colocationAIenergy

Colocation Options for AI Workloads in the PJM Region: Power, Renewables and On-Site Generation Comparisons

ddatacentres

2026-02-01

10 min read

Compare PJM colocation providers for AI by guaranteed power capacity, on-site generation, PPA options and resilience under new 2026 cost rules.

Hook: Power risk is now procurement risk — and AI workloads make it urgent

If you're running large-scale training clusters or latency-sensitive inference in the PJM footprint, your colo choice is no longer just about racks and cross-connects. Power capacity guarantees, on-site generation options, and renewable procurement mechanisms now determine whether your project delivers on SLA commitments and stays within budget under new cost-allocation pressures introduced in late 2025–early 2026.

In January 2026 the federal government signaled a shift: data-center operators could be required to fund transmission and generation upgrades tied to rapid AI-driven demand growth. That changes commercial negotiations and elevates energy architecture to a primary vendor selection criterion. This article gives technology and procurement teams a practical framework to compare colocation providers in the PJM region — emphasizing the metrics that matter to AI tenants.

The 2026 context: PJM, AI growth and new cost-allocation dynamics

PJM is the largest U.S. RTO by load and covers critical metro hubs for AI deployments (Northern Virginia, parts of Pennsylvania, New Jersey, Maryland). AI compute demand — especially dense GPU training clusters — places sustained, high-PUE loads on local distribution and transmission assets. Late 2025 and early 2026 saw accelerated announcements for large campuses and wholesale builds inside PJM.

"A federal plan unveiled in January 2026 proposes that data centers pay for incremental generation and transmission to handle AI-driven load growth — shifting a portion of grid upgrade costs to large electricity consumers."

The practical effect for colo tenants: carriers and operators will increasingly assign or pass through allocation and interconnection costs. Some operators will absorb upgrades and monetize the service later; others will require tenants to underwrite capacity. Either way, power procurement and on-site resilience are now core commercial levers.

Why AI workloads change the colocation evaluation criteria

AI workloads differ from typical enterprise or web hosting in four ways that affect colo selection:

Sustained high-density power draw: racks frequently draw multiple kilowatts each — not intermittent bursts.
Predictability demands: training jobs run for days; any unplanned derate or outage is costly.
Energy cost sensitivity: training is energy-dense and highly exposed to variable energy charges and capacity allocation rules.
Sustainability procurement: many enterprise AI buyers require 24/7 carbon accounting and hourly-matching renewable attributes.

Key metrics AI tenants must compare

When you issue an RFP or shortlist providers, evaluate across these dimensions and quantify every answer.

Guaranteed power capacity: Specify kW/rack, kW/cage, and the provider's minimum contiguous MW reservation. Ask: is capacity firm or subject to utility upgrade timelines? What is the lead time for additional MW?
Power SLA and remediation: Uptime guarantees, credit formulas, and the measurement points for SLA are essential. Also ask how power derates are handled and what constitutes an SLA event.
On-site generation & fuel strategy: Generator type (diesel gensets, gas turbines, reciprocating engines), total on-site MW, run-hours guaranteed, blackstart capability, and firm fuel supply contracts (on-site tanks, road access for refueling, dual-fuel capability).
Battery energy storage systems (BESS): BESS capacity and discharge duration for critical bridging versus sustained resiliency, and whether they support fast frequency response or 24/7 load shifting.
PPA & renewable options: Are tenant-specific PPAs supported (virtual or sleeved), is hourly 24/7 matching offered, and can the provider bundle storage + renewable projects to deliver carbon-free energy at machine-hour granularity?
Cost allocation and interconnection liabilities: Who pays for grid upgrades and interconnection studies? Is there a formula for split costs, or is the tenant on the hook for direct assignment?
Operational resilience: On-site microgrids, N+1 design for generators and UPS, fuel contracts, and maintenance SLAs.
Transparency & metering: Real-time energy metering down to rack-level, interval data access (API), and granular chargebacks for energy and demand.

How the major colo players typically position in PJM (2026 lens)

Rather than a fully numerical ranking, here are practical, evidence-based profiles of the typical approaches you’ll encounter in PJM from major providers. Use these as negotiation starting points.

