AI data centers are picking fights with utilities over nuclear reactors, water rights, and who eats the bill when the campus never gets built. This is The Data Center Daily — and today we’re tracking how hyperscaler load growth is forcing every layer of the grid stack to make decisions it was never really built to make. Kentucky’s betting on SMRs, somebody drained thirty million gallons without writing a check, and now PUCs are finally asking the real question: what happens when the AI campus ghosts the interconnection queue? That tariff-design question is the one that actually matters for ratepayers. We’re getting into upfront payments, exit fees, curtailment rights, all of it. From Jennifer Green at The Cool Down:
Louisville Gas and Electric and Kentucky Utilities are drawing fresh attention after announcing a collaboration with X-energy to explore leveraging small modular reactors — or SMRs — as data center growth and other large electricity demands put increasing pressure on the grid. In early May, the Kentucky utilities began "early project feasibility activities" with X-energy to study whether SMRs could help support "long-term grid reliability across the Commonwealth," according to the Lane Report.
Louisville Gas and Electric and Kentucky Utilities are still in early feasibility work with X-energy on SMRs — and I mean early. This is a study, not a signed contract, not a licensed site, not a construction timeline. Every utility in the country is waving the SMR flag right now because data center load is blowing up their planning assumptions. Kentucky’s at least being honest that this is a study — but ‘early project feasibility activities’ is a very long way from a reactor on the grid. X-energy is one of the more serious players in the space — they’ve got a DOE cost-share agreement and a deployment target with Dow in Texas — but Kentucky ratepayers should know no SMR has been energized on a U.S. commercial grid yet. That’s the baseline. And when the utility says SMRs support ‘long-term grid reliability across the Commonwealth,’ I want to know who’s on the hook if the timeline slips five years. Because the data centers aren’t waiting five years. Here's Jennifer Green at The Cooldown:
In a 2025 letter to QTS, the county said the company owed $147,474 for more than 29 million gallons of water. According to Politico, the utility director estimated roughly four months of unpaid use, while QTS put the span at about nine to 15 months.
QTS in Fayetteville, Georgia — 29 million gallons, two industrial hookups, one of them connected without notice to the utility, the other not on the billing account. The county sent a letter, and QTS paid $147,474 in back charges. Neighbors had low water pressure for months while a data center campus was pulling tens of millions of gallons off the books. And QTS’s explanation is basically, ‘the smart-meter rollout confused us.’ Thirty million gallons of confusion. The dispute over the duration matters. The county says four months of unpaid use, QTS says nine to fifteen. That’s not a rounding error — that’s a completely different story about how long this went undetected. And $147,000 for 29 million gallons is a rounding error for a QTS campus. The real cost landed on the subdivision — low pressure, no explanation, no compensation. That’s who ate it. As PUCs move from one-off negotiations to formal large-load tariff frameworks, what’s actually in the rules that keeps other ratepayers whole if an AI campus never ramps, or walks away mid-contract? The short answer is: the toolkit exists, but the rollout is uneven, and the math is only now getting stress-tested. E3 published a guidebook late last year laying out the core instruments — upfront contributions in aid of construction, minimum bill guarantees, exit fees, collateral requirements, and explicit curtailment rights — and they’re clear that no single lever is enough; the question is how they stack. On the construction-cost side, Wisconsin’s PSC just rewrote We Energies’ proposal to require large-load customers to cover the full cost of generation and transmission infrastructure built specifically for them, with no cost socialization onto the general rate base. AEP Ohio went a similar direction: PUCO approved a data-center-specific tariff that, per the KJK legal analysis, explicitly rejects cost shifting to other ratepayers and puts a higher infrastructure-cost burden on the data center operator — tech firms appealed and lost. The Oklahoma Corporation Commission’s Public Utility Division flagged the asymmetry clearly in a June 2025 staff memo: these loads top 100 MW per site, are unusually consistent, and demand high reliability, but the economic-development case is weaker than advertised because headcount is low and there’s limited supply-chain multiplier effect — which is exactly why ratepayer protection, not just economic-development incentives, has to be the design