Power & Infrastructure Weekly — Mar 09, 2026

Power, cooling, and water stopped being separate planning problems this week — and the paperwork finally caught up. Google and Xcel announced what would be the world's largest grid battery by energy capacity, a 100-hour iron-air monster in Minnesota, while Panasonic started taking commercial orders for liquid-cooling hardware aimed at AI data centers in Europe. Meanwhile, the federal PFAS regulatory picture fractured into two contradictory tracks that will force water utilities to spend money they weren't sure they'd need to spend. The through-line: the physical systems that keep civilization running are being redesigned simultaneously, under time pressure, with capital flowing faster than the rules can keep up.

If you want to see what a renewables-heavy AI grid actually looks like in hardware, here it is. Google and Xcel Energy announced plans for a 300 MW / 30 GWh iron-air battery from Form Energy in Minnesota — paired with roughly 1.6 GW of new wind and solar — to firm power for a new Google data center. Xcel calls it the largest grid battery by energy capacity announced to date, anywhere.

The chemistry matters. Iron-air batteries discharge for roughly 100 hours, which makes them a fundamentally different tool than the 2–4 hour lithium-ion systems that dominate today's market. This isn't about smoothing an afternoon solar dip; it's about bridging multiday wind droughts in a cold, winter-peaking grid. Form Energy's domestic factory plans also sidestep the tariff and foreign-entity-of-concern headaches plaguing lithium-ion supply chains.

What's easy to miss: Google's Minnesota buildout also includes 400 MW of grid-scale batteries from Minnesota Power, meaning multiple storage providers and operating modes will be orchestrated around a single hyperscale campus. That's a preview of the multi-asset, multi-utility coordination model that every large data center will eventually require. Watch for final regulatory approvals, cost recovery treatment, and any disclosed capex per kilowatt-hour — those numbers will determine whether iron-air moves from interesting pilot to standard RFP line item.

Back in December, FERC found — unanimously, 5-0 — that PJM's existing rules for generation co-located with large loads behind the interconnection point were unjust and unreasonable. PJM filed its initial tariff brief in February. Responses are due March 18. That's next week.

Here's why this isn't just a regulatory procedural note. FERC's logic says that if a 1,000 MW generator uses 900 MW to serve a co-located data center and only leans on the grid for 100 MW, PJM need only study the 100 MW. If that holds, it dramatically compresses interconnection study timelines for on-site generation deals — and the three new transmission service categories being defined right now (firm contract demand, non-firm contract demand, interim non-firm) will become the template every grid operator eventually follows.

PJM is also quietly building complementary fast-lane infrastructure: a price collar on capacity auctions plus an expedited interconnection track explicitly aimed at large new loads. Meanwhile, states are drawing their own lines — Dominion's new large-load tariff class in Virginia requires long contract terms and high reserved-capacity payments to protect retail ratepayers. Federal rules set the template; states and utilities still hold the cost-allocation levers. For IPPs, hyperscalers, and energy lawyers, March 18 is the last clear shot to shape the rate structure before it hardens.

Developers plan to add roughly 24 GW of utility-scale battery storage in 2026 — nearly double last year's record 15 GW — as part of an 86 GW total capacity pipeline dominated by solar and batteries. Texas alone accounts for about half the planned battery capacity. BloombergNEF pegs the four-hour battery benchmark at roughly $78/MWh, which makes storage competitive with many new gas peakers.

The wrinkle nobody's talking about loudly enough: tariffs and foreign-entity-of-concern restrictions have pushed battery cell and pack costs 56–69% higher in certain construction cost assumptions since early 2025. Projects underwritten on 2024 price decks are most exposed. Layer on FEOC rules for federally funded projects and you add supply-chain complexity that can delay procurement or force developers toward higher-cost domestic sourcing.

A 24 GW buildout at higher per-unit costs, competing for transformers and switchgear with the rest of the 86 GW pipeline, is a recipe for cancellations and scope changes. Watch EIA's upcoming generator inventory updates for the first hard evidence of deferrals, and Q1/Q2 earnings from CATL, BYD, and Fluence for margin signals.

The federal PFAS drinking water framework is splitting into two contradictory tracks, and water utilities have to make capital decisions anyway. In February, EPA moved to begin rescinding standards for four of the six regulated PFAS compounds (PFHxS, PFNA, GenX, and PFBS), while simultaneously signaling a spring 2026 rule that would keep PFOA and PFOS MCLs (maximum contaminant levels — the legal ceiling for contaminants in drinking water) in place but extend compliance deadlines to 2031.

Meanwhile, states aren't waiting. Virginia's HB880 would require wastewater plants to test biosolids for PFAS. Illinois is pushing monthly PFAS and microplastics testing in Lake Michigan systems. Every state is moving toward some form of PFAS reporting or restriction, even as the federal floor shrinks. The TSCA reporting window (April 13 – October 13, 2026) will generate the data inventory that underpins enforcement for years.

The practical consequence: utilities that paused treatment capital decisions waiting for regulatory clarity may be waiting for something that never resolves cleanly. The federal floor is shrinking for four compounds while state ceilings are tightening independently. Treatment technology vendors should be watching state dockets more carefully than the Federal Register right now.

MISO's draft 2026 Transmission Expansion Plan, released March 4, totals $8.8 billion — and roughly $3.1 billion is explicitly tagged to handle rapid load growth from data center interconnections in the Midwest. This is one of the first major U.S. transmission plans where data centers show up as a named driver of multi-billion-dollar grid capex instead of being buried in generic load forecasts.

If you're an IPP or storage OEM, the signal is that new 345 kV lines and substations will be co-sited with hyperscale clusters, not just wind farms. If you're a hyperscaler, the White House pledge to fund grid upgrades just got a price tag to negotiate against. The plan is still in draft — expect project lists and cost allocations to shift before final approval — but the directional message is clear: data centers are now a transmission planning category of their own.

Panasonic CDUs and free-cooling chillers for AI data centers — Panasonic's HVAC division began taking orders in Europe on March 4 for coolant distribution units rated 400 kW and 800 kW, plus free-cooling chillers at 800 kW and 1,200 kW, specifically designed for generative AI data centers. The CDUs handle liquid loops to direct-to-chip cold plates; the chillers run compressor-free below about 10°C ambient. This is liquid-ready cooling moving from bespoke engineering to catalog product — expect smaller colos and edge operators to follow hyperscalers into liquid cooling faster than most mechanical contractors are prepared for.

Crestchic 600 kW liquid-cooled loadbank — UK-based Crestchic launched a purpose-built loadbank for commissioning liquid-cooled data center infrastructure, available for sale and rental, with ±0.5°C temperature accuracy regardless of flow variation. When the test equipment market builds dedicated liquid-cooling products and offers them for rental, the commissioning pipeline has grown large enough to justify the capital. This is the concrete-trucks-showing-up signal.

HRL Laboratories Low-Chill™ single-phase cooling — Developed under DOE's ARPA-E COOLERCHIPS program, HRL's technology increases processor cooling capability by 40% or reduces pumping power by more than 10×, extending single-phase liquid cooling to AI-scale densities without two-phase fluid handling complexity. By enabling hotter coolant loops, it can eliminate evaporative cooling entirely — a major water-use win. HRL is seeking commercialization partners; this is lab-validated, not yet in production, but the ARPA-E provenance adds credibility.