Who Should Power AI Data Centers—And How?

Most people see AI as software in the cloud. Grid planners see it as 30 gigawatts of new demand barreling toward them by 2030—the equivalent of powering about 20 million homes on top of everything else already running.

That demand is landing hardest on PJM Interconnection, the largest power market in the United States. PJM coordinates electricity for 65 million people across 13 states and D.C., including some of the most data-center-heavy corridors in the world. And right now, its members can’t agree on something basic: who should pay for the power that AI data centers chew through, and what kind of generation should supply it.

This fight isn’t just an industry squabble. It directly shapes electricity bills, the pace of clean energy buildout, and how “green” AI really is. If your company is betting on AI and green technology, this is your new risk surface.

This article breaks down what’s happening inside PJM, why the proposals collapsed, and—most useful for businesses and policymakers—what a smarter, cleaner pathway for AI data centers looks like.

---

The Core Conflict: AI Demand vs. a Stressed Grid

AI-driven data centers are on track to demand 30 GW of new capacity in PJM by 2030. That’s not a forecast rounding error; it’s a structural shock. Early stages of that growth have already helped push double-digit electricity price increases, including around 20% higher summer bills in New Jersey.

The core conflict is simple:

• Data centers want fast, reliable, cheap power so they can deploy AI at scale.
• Utilities and generators want stable revenue and limited risk when they build new plants.
• Households and small businesses don’t want to subsidize Big Tech’s power bill.
• Climate advocates want the new demand met with clean energy, not another wave of fossil infrastructure.

PJM’s “Critical Issue Fast Path” (CIFP) process was supposed to crank out rules quickly to handle this surge. Instead, after months of debate, none of the dozen proposals earned the two-thirds support needed to count as an endorsed path forward.

The result: the PJM Board now has a blank canvas—and an enormous responsibility—to decide how AI, data centers, and clean energy will coexist in the region.

---

What PJM’s Failed Vote Tells Us About the Future of Power

The failed advisory vote is revealing. It shows where the battle lines sit and what’s at stake for anyone building or hosting AI infrastructure.

The main options on the table

During the CIFP process, PJM stakeholders considered proposals that fell into a few camps:

1. “Bring your own power” for data centers
   Data centers would be pushed—or required—to build or procure their own generation, potentially on-site or via dedicated projects, in exchange for faster interconnection and siting.

2. Fast-tracking energy projects
   Create a special lane in the interconnection queue for generation meant to serve data centers, so those projects don’t sit in limbo for years while AI demand keeps growing.

3. Pause new data center connections
   A stricter idea: don’t connect new data centers unless and until PJM can guarantee reliable power for them, based on clear reliability standards.

4. Conditional service and curtailment rules
   During grid stress—heatwaves, cold snaps, hurricanes—data centers would be required to cut load unless they bring their own additional resources (e.g., on-site generation or storage) into the system.

Notably, none of these reached two-thirds support. Why? Because each shifts costs and risks in ways different groups hate:

• Tougher rules mean generators and data centers carry more risk instead of pushing it onto ratepayers.
• Lax rules mean residential customers subsidize corporate loads and more fossil generation gets locked in.
• Some options constrain gas buildout, others constrain clean energy timelines, and some threaten AI companies’ deployment schedules.

The reality? There’s no neutral, painless choice anymore. PJM’s decision will pick winners and losers.

---

Why This Matters for Green Technology, Not Just Electricity Bills

AI is often sold as an enabler of green technology—optimizing grids, managing renewables, improving efficiency. All of that is true. But AI data centers themselves are now one of the fastest-growing energy loads on the system.

If that load is met with:

• New gas plants and diesel backup generation, AI becomes a climate liability—high emissions, local air pollution, and decades of fossil lock-in.
• Wind, solar, storage, and flexible demand, AI can be part of the clean-energy transition—especially if those data centers actively help balance the grid.

Right now, PJM is at a fork in the road:

• Energy experts warn that most near-term “firm” capacity will likely come from natural gas unless policy and market rules push harder on clean options.
• Climate advocates are pointing to Texas of all places as proof that you can build renewables fast when the rules don’t strangle them. Texas added tens of gigawatts of wind and solar in under a decade—largely because the interconnection and market structure didn’t bog projects down for years.

If PJM handles this poorly, we get more gas peakers and more diesel fumes in neighboring communities every time the grid strains. If PJM and the states get it right, AI could anchor massive new clean energy investment.

---

The Diesel Backup Problem: Hidden Emissions in “Cloud” AI

One of the thorniest proposals floating around PJM is this: during grid emergencies, data centers should be the first to curtail, unless they contribute extra resources to help keep the grid reliable.

On paper, that makes sense. In practice, it creates a dirty risk: diesel backup generators.