Large campus operators (Digital Realty, Vantage, QTS)

These providers focus on multi-megawatt campuses and build-to-suit shells designed for AI scale-outs. Their value propositions include:

High guaranteed capacity: ability to commit contiguous MWs and to build substations.
On-site generation: large genset farms and increasing BESS installations for bridging and grid services.
PPA flexibility: they commonly support large-scope PPAs and can sponsor utility-scale renewables (or offer tenant-tailored PPAs), though pricing and terms vary.
Cost allocation posture: these operators often negotiate to allocate interconnection upgrade costs across campus tenants — read the fine print.

Interconnection-first metro colos (Equinix, CoreSite/American Tower)

Metro-focused colos trade on network density, multi-cloud access and short-haul connectivity. For AI buyers:

Power density: typically good for smaller footprint AI clusters, less suited to wholesale multi-MW campuses unless they coordinate campus scale across sites.
Renewables: many have advanced 24/7 matching programs and VPPAs tied to large RE projects to meet corporate sustainability targets.
Resilience: best-in-class SLA and distributed interconnectivity reduce single-site risk, but on-site generation may be smaller.

Developer and flexible providers (CyrusOne, Iron Mountain)

These providers often offer tailored commercial structures: build-to-suit, flexible PPA passthroughs, and varying on-site generation strategies.

Commercial flexibility: more willing to support tenant PPAs, behind-the-meter generation and customized resiliency contracts.
Operational caution: ask for explicit commitments on run-hours and generator testing cadence.

Concrete resilience strategies under new cost allocation rules

As load-growth costs become more visible and potentially allocated to large consumers, providers will respond in three ways you should account for.

Shift to tenant-funded upgrades: Operators will offer firm capacity but may require tenants to fund the immediate interconnection or substation build — typically amortized over contract term.
Bundled generation and PPAs: Providers will create bundled offers that combine firm on-site generation, storage, and renewable PPAs — shifting capital and operational risk away from the tenant.
Market mechanisms and demand response: Use of BESS for demand charge management and participation in ancillary markets to offset costs is increasing; providers may share revenues with tenants.

Negotiation levers for tenants:

Insist on a clear cost-allocation clause that caps tenant liability for transmission upgrades or ties payments to concrete milestones.
Ask for performance-based capacity commitments (e.g., a guaranteed MW backed by liquid credits if not delivered).
Demand transparent meter data and a pass-through accounting model so you can integrate energy costs into your model.

RFP checklist — specific questions to demand answers for AI workloads

Use this checklist verbatim in procurement and technical RFPs:

What is the maximum contiguous guaranteed capacity (in MW) you can reserve and within what lead time?
Define the power SLA: measurement point(s), uptime percentage, credit formula, and exclusions.
Provide details on on-site generation: type, total MW, blackstart capability, guaranteed run-hours without external refuel, fuel storage capacity (gallons), and dual-fuel options.
What BESS is installed? Specify kWh and discharge duration, and whether BESS supports islanded operation.
Describe your PPA offerings: tenant-tailored VPPA, sleeved PPA, hourly matching, bundled storage+PPA.
Who pays for interconnection and distribution upgrades? Provide typical cost allocation scenarios and escrow options.
Do you permit tenant-sourced on-site generation or behind-the-meter projects? If so, what interface and metering are required?
Provide examples of past incidents and remediation, including time to recover and SLA credit payouts.

Modeling total cost of ownership (TCO) for AI tenants — a practical approach

AI workloads shift energy from an operational line item to a central driver of TCO. Build a model with these components and run sensitivity analyses on two variables: energy price and allocated upgrade costs.

Core model components:

Base rent and power-density premium (per kW)
Energy charge (kWh) and demand charge (kW)
Capacity reservation or capacity charge for committed MW
PPA pass-throughs, renewable attribute costs (REC or 24/7 premium)
Resiliency premium: cost of guaranteed on-site generation access or backup SLA adders
Allocated interconnection/upgrade amortized cost (if tenant-funded)
Operational revenue offsets (demand response payments, ancillary market participation)

Example sensitivity to run:

Base case: 2 MW sustained load, energy $0.06/kWh, demand $15/kW-month, no upgrade allocation.
Stress case: energy +40%, plus a one-time $5M interconnection charge amortized across 5 years and two tenants.

