AI data centers push grid overhaul

Despite a modest uptick in 2024 with about 880 miles of new high-voltage lines completed—mostly long-delayed projects initiated 10-15 years ago—the U.S. transmission grid buildout remains sluggish amid surging AI-driven electricity demand. Federal initiatives like FERC Order 1920 aim to enhance long-term transmission planning, but their future is uncertain given political shifts, while longstanding regulatory and political hurdles continue to hamper timely infrastructure progress, as noted during the Biden administration's efforts.

Texas exemplifies the transmission capacity crisis, where the ERCOT interconnection queue exploded from 42 GW in early 2024 to 226 GW by late 2025—nearly three times the state’s peak load of 85 GW—highlighting severe bottlenecks driven by speculative and rapid AI data center growth. This unprecedented surge, with tech capital expenditure nearing 2% of U.S. GDP, is reshaping grid planning and reliability, forcing utilities to confront demands that outpace infrastructure development timelines by years.

By early 2026, grid operators like PG&E faced a deluge of interconnection requests that overwhelmed systems designed for minimal demand growth, with annual load increases hitting 4%—a stark contrast to historical sub-1% rates—and projections estimating AI data centers could consume up to 9% of U.S. electricity by 2030. This demand is heavily concentrated in regions such as Virginia, Texas, and California, exacerbating existing strain and revealing a massive backlog of over 2,600 GW of generation and storage projects awaiting grid connection, more than twice the nation’s installed capacity.

Utilities are grappling with a widening gap between announced AI data center load requests and deliverable demand, as speculative filings inflate interconnection queues—700 GW reported in 2025 alone, matching U.S. peak demand—and drive up costs passed to consumers. In response, companies like Duke Energy and AEP are restructuring capital plans, raising billions to finance grid expansion, while adopting new planning frameworks that prioritize commercially committed projects over speculative ones, reflecting a fundamental shift in managing grid reliability amid unprecedented and uncertain AI-driven load growth.

Texas has pioneered a comprehensive overhaul of its grid interconnection process to manage the unprecedented surge in large load requests, particularly from AI data centers and crypto miners. The Public Utility Commission of Texas (PUCT) introduced the 'Batch Zero' process, rooted in the 2025 Senate Bill 6 (SB6), which mandates a $50,000 per megawatt financial deposit and milestone-based commitments to screen out speculative projects and ensure only credible loads proceed. This multi-year, batch-based approach, expected to span three to five years, groups over 438 GW of proposed demand into study batches, shifting much of the analytical responsibility to ERCOT and incorporating innovative programs like WL-PUN and PCLR to integrate more load without immediate transmission expansion.

At the federal level, FERC has taken decisive action to accelerate grid access for large electricity users, including AI data centers, by issuing show cause orders in June 2026 to six major Regional Transmission Organizations (RTOs) and Independent System Operators (ISOs). These directives require tariff reforms within tight deadlines to enhance interconnection efficiency, clarify cost allocation, and introduce operational flexibilities such as ramp-up service and conditional interconnection. While industry groups like the Data Center Coalition and Nvidia applaud these reforms for promoting growth and affordability, experts like Ari Peskoe caution that transparency challenges remain, particularly regarding who ultimately bears the costs of network upgrades.

State regulators and utilities nationwide are responding to the speculative nature of AI-driven load growth by tightening financial requirements and revising tariffs to focus on deliverable demand rather than announced projects. For example, American Electric Power’s Data Center Tariff process in Ohio reduced preliminary 30 GW data center load requests to 5.6 GW backed by firm commitments, reflecting a broader shift toward portfolio-based planning that incorporates financial gates, tiered project credibility, and enhanced procurement strategies. Utilities like Duke Energy and AEP are also expanding capital plans and creating procurement subsidiaries to secure critical equipment amid equipment shortages, underscoring the systemic nature of these challenges across major AI data center corridors including Texas, Virginia, and the Carolinas.

Texas’s regulatory focus has evolved from broad post-Uri reforms to targeted strategies addressing rapid large load growth from hyperscalers, emphasizing equitable cost allocation and grid reliability. Chairman Thomas Gleeson highlighted the urgency of meeting hyperscalers’ demand for power within eighteen months and the imperative to shield residential and small business ratepayers from disproportionate transmission upgrade costs. To this end, Texas is considering reforms to cost allocation methods—such as shifting from a four coincident peak (4CP) to a twelve coincident peak (12CP) methodology with minimum demand charges—and expanding demand-side management programs including aggregated distributed energy resources and a $1.8 billion backup power initiative funded by the Texas Energy Fund.

By early 2026, PJM faced a resource adequacy crisis driven largely by surging data center loads, prompting FERC to implement a tariff cleanup that clarified co-location rules and standardized service options. This regulatory adjustment aimed to better accommodate large loads without directly solving generation shortages or accelerating transmission buildout, underscoring the complexity of balancing market clarity with infrastructure constraints. Meanwhile, regional governors, including Josh Shapiro and Glenn Youngkin, pressed the White House for emergency interventions such as auctions and new power plant construction, signaling heightened political urgency around PJM’s reliability challenges.

