Power & Infrastructure Weekly — Apr 25, 2026

This was the week the "just build more gas" answer to AI power demand stopped working on three fronts at once: BloombergNEF clocked combined-cycle construction costs up 66% in two years, FERC pushed its large-load interconnection deadline to the end of June 2026, and TerraPower put a shovel in the ground on the first new utility-scale advanced reactor in the United States. None of these stories is, on its own, a turning point. Together, they describe a system where the cheapest short-term answer is getting more expensive, the regulatory backstop keeps slipping, and capital is genuinely flowing toward the alternatives — nuclear, storage, flexible load — fast enough that the planning math is rearranging in real time.

• Natrium Reactor (Kemmerer Unit 1) (TerraPower): 345 MW sodium-cooled fast reactor with molten-salt storage that boosts dispatchable output to 500 MW; nuclear island construction officially started April 23, 2026.
• Xe-100 Reactor (IPO Capital Raise) (X-energy): Helium-cooled, high-temperature gas reactor using TRISO fuel; the company raised roughly $1 billion in its Nasdaq debut, pricing 21% above the marketed range at pricing.
• Shipyard-Built Grid-Scale Reactor Program (Blue Energy): $380 million raise to assemble 1.5 GW reactor modules in shipyards, with first Texas project targeted to begin construction this year.
• PFAS OUT Initiative (EPA): Proactive engagement program targeting roughly 3,000 drinking water systems with known PFOA or PFOS challenges ahead of compliance deadlines, launched April 14, 2026.
• Large Load Connect-and-Manage Filing (PJM): Filed price collar and expedited interconnection mechanisms tied to a 6,600 MW capacity shortfall against installed reserve margin, principally from data center growth.

The average build cost for a combined-cycle gas plant rose to $2,157 per kilowatt last year from less than $1,500 in 2023, according to BloombergNEF. Lead times are up 23% over the same period. GE Vernova's gas turbine backlog now stretches through 2028, and Jefferies has flagged that the skilled labor pools that built gas plants a decade ago have largely retired or migrated to LNG export, semiconductor fabs, and data center construction.

What changes if this sticks: every utility integrated resource plan filed in the last two years used cost assumptions that are now structurally wrong, and this could lead to revised filings at state public utility commissions — the comparator math against batteries, demand response, and firm low-carbon options shifts meaningfully in those revisions. The BNEF figure also predates any cost impact from the Iran conflict that began February 28, 2026, so the next update could be worse.

What to watch: whether utilities defer planned gas builds in upcoming IRP cycles, and whether NextEra's claim that gas plant costs have effectively tripled (a higher figure than BNEF's) gets corroborated in subsequent filings. If the $2,157/kW number drifts upward in the second half of 2026, the gas-as-bridge thesis is finished as a default planning assumption.

On April 16, 2026, FERC announced it will act on its large-load interconnection docket by the end of June 2026 — a slip from the April 30, 2026, deadline the Department of Energy had set. The proceeding governs how multi-gigawatt loads connect to the interstate transmission system, and according to Engineering News-Record, the proposed framework marks "a sharp departure from the traditional socialized model" of recovering transmission upgrade costs through regional rates.

What changes if this lands well: developers modeling 2027–2028 interconnection get a federal rulebook on cost responsibility, screening, and timing — and the "data centers should pay their own way" principle moves from state-by-state experiment to federal floor. What changes if it doesn't: ratepayer advocates have already told Utility Dive the DOE proposal is an "unprecedented expansion" of FERC's jurisdiction, which means whatever FERC issues in June will land in court regardless of content.

The signal worth watching: any further slip past June 2026. The April 30 date wasn't arbitrary, and a second miss would push the rule past the point where it can shape projects breaking ground next year. PJM is already filing operational workarounds — price collars, expedited interconnection, large-load flexibility products — which suggests regional operators have stopped waiting.

On April 23, 2026, TerraPower officially started construction on the nuclear island at Kemmerer Unit 1 in Wyoming — the first utility-scale advanced reactor under construction in the United States and the first commercial non-light-water reactor to receive an NRC permit in more than 40 years. The Natrium design pairs a 345 MW sodium-cooled fast reactor with molten-salt energy storage that can boost output to 500 MW for flexible dispatch. TerraPower targets first power by 2030 and has a separate deal with Meta for up to eight Natrium reactors.

The detail buried in the celebration: PacifiCorp, the original utility partner, has not announced a power purchase agreement for Kemmerer Unit 1, and TerraPower's announcement quoted Rocky Mountain Power — a PacifiCorp subsidiary — rather than PacifiCorp itself, per ANS Nuclear Newswire. That's an unusual gap for a project pouring concrete on a 2030 commercial operation date.

