During a House Oversight Committee hearing prior to the congressional recess, Rep. Julie Fedorchak (R-ND) questioned energy experts about making artificial intelligence development more efficient.
Category
🗞
NewsTranscript
00:00Thank you. I now recognize the gentlelady from North Dakota, Mrs. Fedorchak.
00:07Thank you, Mr. Chairman, and thank you for the opportunity to sit in on your committee today.
00:13I'm a member from North Dakota, a new member, and I am on the Energy and Commerce Committee,
00:20but am very interested in this issue, particularly the powering of America's AI industry.
00:28I spent 12 years as a utility regulator, most recently was president of the National Association of Utility Regulators,
00:36so I'm very familiar with a lot of the issues regarding our energy markets, the signals we're sending to our energy markets,
00:44the massive increase in demand that we're talking about here.
00:50When I look back on the 12 years as a regulator in my state, and we served during a time of energy boom,
00:58so we saw a demand increase, it was still like 3%.
01:02So when you look, I was in a meeting this morning with PJM, and they said that their peak demand today is 150 gigawatts,
01:12and they're looking in five years to be at 185 gigawatts.
01:16That is mind-blowing. That is astounding.
01:20And to meet that demand, I appreciate, requires a level of development that we have never before done.
01:29There's no time in history that we've grown that much power and connected it to the grid that quickly.
01:36So with that, I have just a couple questions for you experts, and I invite you to participate.
01:41We've got a working group. My office has led on AI and energy.
01:46We want to try to get some solutions and a policy framework for solutions to bring to the table on this very issue.
01:53But what types of efficiency upgrades can we count on, I hope, to mitigate that power demand increase?
02:06Because AI is getting more and more efficient, and hopefully some of these projections will come down
02:12because of efficiency increases and improvements.
02:15So for any of you, I invite you all to comment on that.
02:18Maybe with you, Mr. Mills, you want to start?
02:23The efficiency, the rate of improvement in energy efficiency of compute and GPUs
02:28is faster than the vaunted Moore's Law efficiency gains.
02:32And the rate of efficiency gains in the cloud are even faster yet
02:37because efficiency gains in the transmission of information, in the memory systems,
02:43all the associated are going extremely fast.
02:45What's happening is that the demand for the product software is growing even faster
02:51because what efficiency gains do in computing is it makes the computing cheaper, easier, faster,
02:57which for quite a long time is going to overtake all of the ostensible efficiency gains.
03:03I think the principal solution will come from, and I will say I endorse a blend of what Mr. Slocum has said
03:09and what a lot of other people are saying, to your point, that we can't build that much capacity that quickly.
03:15It looks like we can't for a whole set of supply chain reasons.
03:19So we're going to have to figure out how to meet the demand for the services
03:23while moderating the incredible demand for power.
03:28I think one of the solutions will be increased use of where the utility industry was 100 years ago,
03:35which is loads this big, which are industrial class loads.
03:39They're purely in the sort of refinery steel mill.
03:43We'll increasingly see on-site generation, independent generation, and even co-generation.
03:48Co in the sense of not heat and power, but peak shaving for the grid without costing consumers money.
03:55Every utility I talk to, every utility CEO, does not want to raise the costs of their electricity
04:02for their consumers to serve an industrial load.
04:06This has been a difficult challenge.
04:09For sure. Thank you.
04:11Mr. Jim, go ahead. Were you finished?
04:14Do I have another minute?
04:16Yes.
04:17Mr. Levin?
04:18I appreciate the question, Representative.
04:19Yes, if you can.
04:21I would just add that data centers are highly efficient facilities,
04:23but they also enable energy savings and efficiencies economy-wide.
04:28I talked earlier about just, you know, the changes in the last 15 years walking into a data center,
04:33the ambient temperature inside.
04:34You know, servers generate heat.
04:37You know, early days you had to wear a windbreaker going into a data center because it was that cool.
04:41Right.
04:42Now, ambient temperatures have increased substantially.
04:45You know, you can go into data centers that are 80 degrees or higher,
04:47and that's part of the energy efficiency gains that we've got within the data centers
04:51if you're looking at cooling, but I also think it's important to recognize the role of data centers
04:55in enabling energy efficiency for others, whether you think about smart thermostats,
05:00whether you think about gets and dynamic line monitoring,
05:03or any number of technologies that are utilized by consumers,
05:07that are utilized by businesses, utilized by government.
05:09All of those rely on the digital infrastructure that data centers are providing.
05:14Good point.
05:14And I think there's not only substantial recognition within the data centers,
05:18but certainly economy-wide of the efficiencies gained.
05:21I'm out of time, but I will say North Dakota has cool air and we're flaring gas,
05:26so come there, use that gas, and generate power.
05:30Appreciate it.
05:30I yield back.