To address time to power, one of the biggest constraints currently slowing data center deployment, EPRI is launching Flex MOSAIC, a uniform flexibility classification framework for large electric loads, developed through its DCFlex initiative in collaboration with more than 65 utilities, system operators, regulators, hyperscalers and technology providers.

The voluntary framework is meant to establish a “shared, credible way” to define flexibility from large loads (particularly data centers) based on the magnitude, timing, duration and frequency of their response. By enabling a common understanding of what flexibility a load can deliver, EPRI argues the framework could help shorten interconnection timelines, improve grid planning confidence and accelerate access to power without compromising reliability or affordability.

“As demand from AI and data centers grows at unprecedented speed, flexibility is becoming the third leg of the speed-to-power stool, alongside generation and transmission,” said EPRI President and CEO Arshad Mansoor. “This framework allows everyone — utilities, regulators, and large‑load developers — to have common language about flexibility and to trust what that language means. That shared understanding is essential to moving faster while maintaining reliability.”

Data center construction and the advent of artificial intelligence (AI) are driving unprecedented electric load growth across the United States. Massive hyperscalers with deep pockets and bold aspirations need power, and they need it fast.

From May 12-14, 2026, DTECH Data Centers & AI will assemble utilities, engineers, and technical decision-makers from across this emerging ecosphere in Scottsdale, Arizona, to discuss everything from capacity constraints to streamlining studies, from modernizing infrastructure to integrating onsite generation into both utility and customer-side systems.

Register for DTECH Data Centers & AI now before early-bird pricing ends on April 1, 2026.

The framework defines flexibility through practical performance characteristics, including how quickly a load can respond, how long adjustments can last and how much power can be reduced or shifted. These characteristics are organized into a set of uniform flexibility classes that utilities, system operators and data centers can apply consistently across regions.

The framework is meant to provide a technical foundation that jurisdictions and market participants can adapt to their local needs. “As large, flexible loads play a growing role in the power system, having clear, technically grounded definitions of flexibility is critical for reliability,” said North American Electric Reliability Corporation President Jim Robb. “A common framework like this can help system operators and planners speak the same language, essential for maintaining a reliable grid.”

“As demand from data centers accelerates, state regulators are focused on ensuring customers are not burdened by the costs of serving new, large loads, as well as maintaining grid reliability,” said NARUC President Ann Rendahl. “NARUC looks forward to engaging with EPRI and others on how a voluntary, standardized framework like Flex MOSAIC can create a common language and shared understanding of flexibility, and provide benefits to state regulators when evaluating data center integration, without shifting costs to customers or compromising grid reliability.”

Initial framework participants include Alliant Energy, Arizona Public Service, California ISO, El Centro Nacional de Control de Energía (CENACE), Compass Datacenters, Constellation Energy, DTE Energy, Entergy, Exelon, Georgia Transmission Corporation, Google, Honeywell, Independent Electricity System Operator (IESO), ING, Jenbacher, Korea Power Exchange (KPX), KPMG, LG Pado, Lincoln Electric System, Lower Colorado River Authority, Meta, Midcontinent Independent System Operator (MISO), Nebraska Public Power District, NERC, NVIDIA, Portland General Electric, PSEG, Rayburn Electric, Salt River Project, Siemens, Southern Company, Southwest Power Pool and United Power.

Caterpillar moved to the center of the AI infrastructure buildout this week as developer Nscale said it would use the company’s natural gas generator sets to power a major new West Virginia data center campus tied to Microsoft and NVIDIA.

Monday’s announcement positions Caterpillar’s G3500 series reciprocating engine platform as core infrastructure for what Nscale said could become one of the country’s largest dedicated AI compute developments.

Under the plan, Caterpillar equipment would support 2 GW of onsite generation by the first half of 2028 at the Monarch Compute Campus in Mason County, West Virginia, giving the project a faster path to power as grid access and transmission upgrades remain a constraint for large data center loads.

Nscale said the campus would host up to 1.35 GW of AI compute capacity for Microsoft under a letter of intent tied to NVIDIA Vera Rubin NVL72 systems and the NVIDIA DSX AI Factory reference design. The company also announced it had acquired American Intelligence & Power Corp., which includes the 2,250-acre Monarch site and what Nscale described as the first state-certified AI microgrid in the U.S., with expansion potential beyond 8 GW.

---

Trump urges tech companies to build their own power plants for data centers

---

It’s the latest example of a data center project being structured around large-scale onsite natural gas generation, rather than waiting solely on utility service.

In a recent report, Cleanview identified 46 U.S. data center projects representing a combined 56 GW of planned behind-the-meter power capacity, which it estimated at roughly 30% of planned U.S. data center capacity in its tracker.

The research company also said 90% of the projects it identified were announced in 2025, indicating that “Bring Your Own Power” has shifted from a niche workaround to a more mainstream development path as grid interconnection timelines lengthen.

“A year ago, behind-the-meter data center power was a curiosity, embodied by xAI’s controversial decision to truck mobile generators into Memphis,” said Cleanview. “Now it’s an increasingly common development strategy.”

Cleanview added many of the projects it identified have secured equipment partners and are already under construction.

“Projects like Monarch demonstrate how Caterpillar’s natural gas generation platforms are being deployed as core infrastructure for data centers and other power intensive applications where reliability, speed of deployment, and lifecycle performance are critical,” Melissa Busen, Caterpillar senior vice president of Electric Power, said in a statement.

Nscale said the West Virginia site would operate independently of the local grid, which it argued would avoid adding costs to existing utility customers, while preserving the option of a future grid interconnection that could allow exports. The company also said it is pursuing carbon sequestration and a design intended to reduce water use.

---