If you're planning new data center capacity anywhere inside PJM Interconnection's footprint — the grid operator covering a large stretch of the mid-Atlantic and Midwest, from Chicago to Northern Virginia to the Carolinas — the ground under that plan just shifted. On July 27, 2026, PJM proposed a one-time backstop capacity auction to close a 6.8-gigawatt supply shortfall, and alongside it, a new framework that would push large electricity users, data centers chief among them, to the front of the curtailment line if they haven't secured their own generation. This is PJM data center power policy moving from theoretical grid-planning conversation to a concrete deadline with a concrete number attached, and every enterprise infrastructure team with capacity plans in this territory needs to understand exactly what it means before June 1, 2027 arrives.
What PJM actually proposed, and why now
PJM's backstop auction is the reactive half of the announcement: a one-time mechanism to buy enough additional generation capacity to cover the 6.8 GW gap between what the grid needs and what's currently committed. That gap didn't appear overnight. PJM estimates roughly 70 GW of new large-load demand arriving by 2038, driven substantially by AI data centers, while about 15 GW of generation capacity has retired from the grid since 2022. Run those two trends against each other and you get a widening structural gap between supply and demand that a single backstop auction only patches temporarily.
The forward-looking half of the announcement is the more consequential one for IT strategy: a new curtailment framework specifically targeting "Large Load" customers that haven't locked in their own power supply. PJM defines a Large Load as any customer with 50 megawatts or more of cumulative peak demand at a single site within a one-mile radius — a threshold that captures the overwhelming majority of hyperscale and enterprise AI data center campuses currently being planned or built across PJM territory. Starting June 1, 2027, any new Large Load in that category that hasn't secured its own generation capacity or otherwise locked in its own power supply will be first in line for curtailment when the grid approaches emergency conditions — ahead of PJM's existing pre-emergency load-management steps that currently apply to other customer classes.
That ordering matters. It isn't a blanket ban on new data centers, and it isn't a moratorium. It's a reordering of who gets cut first when supply gets tight, and it specifically penalizes exactly the customer profile that has driven most of the recent demand growth: large, power-hungry AI infrastructure that plugged into the grid without bringing new generation along with it.
What "bring your own power" curtailment risk means in practice
For an organization evaluating a new data center site or a large colocation footprint in PJM territory, this changes the calculus in a way that's easy to underestimate if you're reading it as a distant regulatory filing rather than an operational constraint. A facility crossing the 50 MW cumulative peak demand threshold at a single site — which is a modest size by modern AI training or inference cluster standards — is now facing a binary choice well before it ever signs a colocation contract or breaks ground: either arrange dedicated generation capacity (on-site gas turbines, a power purchase agreement tied to specific new generation, or another arrangement PJM recognizes as "bringing your own power"), or accept that this facility will be curtailed before essentially everyone else on the grid during a supply emergency.
Curtailment in this context isn't a hypothetical edge case. PJM's own capacity market data for the 2026–2027 delivery year shows the system has just enough supply to remain reliable, with little margin to spare, at a capacity market cost of roughly $16 billion for that delivery year alone. Tight margins mean emergency conditions are a realistic, recurring risk rather than a once-a-decade tail event. A data center that's first in the curtailment queue during a tight-margin grid isn't buying insurance against an unlikely scenario — it's buying insurance against a scenario PJM itself is telling the market to expect.
For IT leaders, the practical translation is this: any new PJM-territory site in the planning stage now carries a power-sourcing decision that used to be optional and is becoming close to mandatory for any deployment that matters operationally. If your workload is latency-sensitive, revenue-critical, or supports a service with uptime commitments to your own customers, running it in a facility that hasn't secured its own generation is now a deliberate risk acceptance, not a default.
The broader 2026 pattern: AI ambition outrunning grid physics
This PJM announcement doesn't exist in isolation — it's the latest and most concrete instance of a pattern that's been building across the data center industry through 2026: AI infrastructure buildout plans colliding with the physical limits of the electric grid. The energy intelligence firm Currence has estimated that between 30% and 50% of large-scale data center capacity expected to come online in 2026 will likely be delayed, with power constraints cited as one of the primary reasons. That's not a niche supply-chain hiccup; it's roughly a third to half of the industry's near-term capacity plans running into the same wall PJM is now formalizing rules around.