Run both scenarios to compute per-training-job energy costs and compare with alternative strategies: on-premises build, cloud bursting for training, or a hybrid mix of colo for inference with cloud for training.

Practical deployment patterns to reduce risk and cost

These are proven strategies used by experienced AI tenants to mitigate power and allocation risks:

Multi-site distribution: Spread critical workloads across two metro colos with diverse utility feeds and independent on-site generation to avoid single-point transmission risk.
Hybrid resilience: Keep short-term cloud bursting capacity for critical retraining windows, rather than committing all capacity to an untested single site.
Firming with storage: Use BESS to smooth demand spikes, reduce peak demand charges and participate in ancillary markets to offset costs.
Tenant PPAs and contractual green tags: Structure PPAs with hourly matching to meet compliance and customer reporting requirements, and include terms that indemnify tenants from renewable project non-performance.
Negotiate fuel and generator SLAs: Require minimum onsite fuel hours (e.g., 72-120 hours) and guaranteed refueling response windows built into the contract.

2026–2028 trends and what to expect next

Expect these trends to accelerate through 2026 and into 2028:

More bundled offers: colos will increasingly bundle BESS, on-site generation and renewable PPAs to offer “firm carbon-free” compute packages.
Regulatory clarity: PJM and FERC will publish clearer cost allocation frameworks; those will affect contractual risk language and may standardize tenant liability caps.
New fuel vectors: pilot hydrogen and RNG projects for long-duration resiliency will grow, but diesel remains the fallback for most campuses — watch evolving standards like EV and charging standards as infrastructure adapts.
Market structures: participation in capacity and ancillary markets will be common — providers will offer revenue-sharing to defray tenant costs.

Actionable takeaways — 10 immediate steps for AI procurement teams

Start RFPs with precise power profile requirements (sustained kW/rack and peak behavior).
Demand an explicit breakdown of interconnection liabilities and cap tenant exposure.
Require hourly energy and carbon attribute data via API for 24/7 matching.
Insist on firm on-site generation run-hours and a documented fuel logistics plan.
Negotiate BESS sizing and service rights for demand-charge management.
Use multi-site deployments for mission-critical training pipelines.
Evaluate bundled PPA+storage offers against tenant-sourced VPPA economics.
Include SLA credit formulas that connect directly to your business loss calculations.
Model TCO with interconnection uplift as a sensitivity case.
Work with legal to include carve-outs for regulatory changes tied to federal or regional cost allocation rules; consider hybrid oracle approaches for regulated data and compliance integrations.

Final assessment: what a winning colo partner looks like in PJM for AI

A partner that meets AI needs in PJM in 2026 will combine three capabilities:

Firm, scalable power — contiguous MW commitments with operational guarantees and short lead times for scaling.
Resilient on-site resources — generators with guaranteed run-hours, meaningful BESS, and tested blackstart procedures.
Transparent commercial mechanics — clear PPA offerings, transparent cost-allocation treatment, fine-grained metering, and contractual protections against unforeseen grid-assigned costs.

Providers differ in how they balance these elements. Campus operators deliver scale and on-site generation; interconnection-centric colos deliver network reach and often best-in-class renewable programs; developers provide contractual flexibility. Your selection should be driven by a technical fit for sustained density, clear cost-allocation terms, and the provider’s willingness to structure energy contracts aligned with long-term AI economics.

Call to action

Start your RFP today with our downloadable AI-colo RFP checklist tailored to PJM (includes question templates for power SLA, on-site generation, PPA structuring and interconnection liability language). Contact our editorial team for a vendor briefing or to commission a site-level assessment for your planned deployment.

Make power your procurement priority — because in 2026 it’s the single biggest determinant of AI performance, cost and compliance.

datacentres

Contributor

Senior editor and content strategist. Writing about technology, design, and the future of digital media. Follow along for deep dives into the industry's moving parts.