operations. That precedent invited exactly the kind of legal challenge critics warned about. Nuclear equipment itself poses operational constraints too, separate from regulation. Running reactors below rated output for extended periods risks elevated vibration and wear. Constellation cited exactly this concern when requesting its FERC deliverability waiver.
Riskier Lending Structures Emerge at the Margins
Not every financing structure enjoys blue-chip credit backing, however. CoreWeave’s $7.5 billion debt facility, arranged in 2024, used GPUs themselves as collateral. That loan carried a variable rate averaging roughly 11%, split across investment-grade and speculative tranches. Repayments began in January 2026, just as GPU collateral values started softening. This timing mismatch illustrates a genuine structural risk within the broader financing ecosystem. Not every AI infrastructure loan carries the same fortress-balance-sheet backing as Microsoft or Google. Convertible bonds offer a cheaper, if riskier, alternative for some borrowers too. CoreWeave separately sold $2.25 billion in convertibles at a coupon of just 1.75%. That low rate depends entirely on equity prices eventually rising as promised. Power procurement risk and financing risk, in practice, sit closely intertwined. A facility without secured power generates no revenue to service its debt. Lenders increasingly scrutinise energy contracts as closely as they scrutinise customer agreements themselves.
Bond Markets Are Funding This Entire Shift
Financing this power scramble requires capital far beyond typical corporate treasuries. Hyperscalers issued roughly $121 billion in bonds during 2025 alone. That figure runs more than four times the previous five-year average. Morgan Stanley now expects $250 to $300 billion more in hyperscaler issuance during 2026. AI-related investments already made up roughly 30% of total US investment-grade net issuance last year. Analysts project that share could reach 15 to 20% of major bond indexes within five years. Credit quality varies considerably across companies pursuing this strategy, however. Microsoft holds a AAA rating, while Meta and Amazon sit at AA-. Oracle, by contrast, sits at BBB, having suffered two downgrades toward junk status already.
Off-Balance-Sheet Structures Hide the Real Exposure
Beyond simple bonds, hyperscalers increasingly use special purpose vehicles to isolate risk. Meta’s Hyperion campus in Louisiana illustrates this structure clearly. A joint venture, 80% owned by Blue Owl, borrowed $27 billion to fund construction. Crucially, this debt doesn’t touch Meta’s own credit rating directly. Meta instead signed renewable operating leases beginning once construction finishes. A residual value guarantee effectively functions like a lease obligation for debt investors. This structure spreads well beyond Meta’s single project regardless. Asset-backed securities and commercial mortgage-backed structures now routinely finance hyperscale campuses. Investors favour these hyperscale deals specifically because underwriting stays easier than for smaller, sub-investment-grade tenants.
Community Opposition Adds a Final Layer of Friction
Even fully financed, permitted, and power-secured projects face one more obstacle. Community opposition has genuinely become a significant headwind for new construction. Roughly $64 billion in projects have been blocked or delayed over the past two years. Organised activist groups now number over 142 across 24 states. Northern Virginia alone hosts more than 40 active local campaigns against new facilities. Lenders increasingly factor this opposition into their underwriting decisions directly. This friction adds yet another reason legacy power looks comparatively attractive. Buying output from an already-permitted nuclear plant avoids most siting battles entirely. The community fight, in many cases, happened decades ago when the original plant was first built.
What This Means for the Broader Grid
Stepping back, this buyout wave reshapes electricity markets well beyond hyperscalers themselves. PJM’s peak load is now expected to grow 3.6% annually through 2036. Data centres drive the overwhelming majority of that new demand. Legacy generation, once viewed as a depreciating stranded asset, has become genuinely strategic infrastructure instead. Plants slated for closure now attract billion-dollar restart investments and federal loan guarantees. This reversal would have seemed implausible before the AI supercycle began. Ultimately, physics dictated this outcome more than any single company’s preference. Building genuinely new generation, clean or otherwise, still takes years nobody currently has. Buying, restarting, and repurposing existing plants remains the only path matching AI’s actual timeline.
Demand Flexibility as a Temporary Stopgap
While supply-side deals dominate headlines, demand-side flexibility offers a quieter parallel strategy. Google has worked to make its own data centres more flexible operationally. Its global head of data centre energy noted it’s often cheaper to pay other customers to shift usage instead. Virtual power plant programmes illustrate this approach at a smaller, consumer-facing scale. Light-duty vehicles can earn roughly $3,000 per summer through enrollment in such programmes. School buses, with larger batteries, can earn as much as $12,000 seasonally. These flexibility mechanisms won’t replace gigawatt-scale generation deals anytime soon regardless. They do, however, buy grid operators valuable breathing room during peak demand hours. Every megawatt shifted voluntarily is a megawatt hyperscalers don’t need to procure through harder-fought channels.
A Familiar Historical Pattern Emerges
Skeptics sometimes compare today’s AI buildout to the dot-com telecom bubble. That comparison holds only partially, according to credit analysts tracking both eras closely. Today’s borrowers carry genuinely profitable existing businesses, unlike many speculative 1990s telecom startups. Still, one parallel deserves attention regardless of that distinction. Both booms outran the physical infrastructure needed to support them initially. Telecom firms overbuilt fibre before demand caught up; hyperscalers now chase power before supply catches up. The difference lies in which side of that equation moves first this time. Fibre capacity sat unused for years after telecom’s crash eventually. Power capacity, by contrast, remains the binding constraint hyperscalers actively fight to secure today.
The Road Ahead for Power and Compute
Looking forward, expect this trend to intensify rather than fade. Turbine backlogs stretch toward 2030 already, with hyperscaler volume agreements reportedly reaching 2035. New nuclear construction, including small modular reactors, still sits years from meaningful scale. Consequently, more shuttered plants will likely receive restart evaluations soon. Duane Arnold in Iowa already received FERC approval for its own 2030 restart. Expect similar announcements as hyperscalers exhaust the current supply of viable candidates. For now, the industry’s honest answer looks less elegant than the clean-sheet narrative suggests. Old iron, carefully restarted and contractually secured, is genuinely saving the AI supercycle’s near-term timeline. Whether that solution proves sustainable depends on choices regulators and ratepayers are only beginning to negotiate. That negotiation will define electricity markets for at least the next decade. Every restarted reactor, every uprated turbine, and every renegotiated tariff represents a small settlement. Somewhere between hyperscaler urgency and ratepayer patience, the actual shape of America’s power grid gets decided.