frame. E3’s comparative analysis of tariffs across U.S. utilities, published this spring, found that while these tariffs have proliferated fast, almost no one has translated the tariff structures into actual levelized-cost numbers — so regulators are approving mechanisms without a rigorous public accounting of whether the protection is actually sized to the stranded-cost risk. If the E3 analysis shows no one’s actually run the numbers on whether these tariff structures cover the stranded-cost exposure, how are PUCs deciding whether collateral levels or minimum bills are sized correctly? Mostly on precedent and negotiation, not bottoms-up modeling — which is the core vulnerability E3 is pointing to. What to watch: Colorado’s PUC directed Xcel Energy to develop new large-load interconnection provisions in January, and FERC simultaneously ordered PJM to stand up co-location tariff rules, so federal and state frameworks are converging in real time. The cost-allocation methodology in those proceedings is going to set a national reference case. The next 12 to 18 months of rate cases — Wisconsin, Ohio, Colorado, and whatever follows PJM’s co-location order — are where the protection architecture either gets properly quantified or doesn’t. Here's Catherine Muccigrosso at Charlotte Observer:
The project, now nearly three years into development, has become a lightning rod for the Charlotte region’s debate over technology growth, economic incentives and environmental sustainability. The accelerated demand for artificial intelligence is driving the hyperscale data center rise for more complex computing and apps like ChatGPT and Gemini. And the Charlotte region is becoming a rising development hub.
QTS is three years into an eight-billion-dollar campus on nearly 800 acres in York County, South Carolina — and they’re now on their third public town hall, this one at an elementary school in Lake Wylie with 60 residents and 56 questions in 90 minutes. Fifty-six questions, seven panelists, Central Electric, York Electric, and Duke Energy all in the room — that’s a pretty complicated power stack for a project that’s been described as a straightforward economic win. Somebody in that room was asking who picks up the transmission upgrade tab. Duke is handling most of the transmission line upgrades, per the Observer. Whether that cost gets socialized onto the ratepayer base or stays with QTS is exactly the kind of detail that never makes the press release. Third public meeting and they’re still answering basic questions about daily-life impacts. That’s not a company winning hearts — that’s a company doing the minimum to keep the permits moving. Here's Mike Clair at International Business Times Australia:
Under the agreement, IREN will provide NVIDIA with a five-year managed GPU cloud services contract valued at approximately $3.4 billion for the chipmaker's internal AI and research workloads. In return, NVIDIA received a five-year warrant to purchase up to 30 million IREN shares at $70 each, representing a potential $2.1 billion equity investment subject to regulatory approvals and performance milestones.
NVIDIA and IREN announced a strategic partnership on May 7th — up to five gigawatts of AI infrastructure, anchored by IREN’s two-gigawatt Sweetwater campus in Texas. The commercial anchor is a five-year managed GPU cloud contract, three-point-four billion dollars, for NVIDIA’s internal workloads. In return, NVIDIA gets a warrant to buy thirty million IREN shares at seventy dollars a pop — that’s a two-point-one billion dollar potential equity stake, subject to milestones and regulatory sign-off. Let’s be precise about what ‘up to five gigawatts’ means: it means the pipeline, not the energized load. Sweetwater is two gigawatts on paper. How much of that is permitted, how much has transmission queue positions, and how much is just land with a vision deck? The warrant structure is genuinely interesting — NVIDIA isn’t writing a check today, it’s buying the option to own a piece of IREN if the buildout hits milestones. That’s alignment, not a blank check. The IBT headline calling this a reason for both stocks to ‘explode’ is doing a lot of work that the actual filing language does not. And the three-point-four billion cloud-services contract is NVIDIA paying IREN to run NVIDIA’s own GPUs — which is a real revenue commitment, but let’s not confuse it with third-party demand. The customer and the chip supplier are the same entity. That’s a very specific kind of deal. Got a story idea, a correction, or something you want us to dig into next? Send us a note at datacenterdaily at lantern podcasts dot com. We read the inbox, and your feedback helps shape the show.
You’ll find links to every story we covered today in the show notes, so if something deserves a closer read, that’s the place to start.
That’s The Data Center Daily for this Thursday, May 14th. This is a Lantern Podcast.