Integrating Data Centers into the PJM Grid

In-Depth Explainer: PJM and the Large Load Surge

If you are trying to bring a large data center online anywhere in the PJM footprint, you have probably already felt the tension at the heart of this article: you want to connect quickly, and the grid operator wants to make sure the lights stay on for everyone else while you do it. Over the course of 2025, PJM Interconnection moved from talking about that tension to filing concrete rules to manage it. This explainer walks through what PJM is proposing, why, and what it means for anyone planning a large load addition.

The story unfolds in three altitudes: PJM's May 2025 stakeholder workshop laid out the concepts; the November 2025 Critical Issue Fast Path (CIFP) Stage 4 package hardened them into filed proposals with dates and dollar figures; and a real-world forecasting dispute shows exactly where the friction lands for a developer's pipeline.

Why this is happening: a demand curve that bent upward

For most of the last fifteen years, electricity demand in the PJM region was essentially flat. That era is over. PJM's 2025 summer-peak forecast now climbs from roughly 155,000 MW today toward more than 230,000 MW by the mid-2040s. More striking than the absolute number is the shape of the curve: laid against every prior year's forecast, the older lines crawl along gently while the 2025 line bends sharply upward beginning around 2026. The growth is both large and recent — it appeared in the forecasts faster than the system was built to absorb it.

The cause is no secret. The new demand is dominated by data centers, increasingly by very large hyperscale facilities tied to artificial intelligence and cloud computing. PJM frames this growth as a genuine opportunity — economic development, jobs, investment, technological leadership, even national security — and has been explicit that it does not see its role as turning these loads away. PJM cannot refuse load integrations; utilities and load-serving entities tell PJM what is coming, and PJM has to plan to serve it.

The difficulty is that wanting to serve large loads and being able to serve them reliably are different things. PJM has identified four problems that must be solved together:

• Supply may not keep up. Forecasted generation may be insufficient to meet forecasted load as the curve steepens.
• The timing is misaligned. Load is growing fast at the same moment older generators retire, and new generation and transmission take years to build. Transitional mechanisms may be needed to bridge the gap.
• Large loads do not want to be flexible. Existing demand-side products give them too little reason to participate; data centers prefer to run continuously, and on-site back-up units face environmental limits on run time.
• Speed-to-market pushes developers off the grid. Because data centers want to energize quickly, some pursue co-location arrangements outside PJM's markets and planning — which PJM views as less reliable.

PJM's guiding principle: come in as Network Load

PJM's strong preference is for large loads to interconnect as Network Load — fully integrated into the grid, accounted for in planning, and served like any other firm load. PJM argues this delivers less operating complexity, more reliable service for critical facilities, better curtailment-priority management in emergencies, and more holistic planning.

The arrangement that worries PJM is the islanded co-located configuration, where a large load sits behind a generator with protection equipment designed to prevent it from drawing grid energy. In PJM's eight-option framing these are Options 4 and 5 — and PJM explicitly does not prefer them. Under those arrangements neither the generator nor the load pays for transmission or energy and ancillary services, the load is invisible to forward planning, and the complex relay schemes create operational risks such as power swings and transient impacts to voltage and frequency. PJM's pitch to developers is therefore: come inside the tent as Network Load, and we will give you flexible, faster ways to do it.

The three paths (and the eight options behind them)

In its response to the FERC Section 206 proceeding on co-located load (Docket No. EL25-49), PJM described eight integration options. Three already exist under its Tariff — variations on network load with separate or shared points of interconnection, and existing behind-the-meter generation rules. Five are newer. PJM has cautioned that the options are not all equally workable, are not mutually exclusive, and that state laws may limit when load can be served by anyone other than the franchised utility. Setting aside the not-preferred islanding options, the constructive menu collapses into three paths.

