PJM says it’s ready for summer, but that only works if its own AI load forecasts are actually right. Welcome to The Data Center Daily. I’m Matt, she’s Cassidy, and today we’re asking the question nobody in the capacity market wants to answer: what happens when the gigawatts on paper never show up on the grid? We’ve got PJM’s summer readiness report, what regulators should actually require before they treat a hyperscale load request as real, NERC moving on supply chain and large-load standards, Pennsylvania’s permitting queue, and Blackstone pulling in 1.75 billion on a data center REIT IPO. It’s Friday, so we’ll try to behave — but I’m making no promises once we get to who eats the stranded asset if those forecasts are wrong. USA TODAY NETWORK, with Matthew Rink:
PJM's summer peak record was set in 2006 when load reached 165,563 megawatts. It has planned for "unlikely but plausible scenarios" of 169,100 megawatts of demand this coming summer.
But Michael Bryson, senior vice president of operations for PJM Interconnection, noted that a "new reality" is emerging: Data centers are driving greater demand for power faster than the grid can add new sources of generation.
PJM says it can handle summer 2026: 180,200 megawatts of capacity against a projected peak of 156,400. That’s about 24,000 megawatts of headroom, plus another 7,800 in demand response. On paper, that looks comfortable. On paper, sure. But that demand response cushion only works if the customers who signed up actually curtail — and data centers are not exactly in the habit of turning off to save the grid. That’s the whole point of the uptime SLA. PJM flagged a possible shortfall last summer and got through it. This year the peak forecast moved up again — 154 gigawatts to 156 — and data center load is the line item that doesn’t flatten out in July heat the way residential AC does. Right. Hyperscaler load is flat and relentless, 365 days, 8,760 hours. The grid was built around peaks. These guys are the peak, every hour. When a hyperscaler drops a multi-gigawatt interconnection request on PJM’s desk, what separates a real load forecast from a land-grab placeholder — and who’s on the hook when the numbers don’t materialize? This is the fight inside PJM right now. The Maryland Office of People’s Counsel flagged it directly to PJM’s board last October, warning that the data center land rush poses extraordinary risks to existing customers and that the 2026 load forecast, which feeds directly into upcoming capacity auction parameters, is being inflated by load adjustment requests with shaky underlying commitments. The core problem, per stakeholder comments filed in PJM’s CIFP-LLA process, is that PJM has been letting large load adjustments into the forecast before answering basic accuracy questions — basically building capacity market obligations and transmission expansion plans on top of unverified demand projections. PJM’s own Stage 3 package, presented in November, says that and proposes significant enhancements to improve confidence in load forecasting, including tighter criteria for what counts as a credible large load adjustment request. PSEG’s proposal in that same process goes further, pushing for explicit criteria tying load inclusion to things like signed interconnection agreements and contractual commitments, not just letters of intent. And per RMI’s analysis of the January 2026 PJM Board decision, the region is staring at roughly 30 gigawatts of new load hoping to connect by 2030, which means the verification problem isn’t academic — bad inputs now can turn into bad transmission capex and bad capacity procurement for years. So if PJM tightens the criteria and a data center project slips or cancels after transmission steel is already in the ground, does that stranded cost land on ratepayers? That’s the live wire in the CIFP-LLA proceeding, and PJM’s Market Monitor has now filed a formal complaint at FERC arguing the RTO hasn’t protected existing customers from exactly that exposure. Watch whether FERC acts on that complaint, and whether the cost-allocation reforms PJM’s board outlined in January actually get codified with teeth — specifically whether large loads have to post meaningful financial security before their projections move transmission and capacity markets. From Holden Mann at RTO Insider:
Standards Committee members are considering a standard authorization request to address reliability risks from the addition of data centers and other large loads. NERC has requested Standards Committee members vote on two projects remotely after their meeting scheduled for May 20 was canceled.
NERC’s Standards Committee is moving toward a formal standard authorization request — a SAR — to address reliability risks from large load additions, and they’re specifically calling out data centers. They’re voting remotely after the May 20 meeting got canceled, so this is moving fast. A SAR is the starting gun, not the finish line. We’re talking years before an enforceable standard lands. Meanwhile data centers are interconnecting right now. The grid operators are already holding the bag. Worth noting, NERC dropped a Level 3 Large Loads Alert back on May 4 — that’s their most urgent advisory tier. A SAR on top of that tells you the committee thinks the existing reliability framework has real gaps, not just edge cases. Supply chain is the other piece here, and honestly that one worries me more than load forecasting. You’ve got foreign-manufactured components sitting inside critical grid infrastructure and NERC is only now scoping a standard. That is not fast. PA Environment Digest Blog, with David E. Hess:
On May 12, the Department of Environmental Protection told the DEP Citizens Advisory Council it is actively considering permits for 17 A.I. data center proposals across the state and is aware of 34 others that have been proposed, but have not yet submitted permit applications.
Pennsylvania DEP told its Citizens Advisory Council on May 12 that it’s actively permitting 17 AI data center proposals statewide, and tracking 34 more that haven’t even filed applications yet. Seventeen in the queue, thirty-four in the rumor pile. Thirty-four proposals that haven’t submitted a single permit app — that’s not a pipeline, that’s a press release graveyard. And DEP’s point man had to clarify that the fast-track program doesn’t mean you skip the line, you just get to stand in it faster. DEP Deputy Secretary John Ryder put it bluntly: data centers get the same permits and meet the same standards as any large-scale commercial development. Which apparently had to be said out loud to the Citizens Advisory Council. The fact that he led with “you can’t turn on the evening news in Pennsylvania without hearing about data centers” tells you how much political pressure is landing on a permitting agency that was not built for this volume. Here's The Economic Times:
New York: Blackstone raised $1.75 billion through the initial public offering of Blackstone Digital Infrastructure Trust, underscoring strong investor appetite for artificial intelligence infrastructure assets. The Blackstone-backed real estate investment trust (REIT), focused on acquiring data centers benefiting from the AI boom, sold 87.5 million shares at $20 apiece, according to a statement on Wednesday.
Blackstone Digital Infrastructure Trust just priced its IPO at 1.75 billion — 87.5 million shares at 20 dollars each. The vehicle is targeting already-built, already-leased data centers in the 250 million to 1.5 billion dollar range, tenanted by investment-grade hyperscalers. So it’s a REIT that buys stabilized assets with locked-in hyperscaler leases and packages the yield for retail investors. Fine. But “AI infrastructure trust” is doing a lot of marketing work here — this isn’t Blackstone building anything, it’s Blackstone buying what somebody else already built and sweated through permitting. To be fair, the filing is explicit — they want operating assets, not development risk. Investment-grade tenants, existing leases. That’s a real and distinct product from a development play, even if the AI branding is a little thick. Sure, but I want to know the lease durations and what the power costs are baked into those rents. If the hyperscaler has a short renewal window and power prices move, who eats that? The REIT unitholders, that’s who. If The Data Center Daily helps you stay ahead, take a moment to subscribe and leave a review wherever you’re listening. It’s a small thing that really helps other people find the show.
We’ve put links to all of today’s stories in the show notes, so if there’s a thread you want to follow more closely, you can pick it up there.
That’s The Data Center Daily for this Friday, May 15th. Thanks for listening, and have a good weekend. This is a Lantern Podcast.