The Lyceum: Power & Infrastructure Weekly — May 23, 2026

Wind and solar generated more electricity globally than gas for the first time ever in April 2026 — and they did it during an active Middle East energy crisis that was supposed to be gas's moment. Meanwhile, a new arXiv preprint quantified what infrastructure planners have been quietly dreading: U.S. data centers may need new water capacity equivalent to what New York City uses every day. And quietly, NERC handed every transmission planner in North America a new rulebook for studying data center risk. Power, cooling, and water aren't three stories this week — they're one story about a civilization that built its AI future without fully accounting for the physical infrastructure underneath it.

• Barakah Unit 4 (Emirates Nuclear Energy Corporation / KEPCO): Fourth and final APR-1400 reactor connected to the UAE grid, completing the Arab world's first nuclear plant at roughly 5.6 GW total capacity.
• Risk Mitigation for Emerging Large Loads Guideline (NERC): New reliability guideline directing transmission planners to run steady-state, dynamic, and short-circuit models specifically for large-load interconnection studies.
• Proposed PFAS Rescission Rule (EPA): Federal Register proposal to rescind drinking water regulations for PFHxS, PFNA, HFPO-DA (GenX), and Hazard Index mixtures; comment period through July 20, 2026.
• Proposed PFOA/PFOS Compliance Extension (EPA): Companion proposal upholding PFOA/PFOS maximum contaminant levels but offering utilities a two-year extension to 2031.
• Bradford Edge Data Center Planning Approval (Deep Green): 5.6 MW facility approved with closed-loop cooling system designed to feed waste heat into the Bradford Energy Network.

The energy transition just crossed a threshold analysts have forecasted for years — and it crossed it at the worst possible moment for the counterargument.

According to new analysis from independent energy think tank Ember, wind and solar generated more electricity than gas globally for the first time ever in April 2026. The two sources produced 22% of the world's electricity on the month — a record 531 terawatt-hours — versus 477 TWh from gas plants at 20%. The growth was broad: wind and solar output rose 13% year-on-year globally in April 2026, with the UK up 35%, Chile up 24%, Australia up 17%, China up 14%, the EU up 13%, the U.S. up 8%, and Brazil up 4%.

What makes this remarkable is the timing. The milestone landed during the first full month of the recent global energy crisis triggered by conflict in the Middle East — the crisis that was supposed to send everyone scrambling back to fossil fuels. Instead it revealed how much the grid has already changed. Ember found no evidence of a broad global shift from gas to coal.

One honest caveat: April is structurally the most likely month for this crossover, combining strong Northern Hemisphere wind with rising solar output and relatively low demand. Summer data will tell us whether this holds when cooling load pulls gas back in. If summer data shows the gap holding or widening, gas peaker economics get rewritten in real time. If it narrows sharply, April was a seasonal artifact dressed up as a trend.

The Arab world's first nuclear plant just completed its full buildout, and the geopolitical implications are bigger than the megawatts.

According to Al-Monitor, the UAE's Barakah Nuclear Energy Plant — built by Korea Electric Power Corporation (KEPCO) in Abu Dhabi — connected its fourth and final reactor unit to the national grid this week. The plant now operates four APR-1400 pressurized water reactors with a combined capacity of approximately 5.6 gigawatts, enough to supply roughly a quarter of the UAE's electricity demand from a single zero-carbon facility. Construction began in 2012.

Barakah is the proof-of-concept that the Middle East can build and operate large-scale nuclear — and it lands as Saudi Arabia, Egypt, and Turkey are in active procurement for their own programs. KEPCO's government-backed, export-oriented model is now a direct competitor to Westinghouse, EDF, and Rosatom, and Barakah is the only large-scale Korean-built reactor program completed on anything close to schedule in the last decade. If KEPCO wins the next Saudi or Egyptian tender on the back of this delivery, the U.S. and French nuclear export programs lose their last credibility argument. The other thing to watch: Barakah's gigawatts free up natural gas previously burned for domestic UAE power, redirecting it toward LNG export — a supply signal worth tracking as European buyers compete for spot cargoes.