Here’s the pattern we’ve seen in data center hubs:

• When grid power is cut or curtailed, data centers fire up rows of diesel gensets.
• Those units are often clustered near residential areas or commercial parks.
• They emit NOx, particulates, and CO₂—undercutting both local air quality and climate goals.

So a naive “cut their grid power and let them self-supply” policy just shifts pollution from regional gas plants to hyper-local diesel stacks. That’s not green technology; it’s a shell game.

Cleaner alternatives to diesel for resilience

If you’re designing or hosting data centers and want resilience and credibility on sustainability, diesel can’t be your long-term backbone. Better options include:

• Battery energy storage systems (BESS) for short-duration backup and grid services.
• On-site solar plus storage for partial self-supply and peak reduction.
• Green hydrogen or renewable gas for longer-duration backup, where available and credible.
• Demand flexibility, where non-critical AI workloads can be shifted in time or location to reduce peak stress.

The companies that win the reputation game will be the ones that treat backup power as a clean-tech design challenge, not just a compliance checkbox.

---

A Smarter Playbook: How Data Centers Can Go Truly Green in PJM

You don’t have to wait for the PJM Board or FERC to sort this out. If you’re planning large-scale AI or data center investments, there’s a better way to approach power.

1. Treat power strategy as core product strategy

For AI-heavy businesses, grid-aware design is now a competitive advantage, not an afterthought.

Practical moves:

• Put energy and sustainability teams at the table with site selection, finance, and product leads.
• Model total cost of power over 20–30 years, not just today’s tariff—especially if gas-heavy buildouts risk future carbon costs or stranded assets.
• Build scenarios where your load can ramp up or down based on grid conditions, with clear SLAs on which workloads are flexible.

2. “Bring your own clean power” the right way

The governors’ coalition proposal—encouraging data centers to supply their own power in exchange for priority permitting—points in the right direction. The trick is how you execute it.

A robust strategy blends:

• Long-term contracts (PPAs or similar) with new wind, solar, and storage inside PJM, so your loads actually help grow clean capacity in the regions you stress.
• On-site or near-site generation where feasible, especially solar plus storage for daytime loads and peak shaving.
• Participation in capacity and ancillary markets, so your assets (or your flexibility) are seen as grid resources, not just load.

This is where AI can be at the heart of green technology: using intelligent forecasting, optimization, and control to align compute demand with real-time clean energy availability.

3. Build flexibility into AI workloads

Not every AI task is latency-critical. Training runs, batch inference, and internal analytics can often shift in time or location.

Concrete options:

• Geographic load shifting: route lower-priority jobs to regions and time windows with higher renewable output and cheaper, cleaner power.
• Time-of-day scheduling: align flexible tasks with solar peaks or off-peak hours.
• Grid-signal integration: respond automatically to price, carbon-intensity, or reliability signals with pre-agreed throttling or rescheduling.

Companies that can say, “Our data centers reduce demand when the grid is stressed and ramp up when renewables are plentiful” will stand out to regulators, communities, and climate-conscious customers.

4. Design for community and political resilience, not just electrical resilience

The PJM fight has already become a political issue in places like New Jersey, where voters of both parties say data centers should pay more of the grid costs. That’s a flashing red light for any operator trying to quietly expand.

If you want smoother permitting and less backlash:

• Be transparent about energy use, emissions, and local air quality impacts.
• Invest in local benefits: community solar, workforce programs, or grid upgrades that visibly help residents.
• Proactively commit to clean backup solutions instead of diesel, even before rules force your hand.

When public sentiment shifts, regulators follow. The companies that anticipated that shift will be the ones still growing comfortably in 5–10 years.

---

How PJM’s Decision Could Redefine “Green AI” in 2026 and Beyond

By the end of this year, the PJM Board is expected to submit its chosen rule package to federal regulators. FERC will then judge whether the proposal is “just and reasonable” and not discriminatory. That process will shape:

• How fast new AI data centers can connect.
• How much of their cost is shifted to everyday ratepayers.
• Whether gas dominates the buildout—or clean energy gets a real shot.

For our Green Technology series, this is the crux: AI and data centers aren’t automatically sustainable just because they enable digital efficiency. Their climate impact depends almost entirely on how we power them.

There’s a better path than endless gas plants and diesel backup:

• Use AI to actively support the clean energy transition—forecast, optimize, and flex demand.
• Commit to “bring your own power”—but make that power clean and integrated with the broader grid.
• Treat community trust and political risk as seriously as latency and uptime.

The companies that move first on this will own the narrative of “green AI infrastructure” instead of scrambling to defend it later.

If you’re planning large-scale AI or data center investments in the next 2–3 years, this is the moment to stress-test your power strategy. Are you designing for a fossil-heavy status quo that’s already under attack—or for a grid where AI and clean energy actually pull in the same direction?