What makes the PJM move notable is that it's a grid operator turning that abstract power-constraint narrative into an enforceable rule with a specific effective date. Previous discussion of the data center power crunch tended to focus on transformer lead times, interconnection queue backlogs, and generation retirements — real constraints, but ones that mostly showed up as delay, not as an explicit curtailment hierarchy. PJM's framework is different: it's not just saying "there isn't enough power," it's saying "and when there isn't enough, here's exactly whose power gets cut first." That's a meaningfully sharper signal for anyone doing AI data center curtailment risk assessment as part of site selection, because it converts a diffuse industry-wide constraint into a jurisdiction-specific, date-specific compliance question.
It's also a preview of where other grid operators are likely headed. PJM is one of the largest and most closely watched regional transmission organizations in the country, and its capacity market dynamics — the widening gap between retiring generation and AI-driven load growth — aren't unique to its footprint. Enterprises with data center or colocation exposure in other regions should read this as an early signal, not a PJM-specific oddity.
Cloud versus colocation versus on-prem, reconsidered
This changes the input variables for the perennial cloud-versus-on-prem-versus-colocation debate in a specific way. Historically, that decision has weighed cost, control, compliance, and latency. Power availability and curtailment exposure now belong on that list as a first-class factor, at least for any deployment sized and located to fall inside PJM's Large Load definition.
For organizations leaning toward colocation or purpose-built on-prem capacity in PJM territory, the question is no longer just "does this provider have power," but "has this provider — or will this provider, for my specific footprint — secure generation that exempts us from the new curtailment priority." That's a materially different due-diligence question, and it's one most colocation contracts signed before this announcement almost certainly don't address.
For organizations using major public cloud providers, the exposure is more indirect but not zero. Hyperscalers operating in PJM territory face the same Large Load thresholds for their new capacity, and if a meaningful share of a provider's regional footprint ends up curtailment-exposed, that risk eventually surfaces as regional capacity constraints, pricing changes, or provisioning delays for customers — echoing the same power-driven capacity friction already showing up industry-wide. The practical implication is that "we use the cloud, so this is someone else's problem" is a weaker argument in 2026 than it would have been a few years ago. It's worth asking your cloud provider directly whether their PJM-region capacity — current and planned — is covered by dedicated generation arrangements, rather than assuming the question is irrelevant because you don't own the building.
For workloads that can tolerate looser latency requirements or that don't need to sit in PJM territory specifically, geographic diversification away from PJM's most constrained sub-regions becomes a more concrete mitigation than it was a year ago. That's not a universal answer — plenty of workloads have real reasons to stay close to PJM's dense population centers and existing infrastructure — but it belongs in the conversation now in a way it didn't before this proposal.
The questions to ask your provider before June 2027
Whatever combination of cloud, colocation, and on-prem your organization runs in PJM territory, there's a concrete set of questions worth raising with providers and internal facilities teams now, well ahead of the June 1, 2027 threshold, rather than after a curtailment event forces the conversation.
Ask whether your current or planned site's cumulative peak demand crosses the 50 MW threshold within a one-mile radius — including any nearby facilities that might combine with yours under PJM's definition in ways that aren't obvious from your own metering alone. Ask whether the facility has secured, or has a credible plan to secure, its own generation capacity or a power arrangement that PJM would recognize as satisfying the bring-your-own-power requirement, and ask for the actual documentation rather than a verbal assurance. Ask what the provider's contractual curtailment exposure looks like today, and whether that changes after June 2027 — specifically, whether your workload's contract includes any protection against being deprioritized ahead of other customers on the same grid. And ask what the provider's own timeline is for closing that gap, since "we're working on it" is a materially different answer twelve months before the deadline than it is one month before.
None of these questions require exotic engineering answers. They require providers to be specific about power sourcing in a way that, until this year, most data center and colocation contracts simply didn't need to be. The organizations that ask them now, with more than nine months of runway before the deadline, will have real options — renegotiating contracts, relocating planned capacity, or securing generation commitments of their own. The organizations that wait until the deadline is close will be negotiating from a position where the leverage has already shifted to whoever secured power first.
PJM's proposal is a regulatory filing today, but it's a countdown clock for anyone with data center capacity plans in its territory. The underlying physics — AI demand growth outpacing grid generation — isn't going away regardless of how this specific proposal is finalized, which means the bring-your-own-power logic it's built around is likely to outlast the exact mechanics PJM lands on. Treat June 1, 2027 as the deadline to have already answered these questions, not the date to start asking them.