It’s obvious that energy industry leaders have been talking about artificial intelligence a great deal late, said Benjamin Della Rocca, national security lead at Anthropic, a leading AI company.

But it’s still uncertain, Della Rocca said, exactly what AI’s energy needs are going to be in the upcoming years.

“AI’s adoption is really changing the way that the core parts of our economy work,” Della Rocca said Wednesday at a Carnegie Mellon University Energy Week panel discussion.

“It’s vital, of course, for the energy transition and for the way the power sector evolves,” he said. “It’s vital for national security.”

The discussion came as the Defense Department has deemed Anthropic an “unacceptable risk” to national security, asserting that the startup could disable or alter its technology to suit its own interests, rather than the country’s priorities.

Della Rocca said he was unable to comment when asked about those developments, but did say Anthropic’s “commitment and product has been to work productively with the government, across all sorts of government missions.”

Before joining Anthropic last year, Della Rocca served in the Biden administration as director for technology and national security on the National Security Council and as a senior policy advisor for the National Economic Council.

“Scaling energy is essential for the United States and our own company’s leadership in artificial intelligence. There’s a wide range of challenges needed to bring energy online at the scale that we need,” Della Rocca said. “And on the other hand, there’s a lot of really important work to be done to make sure that as we meet those challenges, we do it the right way.”

Doing it the right way, Della Rocca said, includes addressing risks or challenges that might come to locations where data centers and resources are built. Locally, data centers are planned in Upper Burrell and Springdale.

It’s also important, Della Rocca said, for data centers to be designed in a way that they’re consistent with grid reliability and can give power back to the grid.

A wide range of energy solutions will be needed to meet AI’s energy demands, Della Rocca said. In the United States, the AI sector needs electric capacity of at least 50 gigawatts by 2028, he said. By comparison New York City, during peak summer months, needs a capacity of 11 gigawatts.

Expanding capacity to meet AI’s needs requires innovation to access more efficient and effective sources of power and leveraging the existing grid to get more out of it, Della Rocca said. There could be opportunities with nuclear or geothermal energy, he said.

“Building all of this infrastructure is very complex. It requires not just building data centers, but generation and transmission infrastructure, which involves lots of complications,” he said.

Della Rocca acknowledged that an unintended effect of rapid AI data center development is rising electricity prices.

He cited Anthropic’s response — the company announced last month it would cover costs of electricity price increases from its data centers. That could range from the company bringing on new generation to meet demand, or working with utilities for its own pricing, or investing in curtailment systems that cut Anthropic data centers’ power usage during periods of peak demand.