00:00Kevin, I want to start where we left off with Emily just now and this bottleneck that we're
00:04seeing when it comes to developing these data centers. What, in your view, would be the right
00:08policy to decrease the U.S. reliance on China for these key parts? Yeah, so first and foremost,
00:14we need to consider this AI infrastructure build out as a national challenge. We've seen that
00:19across 11 different states, there have been considerations of bans or years-long moratoria
00:25on data center creation. And as you know well, data centers aren't just for AI. We need data
00:31centers for health care, for commerce, for all the things that make the internet go. And so we need
00:37to really emphasize that states, cities, and the national government all have a role in making sure
00:42we can build AI infrastructure that's aligned with our AI ambitions. Okay, so yes, in my backyard is
00:48certainly, in your view, a big part of this, but that still doesn't address the component problem
00:53that Emily and the team highlighted in her piece. How do you make sure that the U.S. actually has
00:58access to these key components, these key parts, when China has a stronghold on them? So this is
01:04where we really do need to work with our allies in developing and strong-arming our supply chain
01:10network across the entire AI tech stack. As you've noted, we've seen that when it comes to chips, when it
01:16comes to things like helium that's being locked up in Qatar, we have a long ways to go in terms
01:21of
01:21making sure that we have that sort of resilience and redundancy that's required to build AI at scale.
01:27And so this requires, ultimately, unsurprisingly, funding and political leadership. And right now
01:32we're seeing that there's a sort of tendency to not be swimming in the same direction on this AI
01:38question. And that's in part for understandable reasons, but ultimately we need to make this a
01:42national priority. Funding from whom, though, Kevin? A national priority, does that mean you want to
01:46see the U.S. government and taxpayers shell out money for this stuff? I do want to see increased
01:51investment from the U.S. government in this domain. We've seen over the history of our nation that when
01:56it comes to critical infrastructure, ultimately the central government does have a role in making
02:01sure we have a national highway system and making sure that our port system is resilient to go all
02:06the way back to the Articles of Confederation. So, yes, this ultimately is a core purpose of the
02:11government is to make sure that basic infrastructure for the technologies of an
02:16age is available and not subject to geopolitical conflict.
02:20So your view is that infrastructure such as this is as important, at least historically in the
02:25current context, as highways, as ports, as other huge national projects. How do you get leaders on board
02:31with this? So first and foremost, we have to do a lot of work in terms of just making sure
02:36we're
02:37addressing a lot of the myths that are being spread about this AI build out. So, for example, there was
02:42an article recently that made headwinds and made the rounds on social media about how data centers
02:48were creating heat islands wherever they were being built. Well, if you talk to folks like Andy
02:53Masley, who studied this issue closer than just about anyone, we see that a lot of this information is
02:58in fact bunk. And so the more that we have sort of misinformation flowing about how AI is being used
03:04and
03:05developed, those are ultimately distracting questions. And we need to move on to the questions
03:09related to the fact that AI is increasingly a national security imperative. We're seeing right
03:15now, for example, in the war in Iran, that militaries are using AI in ever more sophisticated domains.
03:21And it's a real national imperative that the U.S. military have access to leading frontier models
03:27that are the most capable and that are the most reliable across the world. Kevin, you study this stuff,
03:32so maybe you have an answer because it certainly doesn't seem like a lot of people do have answers
03:35to this. If we look at increased demand for electricity across the grid and the way that
03:41the grid is being taxed in a way now that many people didn't think it would be in the year
03:452026
03:46and the projections for the future, how do you balance the desire and the need in your view to
03:51build these data centers in the U.S., draw on power from American power plants, but also make sure
03:57that American consumers aren't seeing hiked power bills as a result. Is it even possible?
04:04I would say it is indeed possible. So I'm a man of many tired lines. One of my tired lines
04:09is
04:10today's AI is the worst AI we will ever use. I'll also say that today's data centers are the worst
04:16data centers we'll ever use. There is a huge incentive for these hyperscalers to invest in
04:21means for more efficient data center operations. And we're seeing that states like Oklahoma,
04:27for example, are enabling behind the meter power that allows for these labs to build out more
04:32renewable and efficient data center operations. And so we've actually seen that when you talk to folks
04:38like Gavin McCormick at Climate Trace, that there are opportunities for labs to actually assist
04:44with making the grid more resilient and actually increasing the grid capacity by virtue of doing training
04:50runs, for example, at low hours of typical power usage.
Comments