Path 1 — Bring Your Own Generation (Option 6)

A new large load brings incremental generation that commits to PJM's capacity market (the Reliability Pricing Model, or RPM) alongside the new demand — through ownership or a demonstrated Power Purchase Agreement. The generation need not be physically co-located; it must meet or exceed the load on an unforced-capacity (UCAP) basis. The payoff is favorable treatment: such a load would only be curtailed in a true emergency manual load-dump, much like residential load. PJM also points to Provisional Interconnection Service (enabled under FERC Order 845) as a way to interconnect generation faster — signing a provisional agreement before all studies and network upgrades are complete, saving roughly 6 to 12 months in exchange for accepting the risk that additional upgrades surface later.

Path 2 — Demand Response (Option 8)

Rather than building generation, a large load earns better terms by reducing consumption when the grid is stressed. The problem is that hyperscalers barely participate today — traditional enterprise data centers make up only about 3% of demand-response capacity, and the largest facilities have essentially no track record. PJM's diagnosis is that the barriers are fixable: emissions rules limit back-up generation run time; the ELCC accreditation method undervalues some demand response; and cost-recovery rules make participation unattractive. Proposed fixes include partnering with state and federal authorities to relax emissions limits during pre-emergency conditions, creating a dedicated data-center accreditation class, expanding cost recovery toward revenue-neutrality, and exploring seasonal or transitional products with limits on curtailment frequency.

Path 3 — Non-Capacity Backed Load (Option 7)

This is the explicitly transitional option and the most novel. It becomes available only when a resource-adequacy test fails — when projected supply falls short of the reliability requirement. In that situation, a large load above a threshold can integrate at reduced cost in exchange for agreeing to curtail before emergency procedures begin. The trade is clear: the load does not pay for capacity and has minimal or no impact on capacity prices, but accepts pre-emergency curtailment, with residential customers given priority in extreme conditions. Importantly it is still treated as Network Load for planning and remains fully part of transmission planning, so the grid is still built to serve it. If an electric distribution company declines the option on a load's behalf, that load falls back to status-quo curtailment rules.

From concept to commitment: the CIFP Stage 4 package

The May workshop was PJM thinking out loud. By November 2025 the concepts had hardened into a package filed under the Critical Issue Fast Path process — PJM's expedited route for urgent issues. The trigger was concrete: the capacity auction for the 2025/2026 delivery year returned tight system conditions, confirming the supply-shortfall concern was not hypothetical. The package has five components; three matter most to developers.

Enhanced load forecasting

Beginning with PJM's 2027 Load Forecast, new elements include a step for state commissions to review large-load adjustments before the forecast is finalized; a duplicative-request disclosure rule (with the penalty that unflagged-but-duplicative requests can be removed entirely); added third-party and national cross-check review; and a transparency requirement that customer non-disclosure agreements must permit utilities to share large-load information with PJM. Several rules are already in place and directly shape how much of a project's nameplate lands in the forecast: loads tied to a Construction Commitment or Electric Service Obligation contract are considered for inclusion; projects further out without firm commitments get de-rated; PJM applies a default 3-year ramp rate absent the utility's own; it uses a default utilization factor of around 70% unless given supporting data; and it expects documented financial commitments.

Demand-side products

PJM is reforming Price Responsive Demand (PRD) for the 2028/2029 Base Residual Auction — removing the old dynamic-retail-rate requirement, replacing it with an energy-market bid price, and aligning PRD dispatch order and penalties with demand response. Notably, PJM is not proposing broader new demand-response products in this package; a limited-duration product some stakeholders wanted cannot be implemented until the 2029/2030 auction.

The Expedited Interconnection Track (EIT) — the one to watch

For developers, this is the headline: a new track, targeted for mid-2026, running in parallel to and outside the normal interconnection Cycle Process, built for shovel-ready generation tied to new large load. Its key parameters

The trade is explicit: speed and certainty in exchange for real financial commitment and inflexibility. This is Path 1 turned into an actual filing.

