Powering AI When the Grid Can’t: Inside the New Behind-the-Meter Playbook

The AI infrastructure boom is forcing a hard reset in how the data center industry thinks about power. What was once a relatively straightforward utility procurement exercise is rapidly evolving into a complex, multi-disciplinary strategy problem spanning generation, fuel logistics, finance, and system architecture.

That reality framed a recent special edition of The Data Center Frontier Show Podcast, which recast and updated one of the most consequential sessions from the DCF Trends Summit 2025: From Grid to Onsite Powering: Optimizing Energy Behind the Meter for Data Centers.

Moderating the discussion was Fengrong Li, Senior Managing Director at FTI Consulting, whose questions and analytical framing shaped the conversation’s direction. With more than 20 years of experience across energy and infrastructure—including expert testimony before the Federal Energy Regulatory Commission (FERC), the Federal Communications Commission (FCC), and multiple state bodies—Li brought a systems-level perspective that pushed the panel well beyond a simple technology tour.

Her premise was clear from the outset: the rise of AI is not just increasing data center demand. It is restructuring the entire power delivery paradigm.

A Moderator Focused on the System-Level Shift

Li’s role went well beyond traditional moderation. Drawing on a career that includes 13 years at Siemens focused on grid issues and eight years at Mitsui in commodity trading and infrastructure investment, she constructed the discussion around what she described as “one of the most urgent topics shaping digital infrastructure deployment.”

“Onsite power and the rise of co-located, integrated power and AI campuses,” Li told the panel, “are accelerating data centers beyond traditional hubs and changing how they interact with the grid.”

Throughout the session, Li repeatedly pushed panelists to connect near-term deployment realities with longer-term structural implications particularly around redundancy, financing, and regulatory exposure. The result was a grounded look at an industry that is still early, still fragmented, but moving quickly toward new norms.

Interconnection Delays Meet AI Urgency

The first major theme Li surfaced was the widening gap between AI deployment timelines and utility interconnection schedules.

From the OEM vantage point, David Blank, Director of Distributed Sales North America at Siemens Energy, described the shift as both recent and dramatic.

“Behind-the-meter as a primary power application has really accelerated in the last year or so,” Blank said. “Everyone would prefer grid power. But in many cases, reliable access isn’t available for five, ten, even ten-plus years.”

That mismatch is increasingly forcing developers to treat onsite power not as contingency but as critical path infrastructure.

Brian Gitt, SVP and Head of Business Development at Oklo, framed the moment in historical terms.

“In some ways, this is back to the future,” Gitt said. “Large industrial loads historically built their own power plants. What’s new here is the scale and the speed the AI market is demanding.”

Multi-gigawatt campus roadmaps are now common in hyperscale AI planning. The panel broadly agreed that the grid alone cannot meet that growth on the timelines operators require.

WATCH THE VIDEO OF THE PODCAST

Bridge Strategy—or Structural Break?

One of Li’s most effective framing moves was to probe whether behind-the-meter power represents a temporary bridge or a lasting structural shift.

Here, the panel showed meaningful divergence.

Gitt suggested that while onsite generation is essential for time-to-power today, some deployments may eventually seek partial grid integration.

“Speed-to-power matters now,” he said. “But over time, there will be pressure for these assets to integrate responsibly with the grid.”

Yuval Bachar, CEO of ECL, offered a more definitive view.

“Many sites that go behind the meter will never go back,” Bachar said. “Data center operators have SLAs. Control matters. Once you own the system and control your destiny, the incentive to stay behind the meter is strong.”

Marty Trivette, SVP of Energy Solutions at AlphaStruxure, positioned the outcome as fundamentally economic.

“Historically, some sites built generation and later mothballed it when utility service became cheaper,” Trivette said. “I think we’ll see a mix. But once you have an effective onsite operating model, there’s less reason to move back.”

Li’s questioning consistently emphasized that the answer will likely vary by region, contract structure, and workload profile, rather than resolve into a single industry pattern.

