The Future of Power Grids and AI Data Center Growth

The intense race to supply electricity to artificial intelligence data centers is moving much faster than the development of the power grid, creating severe risks for developers, energy markets, and everyday consumers.

A new Horizons report from energy research firm Wood Mackenzie warns that data center operators face vastly underestimated technical, economic, and regulatory obstacles as they try to bypass traditional five- to 10-year grid expansion timelines.

To secure electricity quickly, developers are turning to collocated generation, which places data centers directly next to power plants, alongside flexible interconnection agreements with regional grid managers.

Wood Mackenzie experts state that while the power sector remains focused on data center flexibility, the true goal is securing reliable grid service backed by massive new transmission networks.

The rapid growth of energy demand creates an existential challenge for deregulated power markets, where load growth and consumer affordability are in direct opposition.

The Mid-Atlantic grid operator, PJM Interconnection, currently faces a massive supply gap with 78 gigawatts of committed data center demand versus only 36 gigawatts of accredited generation capacity in its development pipeline.

Texas faces a different economic hurdle, as current market power prices of $30 to $40 per megawatt-hour are not high enough to attract the necessary investments for new natural gas generation.

Meeting the projected demand under an accelerated growth scenario would require adding 16.4 gigawatts of gas capacity annually through 2035, a massive increase from the four gigawatts averaged per year between 2023 and 2025.

If electricity prices rise enough to incentivize this required power generation, it will inevitably raise utility bills for all consumers and trigger a political backlash.

To address this, PJM is splitting its market into two pricing tiers to charge large new loads more while paying less for existing resources, which could backfire by driving older gas and coal plants into early retirement and threatening overall reliability.

Texas has no comparable incentive plan for new grid-connected generation and is instead relying entirely on competitive market forces and rising prices to attract new investments.

As consumer affordability pressures mount, developers and investors should anticipate further regulatory interventions and market rule changes driven by political pressure rather than standard grid planning.

Although the industry has placed a massive bet on the “bring-your-own-generation” collocation model with over 90 gigawatts in US pipelines, researchers conclude this approach is unscalable and only achievable for the wealthiest tech giants.

The technical obstacles of on-site power generation are formidable because the near-instantaneous fluctuations in AI electricity demand can cause physical damage to gas turbines and reciprocating engines.

Using lithium-ion batteries as a shock absorber to protect generation equipment requires rapid response times from technologies that are not yet widely commercialized, risking a rapid depletion of the battery lifespan.

Additionally, the irregular power consumption of AI graphics processing units and cooling systems introduces power harmonics that cause equipment to overheat, alongside sub-synchronous oscillations that threaten the stability of local and distant generators.

Data centers also face major regulatory exposure from regional grid operators whose ultimate priority is system reliability over corporate speed-to-power demands.

Existing policies from PJM and the Southwest Power Pool give regional grids priority rights over collocated power plants, meaning data centers would be forced to cut electricity use during shortages even while their own on-site generators feed the public grid.

The Electric Reliability Council of Texas is working to update voltage and frequency rules to prevent data centers from prematurely switching to backup power during minor grid issues.

This rule update follows a 2024 incident in Virginia where 60 data centers dropped off the system simultaneously during a minor disturbance, nearly causing a total grid collapse.

Data centers currently in late-stage development may be unable to comply with these pending rules, potentially forcing expensive site redesigns and the purchase of new electrical equipment.

Compounding these issues is a fundamental debate over who will pay for nearly $100 billion in planned transmission investments meant to support data center growth.

These major regional investments include more than $30 billion by the Midcontinent Independent System Operator, $33 billion in Texas, $11.8 billion by PJM, and $8.6 billion by the Southwest Power Pool.

Without updating traditional cost-allocation rules, a massive portion of these infrastructure costs could be shifted onto ordinary ratepayers rather than the tech companies driving the demand.

Grid operators view flexible interconnections as temporary stopgaps, warning that costs for existing customers will rise if transmission systems are forced to provide complete service and data center demand fails to meet forecasts.