What changes if this works: Bechtel's prefabrication-heavy execution model becomes the template for compressing nuclear construction timelines, and HALEU (high-assay low-enriched uranium) fuel supply from DOE becomes the binding constraint rather than reactor physics. What failure looks like: schedule slip on HALEU, no PPA from PacifiCorp by 2027, and the Meta deal becomes the project's commercial anchor by default — which would be a different story about who actually pays for advanced nuclear.

X-energy raised approximately $1 billion in its Nasdaq debut, the largest nuclear public offering on record, with shares pricing 21% above the marketed range at pricing. The Xe-100 is a helium-cooled, high-temperature gas reactor using TRISO fuel — a design that produces industrial process heat as well as electricity, and crucially, requires dramatically less cooling water than light-water reactors.

That last point matters more than the IPO headline. Hyperscalers siting in drought-prone regions of Texas, Arizona, and the desert Southwest have been quietly hitting water-availability ceilings, and a reactor design that can supply firm power without competing for municipal water supply changes the siting map. X-energy's business model emphasizes licensing rather than plant ownership, and the immediate commercial milestone to watch is whether the proposed four-unit plant for Dow in Texas reaches a final investment decision.

If Dow goes FID, advanced nuclear officially has an industrial heat market — cement, steel, chemicals — that doesn't depend on utility offtake at all. If it doesn't, X-energy is a well-capitalized licensing company waiting for someone else to build first.

According to TechCrunch, Blue Energy raised $380 million to pursue grid-scale nuclear plants assembled in shipyards rather than built on-site, with its first 1.5 GW Texas project slated to begin construction later this year. The reactor design is conventional; the manufacturing approach is not.

This is the story advanced nuclear has been avoiding for a decade. Schedule blowouts and cost overruns at Vogtle and elsewhere weren't reactor-physics problems — they were site-built complexity problems. Move fabrication into controlled shipyard environments, barge modules to site, and in theory you get repeatability instead of artisanal megaprojects.

What to watch: whether Blue Energy lands a real utility offtake or EPC partner this year. A financing round is not a commissioned plant, and shipyard-based nuclear has been pitched before. The signal that this is real is a named buyer — without one, it's a well-funded thesis.

Data Center Dynamics' MENA programming this week shifted the liquid cooling conversation from "water good, air bad" toward something more practical: whether the supply chain, integrators, and operators can deploy direct-to-chip cooling at scale without turning every project into a one-off science fair. The thermal physics aren't the bottleneck anymore. Installation labor, controls integration, spare parts inventory, and maintenance practices are.

What changes if cooling becomes an operational competency rather than a facilities afterthought: vendors with installed-base service revenue (Vertiv, Schneider Electric, Trane, the newly combined Eaton/Boyd Thermal stack) start pulling ahead of pure-play hardware shops. The OEM signal to watch is Q1 earnings — specifically whether liquid-ready backlog is converting into recognized equipment revenue, not just pipeline.

Canary Media reported this week that Illinois, New Jersey, and Texas are moving in three different directions on who pays for the grid upgrades AI data centers require. Illinois lawmakers are debating energy and water disclosure plus grid cost contributions. New Jersey lawmakers are considering efficiency and reporting requirements on new data center construction. Texas, hosting the largest concentration of AI compute, is navigating ERCOT reliability against the political pressure to attract investment.

The "data centers pay their own way" consensus is hardening into law — but the mechanisms differ enough that hyperscalers face genuinely inconsistent regimes across states. A 50 MW threshold for on-site generation in one state, a 0.2 gal/kWh water-use efficiency cap in another, and ERCOT's tactical embrace of non-exporting generation registrations (where developers classify campuses as on-site generation paired with batteries to skip transmission queues) all push toward different engineering choices on the same project.

The second-order effect: site selection for the next generation of hyperscale campuses is becoming a regulatory arbitrage exercise, and the states that write the cleanest rules first will get the projects that can't tolerate ambiguity.

This week: a sodium-cooled reactor breaking ground in Wyoming wind gusts, a billion dollars of nuclear IPO money that priced like a meme stock at pricing, and a gas turbine waiting list that now stretches past the next presidential election. The cheapest answer to AI power demand turned out to be the one nobody can build on time, at the price they quoted, in the state that wants it. Until next week.

Forward this to the planner still using 2023 cost assumptions — they need it more than you do.

FERC gave itself until June to rewrite how giant AI loads connect to the grid — and that clock now matters to every data center deal in America. Read → FERC Gives Itself a June Deadline to Rewrite the Rules for AI Power Demand