This year’s Baker Hughes Annual Meeting in Florence, Italy, highlighted just how far interest in data centers has expanded beyond the IT sector. A large crowd of delegates from the energy sector gathered for a breakout session on data centers last week, where the discussion centered on the strain generative AI is placing on power infrastructure. The scene was set with a Goldman Sachs projection that data center power demand will grow by 165% by 2030 – raising concerns about whether local grids can keep pace without compromising reliability.
Panelists from the data center, government, and energy sectors chimed in with their opinions, offering insights into the challenges and potential solutions for balancing AI-driven demand with grid stability.“The data center market was growing steadily due to the cloud and has been supercharged by AI,” said Jeffrey Tapley, COO of data center owner-operator Digital Realty. “Three years ago, we had an oversupply of data centers, but AI is now pushing up price rates for data center space, which is doubling in some markets,” Tapley explained that transmission and power constraints exist in many municipalities. He noted that grid connections for new data centers are often delayed by a year or more, even after initial assurances of power availability.
Location
Hyperscalers are demanding massive power near AI development hubs, where large language models (LLMs) are trained in data centers. Since the ‘learning phase’ of AI isn’t latency-sensitive, proximity to end users isn’t a priority. Even so, Tapley said he was hesitant to build data centers solely dedicated to these workloads.“If you colocate a data center for AI, that site might not be needed in seven years,” he said. “On the other hand, metro areas will always need more power for the ever-expanding cloud, even if AI doesn’t take off as expected. Digital Realty, then, is focused on expanding its metro area footprint as it expects that sector to continue to be strong.”
Tapley wasn’t especially keen on renewables or battery storage either. He said they take up too much land – valuable real estate that Digital Realty prefers to prioritize for its data center halls.
Another key factor in data center site selection is proximity to the natural gas network. Tapley noted that gas power is a more viable option for facilities located near existing pipelines, making it easier to justify investment in onsite generation.
Onsite Efficiency for Tomorrow’s Data Centers
Jonathan Maxwell, founding partner and CEO of Sustainable Development Capital LLP (SDCL), highlighted systemic energy inefficiencies and stressed the need for onsite solutions. SDCL, which has $2.5 billion invested in onsite energy for industry and transport, has significantly increased its focus on data centers.“The dirty little secret is that most energy is wasted: 8% in extracting and converting it, 50% in heat losses, and 5-10% at the point of use,” said Maxwell. “By the time data center operators pay for expensive power, most of it is wasted.”The way to solve this issue, he argued, is by building energy solutions where they are most needed – onsite. In the past, 25% of SDCL’s business has been in data centers. It has now ramped up to 50-60% of current projects.“If you can generate energy locally, it is usually cheaper than the grid,” said Maxwell.
The CEO sees batteries playing an increasingly important role as data centers gradually phase out diesel generators for backup power. After all, batteries are a familiar technology courtesy of uninterruptible power supply (UPS) systems. Natural gas is another option that could be used to replace diesel, he added. Maxwell recommended a review of ‘cogeneration’ options such as the use of waste heat to provide steam for heating and cooling. This is a good way to improve project economics. Greg Joiner, executive vice president of renewable power generation at Shell Energy, concurred.“We need more batteries and gas-peaking units to smooth out demand,” he said. “Hybrid solutions give the balance of affordability, sustainability, and security.”
U.S. Infrastructure Constraints
The final panelist at this year’s Baker Hughes session was Dan Brouillette, U.S. energy secretary during President Trump’s first term. He continued the theme of infrastructure constraints impacting the high-growth tech space. Despite the new president’s recurring “Drill, baby, drill,” mantra, significant hurdles stand in the way of utilizing the abundant energy resources of North America. The region’s huge oil and gas reserves are hard to transport to key internal markets and the coast for export.“We need better pipelines and many more of them to be able to move gas to the ocean,” said Brouillette.
Grid-wise, he cited forecasts that the U.S. would be close to 30 GW short of capacity in the next five years. That was before the ongoing surge in data center demand. Waiting around for power and gas lines is not an option for those hoping to capitalize on AI.“Colocation will be more and more the future,” said Brouillette. “Bring your power is going to be a major trend.”Like Tapley, Brouillette doesn’t see renewables as an immediate answer to AI power demands. Land use needs and a capacity factor of around 30% will limit the role of renewables in AI data centers. Similarly, nuclear development will take too long.The federal government may be easing regulations and accelerating the permitting process for infrastructure, but there will still be plenty of state and local approvals to be earned. And then there is the specter of litigation hovering over data center expansion.
