Powering AI: Modular Data Centers can Unlock Stranded Power

Use modular data centers to unlock stranded grid power for AI, speed deployment, reduce water use, and ease community impact.
By Russell Boyer |
August 26, 2026August 24, 2026

Topics in this article [AI Solutions](/en-us/blog/tags/ai-solutions/)[Data Center](/en-us/blog/tags/data-center/)[Energy](/en-us/blog/tags/energy/)[Public Advocacy](/en-us/blog/tags/public-advocacy/)

Key takeaways 7 min read

-

Modular data centers can unlock stranded power on existing grids to deploy AI capacity faster, with lower community impact.

• Right-sized, liquid-cooled MDCs align well with latency-sensitive AI inferencing workloads across distributed locations.

• This model supports grid reliability, resilience, affordability and sustainability while easing permitting and deployment timelines.

Artificial intelligence is creating a very real infrastructure challenge. Demand for AI capacity is accelerating, yet large new data center proposals face growing resistance over land use, water consumption, noise, visuals and added grid strain. That tension raises an important question: what if some of the power needed to support AI is already available at existing points on the electric grid?

Dell Technologies sees a practical path forward: use modular data centers to tap existing, underused capacity on the grid and bring AI online faster, with fewer tradeoffs for communities.

The stranded-power opportunity is larger than many realize

Stranded power, in this context, is available but underutilized electrical capacity at existing grid locations, such as substations or adjacent parcels, where infrastructure headroom exists but is not currently monetized or used productively. These increments often range from tens of kilowatts to several megawatts and can be aggregated across many sites.

Utilities and grid operators build headroom to handle peaks or future growth that may not fully materialize. The result is pockets of available capacity at substations and similar electrical sites. Multiplied across thousands of locations, that becomes a meaningful, distributed opportunity for new compute. It also matters because the traditional alternative is getting harder. Large transformer lead times remain a major bottleneck, which slows and raises the cost of new substation builds. Co-locating AI infrastructure next to sites that already have transformer capacity can help sidestep that constraint. For background, see the U.S. Department of Energy’s overview of the current transformer supply landscape at the Grid Deployment Office.

There is a practical benefit for utilities and landowners as well. While utilities generally will not place third-party equipment inside the substation yard, adjacent parcels can be viable. That creates a path to new revenue from underused assets without interfering with core utility operations.

Why modular data centers fit this moment

Modular data centers match the scale of the opportunity. Hyperscale facilities are often designed around very large deployments, while stranded-power locations tend to offer capacity in the 15 kW to 10 MW range. Modular designs are built for right-sized, distributed infrastructure.

Right-sizing also changes the economics and the delivery model. Smaller, modular deployments can mean lower upfront capital requirements, faster time to value and less permitting complexity than a traditional large-scale facility. Dell’s modular approach is factory-integrated, tested and shipped ready for deployment. That helps compress the timeline from site selection to production workloads while reducing on-site construction activity. Explore the portfolio and design approach and the MDC brochure.

Cooling is another reason modular matters now. Liquid-cooled modular data centers are especially relevant for AI because they can reduce water use and improve energy efficiency, directly addressing two common community concerns. For a look at how liquid cooling extends to ruggedized edge designs, see this independent overview.

A clearer comparison

The table below summarizes how traditional large-scale AI builds compare to modular deployments at stranded-power sites.

A better answer to the community conversation

Community concern around large projects is now common. Planning boards and councils are weighing aesthetics, noise, water demand, traffic and impact to the grid. A modular data center presents a smaller visual footprint, blends more naturally with industrial or utility-adjacent settings and shortens on-site construction, which reduces disruptions.

Water use is equally important. Liquid-cooled MDCs can drastically reduce water consumption and eliminate dependence on evaporative cooling towers, which preserves local resources. From an energy perspective, the model is different from a conventional large new load: instead of requiring entirely new capacity, these deployments are designed to use unused headroom already present on the grid.

The result is a different kind of infrastructure story. Instead of imposing a campus-scale build, modular data centers at stranded-power sites put dormant assets to work, add local tax and economic value and do so with fewer of the tradeoffs that typically trigger opposition.

Strengthening the grid while supporting AI growth

This model is not only about deploying compute faster. Co-located MDCs can participate in demand-response programs, curtailing load during grid stress events and acting as a more flexible resource for grid operators. For an overview of demand response as a grid tool, see the U.S. Department of Energy’s summary.

Four outcomes that matter to utilities and communities:

-

Reliability: Flexible, right-sized load that can adjust with grid conditions.

• Resilience: Smarter siting and distributed capacity close to demand centers.

• Affordability: Monetizes underused infrastructure and defers major upgrades.

• Sustainability: Enables growth without immediate new generation or large builds.

Why inferencing may be the real near-term fit

Training clusters will remain centralized for scale and interconnect reasons. Inferencing is different. It benefits from proximity to users and data, thrives on distributed capacity and is highly latency-sensitive. Modular deployments across multiple stranded-power sites can place inferencing where it is needed most, close to enterprise demand centers, service provider POPs and sovereign workloads, without concentrating risk or impact in a single location.

A smarter path forward

The AI era does not need to hinge on massive new power plants, drawn-out land battles and strained community relations. There is another path: use modular data centers to unlock stranded power near existing grid infrastructure.

For Dell Technologies, the opportunity is clear. Modular infrastructure can bring AI capacity online faster, more quietly and with less friction than traditional alternatives, while giving utilities a practical way to derive value from existing assets and strengthen grid performance. To explore how this approach could work at your sites, visit the Modular Data Center Solutions page.

About the Author: Russell Boyer

A passionate innovator, Russell leads the Utilities Industry practice at Dell Technologies focused on advancing human progress by developing solutions that accelerate the energy transition. His responsibilities include strategy, solution development, partner management, and go-to-market enabling the digital transformation of the energy industry. With over twenty years’ experience, he is a proven utility solutions expert, applying policy, regulations and standards to drive technology solutions which deliver business outcomes. A firm believer in accelerating innovation, he fosters a global utility industry community and ecosystem collaboration. He is a serving member of the GridWise Alliance developing an innovation framework for advancing grid modernization.

Russell graduated from Texas A&M University with an MBA and a Bachelors in MIS. He enjoys spending time outdoors, riding mountain bikes, and spending time with family at the lake.

Topics in this article [AI Solutions](/en-us/blog/tags/ai-solutions/)[Data Center](/en-us/blog/tags/data-center/)[Energy](/en-us/blog/tags/energy/)[Public Advocacy](/en-us/blog/tags/public-advocacy/)