The question of who pays for grid upgrades that data centers require has moved from academic debate to regulatory docket, and California just put a number on it.

The California Public Utilities Commission is considering rules that would require data centers to bear a larger share of the transmission and distribution costs their load growth triggers — shifting away from the current model, where all ratepayers share upgrade costs, toward a "cost-causer pays" framework targeting large industrial loads. According to Consumer Reports, by 2028 data centers could use 12% of all U.S. electricity, per a 2024 survey by Lawrence Berkeley National Laboratory; in Virginia, already home to nearly 600 facilities with more than 100 more proposed, data centers accounted for almost 40% of state consumption in 2024, per a Bloomberg News analysis.

This is the same fight playing out at FERC in the PJM co-location proceeding, but at the state level with a retail rate angle. If California finalizes a cost-causer framework, it becomes a template that other state PUCs — and eventually FERC — will reference. The signal that tells you which path this is on: whether hyperscalers respond by accelerating behind-the-meter generation and storage (the avoidance signal) or by quietly accepting the new tariff structure as the price of California compute (the acquiescence signal). Watch the formal rulemaking docket for the comment period opening.

The chip shortage got all the headlines. The water shortage is the one that could actually stop construction.

A preprint study published on arXiv this month — not yet peer-reviewed — analyzed public government records and water utility data to put a specific number on data center water demand for the first time. If 2024 water-use intensity persists, U.S. data centers could collectively require 697 to 1,451 million gallons per day of new water capacity through 2030 — comparable to New York City's average daily supply of roughly 1,000 million gallons per day. Under an optimistic scenario with 10% annual reduction in water-use intensity, demand drops to 227–604 million gallons per day. The total valuation of the new water capacity need is on the order of $10 billion, reaching up to $58 billion in the high-growth case.

The mechanism is evaporative cooling — the same technology that has cooled data centers for decades, working by evaporating water to shed heat. As rack densities climb past 100 kW for AI workloads, water consumption climbs with them. Direct-to-chip liquid cooling can dramatically reduce water use, and a related arXiv analysis found that advanced cooling technologies can cut cooling energy by up to 50%, but retrofitting existing facilities is expensive and most new hyperscale campuses still use hybrid air/evaporative systems.

The buried story is jurisdictional. The $10–58 billion in new water infrastructure isn't being built by hyperscalers — it falls on local utilities and ratepayers. Watch whether water availability disclosures become a standard part of data center permitting in drought-stressed states, and whether the paper clears peer review — if it does, expect EPA and state water boards to cite it directly in permitting decisions.

NERC published its 2026 Summer Reliability Assessment this week alongside something that got almost no coverage: a new Reliability Guideline titled Risk Mitigation for Emerging Large Loads.

The guideline directs transmission planners and planning coordinators to run steady-state, dynamic, and short-circuit models specifically to assess large-load reliability impacts. NERC's core concern, stated plainly in the document: large loads frequently seek grid connection faster than transmission can be built, and demand may outstrip generation supply as large loads are added faster than new generation resources come online. The guideline also flags that large loads typically require firm service, limiting their ability to provide demand response or operate as flexible resources — which directly reduces operating reserves.

The summer assessment itself told a more reassuring story. NERC concluded that all assessment areas should have adequate resources to meet normal summer peak demand in 2026, with significant new solar, battery storage, and gas capacity strengthening reserve margins. But the guideline is the more durable signal. It's the first time the reliability organization has codified exactly what studies must be run before a data center gets a queue position. Reliability guidelines aren't enforceable, but they reliably become the basis for future standards. The signal to watch: ISO/RTO compliance filings that start citing this document in interconnection study methodologies over the next 12–18 months. If they do, the queue gets longer before it gets shorter.

A South Korean reactor humming on the edge of the Empty Quarter, a Bradford district heat network warmed by GPUs, and a New York City's worth of water vanishing into evaporative cooling towers somewhere in the American desert. Somewhere in EPA's Federal Register filings, four PFAS compounds are quietly getting unregulated — great news if you sell drinking water, terrible news if you sell granular activated carbon, and either way the molecules don't care what the docket says.

Stay grounded.

Forward this to the one person you know who'll actually read the NERC guideline.