AI’s Hidden Challenge: Load Volatility

Another key dimension Li surfaced, and one rising quickly on operator radar, is AI load behavior.

By asking how onsite generation changes “secondary and tertiary design elements,” she steered the panel toward what may be one of the least appreciated challenges in AI infrastructure: transient load swings.

Blank underscored the complexity.

“You’re effectively building a utility-scale power plant,” he said. “And then you have to make it behave like what the data center expects—especially when AI loads move quickly.”

The panel described a growing toolkit of fast-response technologies designed to manage this volatility.

“When you have sharp swings, you may need supercapacitors, flywheels, battery storage, and load banks to shape the curve,” Trivette explained. “Ultimately, the requirement is to deliver reliability that meets or exceeds grid expectations.”

Li repeatedly returned to this theme, highlighting how AI is forcing the convergence of traditionally separate engineering domains: power systems, mechanical infrastructure, and compute architecture.

Modularity Becomes the Common Language

Despite wide variation in technology pathways spanning advanced nuclear, hydrogen fuel cells, gas turbines and more, the panel aligned strongly around one architectural principle Li spotlighted early: modularity.

Oklo’s small modular reactor strategy is built around 75-megawatt increments designed to scale alongside campus growth.

“Data centers don’t turn on a gigawatt in one day,” Gitt noted. “They build in phases. Modular generation lets you synchronize power and compute growth.”

Bachar described a parallel philosophy at ECL using distributed block-level generation.

“AI is changing on a nine- to twelve-month cycle,” he said. “Power hardware doesn’t evolve that fast. With block-by-block buildouts, you maintain flexibility.”

Li consistently tied modularity back to finance and risk, underscoring that the appeal is not just technical elegance, but capital efficiency and schedule control.

Fuel Strategy Meets Real-World Constraints

Li also guided the discussion into the increasingly complex fuel question, particularly as operators push inference capacity closer to urban load centers.

Blank pointed to rising demand for multi-fuel capable turbines.

“Our turbines are relatively clean already, and they can become cleaner with alternative fuels,” he said.

But Bachar emphasized that logistics can become as constraining as generation technology.

“In cities, natural gas pressure and flow can be limited,” he said. “Hydrogen delivery has scaling limits. The future is going to be blended—we’ll use every energy source we can.”

The tone was notably pragmatic. Long-term decarbonization remains a goal, but near-term deployment is being driven by firm capacity availability.

Contracts as the New Gatekeeper

In the session’s final segment, Li turned to what may be the most underappreciated enabler of the behind-the-meter shift: contract structure.

Her closing questions focused on take-or-pay agreements, credit support, and the timing of hyperscaler commitments as areas increasingly determining whether projects advance.

Gitt described Oklo’s approach.

“We use capacity reservation deposits or prepayments to lock in capacity and support long-lead commitments,” he said.

Trivette noted that financing timelines are now tightly coupled to offtaker certainty.

“The sooner the ultimate offtaker is committed and backstopping the project, the faster it can move,” he said.

Blank confirmed the shift is visible even at the equipment level.

“We want to see clear offtake commitment before major equipment decisions move forward,” he said.

The Industry’s Power Conversation Has Entered Execution Mode

If the discussion revealed one overarching shift, it is this: the industry has moved beyond debating whether behind-the-meter power matters. The focus now is how to execute at scale.

Li’s framing throughout the session highlighted an ecosystem still early but rapidly professionalizing around modular design, financial rigor, and integrated energy planning.

As the AI buildout accelerates, the message from the panel was unmistakable. Power strategy is no longer a supporting function.

It is the gating factor.

And increasingly, the solutions are being built not at the substation but behind the meter.

At Data Center Frontier, we talk the industry talk and walk the industry walk. In that spirit, DCF Staff members may occasionally use AI tools to assist with content. Elements of this article were created with help from OpenAI's GPT5.

Keep pace with the fast-moving world of data centers and cloud computing by connecting with Data Center Frontier on LinkedIn, following us on X/Twitter and Facebook, as well as on BlueSky, and signing up for our weekly newsletters using the form below.