CIFP Phase II and beyond

PJM also recommended its Board invoke a narrower second phase to incentivize load flexibility when PJM is capacity-deficient and to reform the manual load-shedding allocation methodology and the RPM reliability-backstop mechanism. A separate post-CIFP track commits to longer-term resource-adequacy work that will not be ready before the 2028/2029 auction.

Where the friction actually lands: forecast discounting

The most relatable illustration of all this comes from how PJM treats a utility's large-load forecast. A utility working directly with its customers may compile a pipeline of data-center projects backed by signed agreements and binding financial penalties, then submit it for inclusion in the PJM load forecast. PJM applies its own methodology — and can discount that pipeline heavily. In one documented case PJM discounted all signed-agreement projects to zero for in-service dates before 2031 (citing infeasible construction and material timelines) and applied a 33% discount to projects beyond 2031. The utility pushed back, warning that aggressive discounting risks delayed transmission upgrades in-zone reliability violations, and far costlier emergency fixes later.

For a developer the lesson is direct and uncomfortable: a signed contract with your utility does not guarantee that PJM counts your megawatts in its plan. The strength of your milestones, the credibility of your in-service date, your ramp schedule, and your financial commitments all feed into whether — and how much of — your project survives PJM's discounting. The forecasting rules are not bureaucratic footnotes; they are the filter your project passes through.

Timeline and what to do now

Pulling the dates together: forecasting changes begin with the 2027 Load Forecast; the Expedited Interconnection Track and transparency measures target mid-2026; PRD reforms and many large-load rules aim for the 2028/2029 Base Residual Auction in June 2026; a limited-duration demand-response product waits until the 2029/2030 auction; longer-term resource-adequacy work comes afterward. Practical takeaways for developers:

• Plan to be Network Load, not an island. PJM's incentives, faster pathways, and favorable curtailment treatment flow to loads that integrate fully and pair with committed supply.
• Decide early which path fits. Bring-your-own-generation (possibly via the EIT) buys speed and reliability but demands capital and rigidity; demand response and non-capacity backed load trade flexibility for lower cost. They are not mutually exclusive.
• Treat your forecast submission as a deliverable. Firm contracts, demonstrable milestones, realistic ramp rates, and documented financials convert requested megawatts into counted megawatts.
• Watch the dockets and the deposits. The FERC 206 proceeding and CIFP filings will set the final rules, and the EIT deposit structure means early commitment carries real cost.
  

Part 2 — Case Studies (Confidential)

The following are anonymized, composite case studies. They contain no project names, no client names, and no locations, and they combine details from representative engagements to illustrate how PJM's large-load rules play out in practice. Figures are illustrative and rounded.

Case Study A — A hyperscale load takes the bring-your-own-generation route

Profile: A hyperscale operator planning a multi-hundred-megawatt computing campus within the PJM footprint, targeting energization roughly three years out.

Situation. The operator's first instinct was an islanded co-located design — sit behind a dedicated generator and avoid the queue. Early diligence flagged the problems PJM had signaled: the configuration is not preferred, it leaves the load invisible to planning, and the relay schemes needed to prevent grid lean introduce operational risk. It also exposed the campus to weaker curtailment treatment and an uncertain regulatory posture as the FERC 206 proceeding evolved.

Approach. We modeled the campus instead as fully integrated Network Load paired with matching incremental generation under Path 1. Because the generation did not need to be physically co-located, the operator secured supply through a PPA sized to exceed the campus peak on a UCAP basis rather than nameplate — which required deliberately over-procuring nameplate to clear the accreditation de-rate. To compress schedule, the generation interconnected via an expedited path: a large nonrefundable study deposit and readiness deposit at the paired-with-load rate, full responsibility for identified network upgrades, and three years of locked site control at application.

Outcome. The campus achieved a credible path to energization on the operator's timeline while qualifying for residential-like curtailment treatment, meaning interruption only in a genuine emergency manual load-dump. The cost was front-loaded capital and a deliberately inflexible design — no post-agreement changes to fuel type, size, or site.

