In 2026, the data centre industry finds itself caught between two pressures that are pulling it in opposite directions. Demand for compute capacity – driven by AI workloads that are more power-intensive than anything designed for – is accelerating. At the same time, utilities and regulators are scrutinising the grid impact of large-scale facilities as never before. For operators, the question is no longer simply how to build fast, but how to build responsibly.

Giampiero Frisio, President of the Electrification Business Area at ABB, believes that framing the debate as a conflict misses the point. "The biggest challenge today is growing quickly while still being responsible to the local community," he says. "People often think of data centres as energy-hungry buildings that strain the power grid. However, new AI factories can actually help the grid instead of hurting it."

Giampiero identifies three interlocking pillars that define this approach. The first is radical efficiency – reducing the energy lost in the conversion steps between grid power and compute load. The second is distributed energy and on-site generation, enabling facilities to draw less from the utility at peak times. The third is data-driven engagement with regulators, using digital simulation to demonstrate that high-power loads will not compromise local grid stability.

The case for 800V DC architecture in high-density AI workloads

At the centre of the efficiency argument is a shift in the fundamental architecture of how power is distributed inside a building. Traditional data centre infrastructure converts power multiple times between alternating current and direct current – each conversion introducing losses and requiring bulky physical equipment. The emerging alternative is to keep power in DC form from the point it enters the facility until it reaches the server rack.

ABB's research indicates that a DC-based architecture can reduce total energy costs by approximately 9%. Applied to a gigawatt-scale facility, that figure carries significant weight, both commercially and in terms of the total draw placed on the utility grid.

Within that DC architecture, 800 volts is emerging as the technically and economically optimal distribution voltage for AI applications. 

"Higher voltage allows more power to be transmitted through smaller conductors with lower losses," Giampiero explains. "In an AI Factory, 800V DC enables the delivery of massive power blocks directly to the rack while minimising the conversion tax."

The benefits extend beyond efficiency. At 800V, modern components such as Solid State Transformers and Solid State Breakers operate at their most effective, able to isolate electrical faults in microseconds – a critical capability when protecting AI accelerators that represent multi-million-dollar investments. The architecture also simplifies procurement. By standardising on components that are engineered to connect quickly, operators can reduce construction timelines by between 30% and 50%, a material advantage in a market where speed to capacity is a competitive differentiator.

How battery energy storage systems are turning data centres into grid assets

For years, utilities have regarded large data centres with a degree of concern, particularly as AI workloads introduce demand patterns that are harder to predict than traditional compute. Battery energy storage systems (BESS) are beginning to change that dynamic by giving operators a tool to actively manage their grid relationship rather than simply consume.

The most direct application is peak shaving. When grid stress is high – during extreme weather events, for instance – a facility can shift a portion of its load onto on-site batteries, instantly reducing its draw from the utility and protecting supply for other consumers in the area. Over time, as smart grid technology matures, the same systems could allow data centres to sell capacity back to the network. "This turns the data centre into a virtual power plant that supports the broader community's stability," says Giampiero. "Instead of competing with the community for power, the data centre becomes a critical infrastructure partner that helps the utility manage a more complex grid with significant renewable energy."

The integration of DC-native BESS with on-site solar generation is particularly efficient. Because both technologies operate in DC, they can be connected directly without additional conversion losses, creating a resilient local energy system capable of operating independently if grid supply is disrupted.

Commercial barriers to adoption are also falling. New as-a-service payment models allow operators to treat battery infrastructure as an operational rather than capital expenditure, removing the upfront cost that has historically slowed deployment.

The financial and environmental case for retrofitting over replacement

The instinct to demolish and rebuild when facing ageing infrastructure is understandable, but Giampiero argues that it is rarely the most rational course. "The rip-and-replace mentality is often driven by the fear that current designs will be unsuitable in 3 to 5 years," he says. "However, there is a compelling financial and environmental case for hybrid modernisation and retrofitting."

On the financial side, full replacement carries significant drawbacks: high upfront capital, extended construction periods and the operational disruption of taking a facility offline. A phased approach – introducing high-efficiency components selectively, targeting the areas of heaviest AI load – allows operators to improve performance incrementally, with faster returns and without large-scale service interruption. Modern power conversion equipment also takes up less physical space than the systems it replaces, meaning operators can increase effective capacity without expanding the building's footprint.

The environmental calculation is equally persuasive. Demolition generates waste and carbon; new construction consumes raw materials including copper and steel. Upgrading in place avoids those impacts while still delivering meaningful efficiency improvements. "The greenest building is usually the one that is already standing," Giampiero notes. "By making an old building up to 9% more efficient, companies can hit their environmental goals without the waste of a brand-new construction project."

Data centres and utilities: Towards an energy partnership by 2030

The longer-term evolution of the industry points towards a fundamental change in the commercial relationship between data centre operators and utilities. Rather than the transactional dynamic of buyer and seller, Giampiero anticipates something closer to a joint infrastructure partnership, with solar playing a central role in the transition.

Solar is well-suited to this because it generates DC power natively and can be deployed at scale relatively quickly, without the planning and construction lead times associated with large grid infrastructure. The rise of energy-as-a-service models – where a third party or the utility itself owns and operates generation and storage assets, with the data centre paying for consumption – is reducing the barrier to adoption further. "This helps data centres avoid high upfront costs while meeting their green goals and letting energy experts handle the grid," Giampiero says.

For utilities, the model offers a way to accommodate the power demands of AI without committing to large centralised generation or transmission projects. Local micro-grids, built in partnership with data centre operators and validated through digital twin simulation, offer a more flexible and responsive alternative.

The vision Giampiero describes is one in which the data centre is no longer simply an end-user of grid power. "The best data centres won't just be buildings full of computers," he says. "They will be energy hubs where operators, utilities, and tech partners such as ABB and NVIDIA work together to deliver the reliable, efficient power AI needs."