PJM’s capacity auctions revealed historic shortfalls, with a 6.6 GW deficit against Installed Reserve Margin for 2027/28 and record-clearing prices soaring to $333.44 per MW-day—temporarily suppressed by a FERC-approved price collar that, without it, might have reached $530 per MW-day. This price volatility and capacity scarcity spurred mid-Atlantic governors to convene at the White House in January 2026, advocating for federal intervention that included extending price collars and instituting a 15-year procurement framework to incentivize new generation. These moves reflect a strategic shift toward balancing cost fairness with reliability amid unprecedented AI-driven electricity demand growth.

Recognizing that hyperscale data centers are the primary drivers of capacity demand growth, PJM and DOE considered emergency orders to make these large loads financially responsible for new supply and grid integration costs. This approach encourages a more robust bilateral market where hyperscalers proactively procure capacity tailored to minimize grid impacts, potentially reducing reliance on traditional auctions. Adjustments to procurement targets—from 6 GW down to 3 GW post-load updates—and the emphasis on demand response and battery storage as viable short-term solutions highlight evolving strategies to manage capacity shortages pragmatically.

PJM’s accelerated stakeholder process, launched in August 2025, sought consensus on reforms to address overwhelming data center load growth but failed to reach agreement on 12 proposals by November. Governors from PJM states (except D.C.) endorsed principles focusing on improved load forecasting, cost allocation to data centers, expedited interconnection studies, and protecting residential customers from capacity price hikes. The unprecedented involvement of the White House in this regional market reform process underscores the federal government’s high stakes in managing AI-driven demand surges. Despite strong stakeholder engagement, ultimate decision-making authority remains with the PJM board and FERC, reflecting the complex governance layers shaping market evolution.

By early 2026, grid operators like ERCOT began overhauling their interconnection processes to manage an unprecedented 233GW queue driven by AI data center demand, adopting batch processing and two-stage load study methods to improve cost and schedule transparency while screening speculative projects. Complementing these operational reforms, FERC mandated fast-tracking of data center interconnections and encouraged the adoption of advanced technologies such as solid-state transformers and dynamic line rating tools to optimize existing grid capacity and accelerate integration of large AI loads.

Battery storage has emerged as a critical bridge technology amid slow transmission upgrades, with utilities and companies like CBAK Energy developing advanced lithium iron phosphate cells tailored for AI data center backup and uninterruptible power supply needs. These storage solutions, alongside digital grid management tools expanded by firms like Hitachi Energy, are enhancing grid responsiveness and enabling more flexible, reliable integration of volatile AI-driven demand and renewables.

Significant investments in large-scale transmission infrastructure are underway to support massive AI loads, exemplified by FirstEnergy’s joint venture with American Electric Power to build over 300 miles of 765-kV lines in Ohio, reflecting a broader industry shift to treat hyperscale AI campuses as integral computational load entities. This paradigm shift is prompting new reliability standards developed in collaboration with NERC and ReliabilityFirst, recognizing AI data centers as essential components of the bulk power system rather than mere customers.

Federal and private sector initiatives are accelerating grid modernization through coordinated funding and strategic partnerships, such as Texas’s Senate Bill 6 and the $10 billion Texas Energy Fund reallocating transmission costs to large loads, alongside National Grid’s $1.75 billion investment in Joulent’s technology-driven power solutions. Meanwhile, the U.S. Department of Energy’s Grid Modernization Initiative and companies like GE Vernova and Siemens Energy (rebranding as Omterra) are advancing AI-driven grid management, cybersecurity, and integrated renewable strategies to meet the dynamic demands of a rapidly evolving electricity landscape.

The persistent challenge of interconnection backlogs, which have long hindered the timely addition of new power projects amid surging electricity demand, is being directly addressed by Senator Martin Heinrich’s Grid Connection and Congestion Management Act (S. 5008), introduced in July 2024. This legislation mandates Regional Transmission Organizations (RTOs) and Independent System Operators (ISOs) to implement a standardized 'connect and manage' interconnection option—Basic Access Service for Energy-Only Delivery (BASED)—within 12 to 18 months, drawing on the proven success of Texas’s ERCOT system that has interconnected over 100 GW since 2020. By expediting interconnections while managing grid reliability through agreed curtailments, the bill aims to reduce costly delays and accelerate the integration of affordable, reliable power to the U.S. grid, garnering support from industry leaders like SEIA, RMI, and GridLab.

The BASED interconnection model innovatively streamlines the review process by allowing generation projects to connect under operating limits that avoid triggering costly grid upgrades, thereby capping interconnection reviews at one year. This approach requires RTOs and ISOs to focus evaluations solely on necessary studies to identify maximum injection levels without additional transmission facilities, with generators agreeing to curtail output as needed to maintain grid stability. While this method promises faster, simpler interconnections and aligns with FERC’s pro forma tariff revisions to be completed within 18 months, it also raises concerns in capacity market regions where BASED resources might not qualify for capacity revenues or resource adequacy, highlighting the nuanced trade-offs in balancing speed, cost, and market participation.

Despite the promise of the Grid Connection and Congestion Management Act to alleviate interconnection bottlenecks, political resistance and the complexity of coordinating federal leadership remain significant hurdles. The bill’s reliance on a model adapted from ERCOT’s unique market structure underscores the challenge of replicating such success across diverse RTO and ISO regions with varying regulatory and market frameworks. Stakeholders acknowledge that while the 'connect and manage' approach can bring new, cheap energy online faster, its implementation must carefully navigate regional differences and potential impacts on capacity markets to avoid unintended consequences that could shift cost burdens onto consumers.