Lessons for developers. First, the UCAP match, not nameplate, governs how much generation you must commit. Second, the expedited path's lower deposit is a direct reward for pairing generation with load. Third, inflexibility is the price of speed — lock your engineering before you commit.

Case Study B — A development pipeline collides with forecast discounting

Profile: A developer with a multi-gigawatt pipeline of data-center projects across several years, each backed by signed agreements and financial penalties for cancellation, relying on its utility to submit the pipeline as a Large Load Adjustment.

Situation. The developer assumed that signed agreements plus binding penalties would secure inclusion in the PJM load forecast at close to face value. When the forecast was finalized, PJM had applied aggressive haircuts: the earliest in-service years were discounted heavily on the grounds that construction and material lead times made those dates infeasible, and later years carried a substantial percentage discount. The gap between requested and counted megawatts was large enough to threaten the timing of the transmission upgrades the pipeline depended on.

Illustrative requested-vs-counted pattern (rounded, anonymized):

Approach. We treated the forecast submission as an engineering deliverable rather than a paperwork step. That meant supplying project-specific ramp schedules in place of the conservative default; documenting utilization with the contractual minimum-demand basis instead of accepting the default factor; assembling the milestone evidence PJM weighs — site control, procurement status, permitting progress, and quantified financial commitment; and proactively disclosing any potentially duplicative siting to avoid wholesale removal under the duplicative-request rule.

Outcome. The discounting did not vanish — PJM applies its own methodology regardless — but a better-substantiated submission narrowed the gap on the projects with the strongest evidence, and gave the developer a clear-eyed view of which megawatts to rely on for its own infrastructure planning. The exercise also reframed internal expectations: the team stopped treating a signed contract as equivalent to a counted megawatt.

Lessons for developers. A signed contract is necessary but not sufficient. Near-term, fast-energization claims draw the most skepticism, so be ready to defend lead times. The defaults are conservative by design; project-specific, well-documented inputs are the lever you control.

Case Study C — A transitional bridge using flexibility before supply arrives

Profile: A large-load developer needing to energize well before any matching generation could realistically be interconnected and accredited, in a zone where projected supply was tight.

Situation. A pure bring-your-own-generation approach was the right end state but could not meet the energization date — the supply would not be online and accredited in time. The developer needed a way to begin operating sooner without either taking on the not-preferred islanded design or absorbing full capacity costs during a period when capacity was scarce and expensive.

Approach. We structured a sequenced strategy. As a bridge, the load was set up to use a flexibility-based posture: in the resource-adequacy-constrained zone, the non-capacity backed load option allowed integration at reduced cost in exchange for accepting pre-emergency curtailment, while still being planned as Network Load. In parallel, the load enrolled its reliable on-site back-up generation in demand response, positioned to benefit from the proposed data-center accreditation class and expanded cost recovery so that participation trended toward revenue-neutral. The durable plan remained a transition to fully capacity-backed Network Load once committed generation came online.

Outcome. The load energized on the developer's timeline at a materially lower carrying cost during the bridge period, accepting a defined and pre-coordinated curtailment exposure rather than an uncontrolled one. Advance coordination of the back-up facilities minimized expected downtime, and the developer retained a clean migration path to the capacity-backed end state.

Lessons for developers. The paths are sequencing tools, not a one-time choice. Transitional options like non-capacity backed load are gated by a resource-adequacy test, so they are situational — available precisely when supply is short. Reliable, well-documented back-up generation is what makes the flexibility paths viable, and the value of that flexibility improves as the proposed accreditation and cost-recovery reforms take effect.

Part 3 — Technical FAQ

The questions below go a level deeper than a general overview, aimed at engineering and

development teams evaluating a PJM large-load project. Answers reflect PJM's 2025 workshop



materials, the CIFP Stage 4 package, and a representative utility load-adjustment submission.

This is general technical information, not legal or regulatory advice; confirm against current PJM

filings and FERC docket activity.