The current AI boom has quickly created unprecedented demand for data centers, and with it, exponentially higher demands for power.
The estimates and trends would have seemed impossible just a short time ago. Recent research projects that power requirements from AI data centers could grow more than 30 times by 2035, with some facilities scaling from traditional 10-50MW power requirements to 100-500MW or more.
Additionally, NVIDIA and other GPU chipmakers have released product roadmaps calling for 1MW per server rack, raising new questions related to power and cooling:
- Where will all this power come from?
- Can the current grid handle such explosive growth without triggering brownouts or forcing a return to fossil fuels?
- Is this growth sustainable in the long term?
While many of today’s headlines seem to paint a gloomy picture of these new power demands—and data centers’ role—a closer look shows that there’s no reason to panic.
Innovative Power Strategies Taking Hold
The industry isn’t waiting for these problems to solve themselves. Data center operators are already rolling out proven technologies that tackle power constraints head-on. These new innovations are proving to be effective responses to growing power demands and are helping prevent what could become an infrastructure crisis.
A 2025 Duke University study found that if data centers can be flexible for just 0.25% of their operating time—representing just 22 hours per year—the U.S. grid can accommodate 76GW of new data center load without building new power plants.
Forward-thinking data center operators are now implementing a wide range of new strategies to deliver exactly this type of grid-connected flexibility.
Energy Storage Systems
For example, advanced storage technologies are quickly revolutionizing how data centers manage power fluctuations and grid interactions. Beyond traditional lithium-ion systems, data centers are implementing diverse storage solutions, including alternative chemistries like aluminum-ion and sodium-ion systems, solid-state technologies that offer enhanced safety profiles, and flow batteries that provide scalable, long-duration storage capabilities.
Additionally, emerging approaches such as thermal storage systems and gravity-based storage are also gaining traction for their ability to store large amounts of energy with minimal degradation over time.
These systems excel at managing the daily “duck curve” of renewable energy availability: storing excess solar generation during peak production hours and releasing it during evening demand spikes. This capability allows data centers to maintain consistent operations, reduce strain on the grid during peak demand periods, and maximize their use of renewable energy sources.
Virtual Power Plants
Additionally, virtual power plant (VPP) networks are transforming facilities from simple energy consumers into active grid participants. By connecting their storage or backup generation systems into coordinated networks, data centers provide grid stability services, participate in demand response programs, and optimize renewable energy utilization across entire regions.
While these flexibility innovations show significant promise, they also highlight the fundamental infrastructure challenges that must be addressed to fully unlock AI’s potential.
Why We Need to Focus on the Grid
While so much attention tends to focus on whether the U.S. can generate enough power for AI’s explosive growth, the real challenges lie in two areas: making data centers more flexible so they can work with existing grid capacity and upgrading transmission infrastructure to deliver power where it’s needed.
Flexibility Over Generation
The data center industry isn’t necessarily facing a fundamental power generation problem. The U.S. grid already maintains substantial reserves, with operators typically utilizing only about 50% of available generating capacity by design. This conservative approach ensures adequate power remains available to meet peak demands that occur during a small fraction of the time, such as the coldest winter mornings and the hottest summer nights.
This creates real potential for new operations strategies, such as demand response capabilities, energy storage systems that shift power usage to off-peak hours, and workload management that schedules non-urgent computing during times of grid abundance. Google’s recent agreements with Indiana Michigan Power and Tennessee Valley Authority represent the first implementations of data center demand response targeting machine learning workloads specifically.
In these cases, Google will temporarily reduce or pause energy-intensive AI and ML tasks during peak demand periods, enabling their data centers to operate as more flexible resources that don’t further strain the electrical grid.
The Transmission Bottleneck
While flexibility solutions help, the real constraint isn’t power generation capacity but utilities’ ability to deliver electricity where it’s needed. Transmission and distribution limitations, rather than generation shortfalls, are creating the bottlenecks that utilities must address through infrastructure investments.
The numbers tell the story. For example, Dominion Energy’s transmission constraints have created multiyear backlogs for Northern Virginia data centers, with relief coming only when new grid infrastructure comes online in 2026. This pattern repeats across major markets, where planning and construction of new transmission lines can take five to ten years, depending on regulatory approval, permitting, and right-of-way acquisitions.
In California, PG&E has stated that new substation work for large load interconnections may take five years or more. Additionally, all six major regional grid operators outside of Texas have reported they cannot meet the Federal Energy Regulatory Commission’s (FERC) deadline for critical transmission upgrades.
These timelines simply don’t line up. While data center developers can build facilities in two to three years, the time required to complete associated interconnection studies and infrastructure upgrades usually spans four to eight years. Goldman Sachs Research estimates that $720 billion of grid spending through 2030 may be needed, but the infrastructure development cycle simply can’t keep pace. This explains why operators are increasingly forced into expensive behind-the-meter (BTM) solutions rather than waiting for grid connections that may take half a decade or more.
Next Steps for Stakeholders
Moving forward means addressing both sides of the equation: transmission bottlenecks and data center innovation working together. While grid modernization, faster permitting, and better transmission technology are essential, data center operators are also stepping up with solutions that reduce grid dependence. The combination of smarter infrastructure and more flexible facilities will unlock sustainable AI growth.
The industry response shows that progress requires both utility infrastructure and data center adaptation. As the Duke University research demonstrates, even minimal data center flexibility can accommodate massive new capacity without additional generation. Meanwhile, behind-the-meter power solutions, clean energy adoption, and strategic facility placement are creating new pathways that work around current grid limitations.
Making this work requires coordination on multiple fronts. Utilities need to update their planning and speed up approvals while data center operators continue to utilize flexible power solutions such as energy storage, demand response, and behind-the-meter solutions. Regulators should prioritize transmission investment while encouraging data center innovations that support grid stability. Both sides are bringing solutions to the table.
The data center industry is adapting with proven technologies while pushing for better grid infrastructure. This isn’t a crisis spinning out of control but more of a two-sided infrastructure challenge where both transmission upgrades and smarter data center design will create more efficient and sustainable digital infrastructure. The question isn’ t whether one side can solve this alone, but how quickly both transmission modernization and data center innovation can work together to power AI growth sustainably.
