Balancing AI’s Energy Demands: The Role of Gas, Nuclear, and Renewables

• The rise of AI requires an increase in reliable baseload power.
• Oil and gas executives have predicted that natural gas could see increased demand from AI-driven power demand.
• To avoid a return to hydrocarbons, some have suggested nuclear as a source of the reliable baseload electricity supply that artificial intelligence development needs.

Artificial intelligence is all the rage these days. So is its carbon footprint, which has sparked an awkward conversation about its supposed benefits for the transition and how they stack up against its substantial energy needs. The bigger question, however, is how these needs are going to be satisfied, because it seems AI is not going anywhere, whatever its footprint.

Advocates claim AI will help drive the energy transition forward. Critics point to the fact that data centers servicing AI consume even more electricity—and water, by the way—than regular data centers, which are already considerable electricity consumers.

This has led to a kind of split in the transition camp, with activists speaking out against AI because of that footprint and corporate transition supporters backing the technology, claiming it would be instrumental in making the transition happen. As for that carbon footprint, opinions vary.

One popular opinion is that artificial intelligence would reverse the shift away from hydrocarbons because those data centers need a stable, reliable, uninterrupted supply of electricity—and hydrocarbons are the only such easily accessible option out there.

“It will not be done without gas,” EQT’s Toby Rice told the Financial Times about the AI revolution, and it is difficult to argue with that, even accounting for the fact that Rice has a significant interest in that being true.

Related: ADNOC Moves Ahead With Huge LNG Export Project in UAE

To avoid a return to hydrocarbons, some have suggested nuclear as a source of the reliable baseload electricity supply that artificial intelligence development needs. As a result, nuclear developers are, like gas producers, chomping at the bit to get going with building more generating capacity.

Small modular reactors are the talk of the town when it comes to nuclear for data centers. A recent article in Forbes outlined the latest in this field, reporting on the story of NuScale Power, a small modular reactor developer, which had recently struck a deal to supply 24 of these reactors to data center developer Standard Power. These 24 small modular reactors would be able to generate close to 2 GW of low-carbon electricity, the report said, although it also added that one short seller had questioned the validity of the deal.

However, it was NuScale again that last November said it would terminate its first small modular reactor power plant project that it was developing with Utah power utilities. The termination came after a decade of work on the project and was considered a heavy blow to the prospects of small modular reactors. They appear to be a bit similar to nuclear fusion – still around the corner and likely to remain there for longer than many hope.

But what about standard, full-size nuclear? Even the International Energy Agency likes nuclear because it does not produce any carbon emissions once it starts operation. In fact, the IEA and other transition forecasters have made a point of noting that the transition would be a lot more challenging without nuclear in the energy mix. And with artificial intelligence draining the grid, nuclear has become even more topical.

According to one energy sector executive, however, nuclear power is “overblown” as a source of stable energy for the data centers housing AI. “The question is, going forward, what’s the price of new nuclear,” said AES Corporation’s Andres Gluski, speaking to CNBC. The comment suggests that nuclear is too expensive to deploy more of it, with Gluski predicting that most of the energy supply for data centers will come from “renewables”, meaning wind and solar.

He pointed to the recent deal between Microsoft and Brookfield Asset Management, which agreed on 10.5 GW of low-carbon electricity capacity that Brookfield will somehow “deliver” to the IT giant as part of the latter’s plans to eventually source 100% of the electricity it uses 100% of the time from low-carbon sources, either directly or indirectly. According to Gluski, the deal shows that the future is in wind and solar, and not nuclear.

Yet wind and solar have proven time and again to have certain shortcomings, chief among which is their variable, weather-dependent output. This is not the sort of electricity data centers—or anyone else, really—needs. Battery backup at this scale would have a price tag of cosmic proportions, so that is not an option at the moment.

There is also the detail that even if Brookfield builds those 10.5 GW across the world, Microsoft will not be sourcing electricity from them to power its data centers. Microsoft will most probably be buying the clean energy credits that governments award these generators in order to claim emission-free electricity, regardless of where it sources its actual supply from.

The marriage of artificial intelligence and the energy transition would be a tricky one. Doubts that AI can derail the transition are multiplying. Nuclear power, for all its upfront costs, could be the solution to the problem.

By Irina Slav for Oilprice.com

More Top Reads From Oilprice.com:

• OPEC’s Oil Production Edged Higher Ahead of Key Policy Meeting
• Mexico Raises Proven Hydrocarbon Reserves Figure
• Europe's Shift from Russian Gas to Pricey LNG