Policymakers have spent years talking about rebuilding America’s industrial base, reshoring critical supply chains, strengthening defense production, and reducing U.S. dependence on China. But recognizing the need to build is not the same as having the ability to do it.
Erin Price-Wright, general partner on Andreessen Horowitz’s American Dynamism practice, joins the AI Policy Brief to make the case that AI isn’t just a software story. It’s the defining factor in the sectors that determine whether the U.S. can build, power, and defend itself in the decades ahead.
Erin invests in companies applying AI to the physical world, with a focus on robotics, energy, manufacturing, industrials, and defense. She and Matt Perault discuss how AI can help make the math work for building in the U.S. again—from accelerating mine permitting and coordinating complex industrial projects to designing factories, lowering the cost of automation, and bringing robotics to more factory floors.
They also discuss where policy needs to catch up: the laws and regulations that make it too hard and slow to build new factories in the U.S. and defense procurement that still favors incumbents over startups.
Finally, they discuss how the debates over data centers and jobs will shape whether America’s reindustrialization effort succeeds.
The takeaway: if the U.S. gets the policy environment right, AI can strengthen the industrial base, help create new kinds of jobs, and give America a powerful competitive advantage.
Topics covered:
00:00: Intro
00:54: Erin’s work investing in AI for the physical world
01:47: Why AI and reindustrialization are converging now
03:07: Applying AI to mining and critical minerals
08:28: What Ukraine reveals about defense production
18:13: How startups are breaking into government markets
20:16: Bringing a factory mindset to critical sectors and complex systems
24:39: How robotics can expand factory automation
30:22: What still makes it too hard to build in the U.S.
32:49: Using AI to design better, cheaper manufactured goods
35:05: Why data centers matter for reindustrialization
39:46: How compute could help modernize the grid and lower costs for consumers
42:50: Why AI could create new industrial jobs
This transcript has been edited lightly for readability.
Erin Price-Wright (00:00)
The intersection of AI and the physical world has been my lens for how I think about a lot of my investing here. It’s a really important and interesting moment in AI. It’s also a really important and interesting moment in the kind of re-industrialization of the United States. And these two waves are coming together at sort of just the right time.
We’re thinking about it in the sense of like, what things could we not build in the U.S. before that we could build now? There’s this big moment and large appetite to figure out how to bring more of that back and reshore in the U.S. And frankly, there’s no universe where we can do that without heavily relying on AI.
Matt Perault (00:38)
Erin welcome to the podcast.
Erin Price-Wright (00:46)
Thanks, it’s so fun to be here.
Matt Perault (00:48)
Yeah, it’s great to have you on. So first things first, can you talk a little bit about what you do?
Erin Price-Wright (00:54)
I am Erin, I am one of the general partners on our American Dynamism team, which means, I invest in companies that are supporting the interests of the U.S. and our allies. So we invest, you know, something like half of our portfolio companies sell directly to the government, primarily the Department of War, but also public safety, gov tech more generally. And then the other half of our companies are really supporting the build out of the broader Western industrial base. So how do we bring the latest tech into problems like energy, manufacturing, supply chain, logistics, aerospace, and many other areas that affect the day-to-day lives of Americans in this country.
Matt Perault (01:37)
So there’s a ton in there. What’s the unifying idea? Like how do you get from energy stuff to defense to gov tech? Like you listed a bunch of different things. What’s the unifying concept?
Erin Price-Wright (01:47)
Yeah, totally. So all of us on the team probably think about it a little bit differently. I spent my career before investing at Palantir and I’m just really interested in taking very hard, complicated, messy problems that we’ve been solving one way for a really long time and figuring out how to bring tech and in particular AI to thinking about how we might do those things differently.
And so the intersection of the AI and the physical world has been my lens for how I think about a lot of my investing here. It’s a really important and interesting moment in AI. It’s also a really important and interesting moment in the kind of re-industrialization of the United States. And these two waves are coming together at sort of just the right time. And I joke like, I don’t know if I am a VC because I’m an optimist or if I’m an optimist because I’m a VC and I just listen to all these like really exciting founder pitches all day long. But I think there’s no more exciting time to be investing in these kinds of physical world problems than right now.
Matt Perault (02:54)
So can you talk about a few of them? It would be great to hear some examples, because I think when people hear like manufacturing, for instance, they’re not thinking, that’s up the cutting edge of tech. What are some examples of how the portfolio is investing in the physical world?
Erin Price-Wright (03:07)
Yeah, sure. I’ll give you one example that I love because I think it sort of captures a pretty broad spectrum of what we’re talking about here. We have a portfolio company called Mariana Minerals. It’s founded by a former long-time Tesla engineer. He was at Tesla for a decade working across the Gigafactory and building out the lithium plant in Corpus Christi and the battery metals and minerals supply chain.
He did lots of really important things working very closely with Elon Musk. And he recognized, which has been a topic of conversation over the last year, that we are dangerously dependent on China for our critical minerals and materials that are important for powering the economy. And not just our defense base, but our broader industrial base. So the materials that go into the things that we use every single day, 60 something percent of them are probably mined in China and 90 something percent of them are refined in China. So all roads lead through China when it comes to things like copper, zinc, aluminum, magnesium, steel, many of the materials that we use in our day-to-day lives as critical materials. So he was like, how do I solve this? How do I bring technology to this problem, to this industry that basically hasn’t changed for a hundred years that’s using lots of the same techniques that it used in the beginning of the industrial age.
And he took his experience at Tesla, which is this real factory mindset. Like, do we construct everything as a factory and bring it to verticalized mining and refining. So the way he’s thinking about it is like, how do we deconstruct the process of what it means to build and operate a mine and a refinery into its kind of core components and understand how to turn those core components into mini factories using technology and how do we basically inject technology into every single process of owning and operating a mine, which is just not how current mining companies work. Everything from how do you use, like LLMs to basically handle a lot of the complicated coordination between all of the different parties that you have operating at a mine. You have EPCs, you have lots of different types of contractors, you have internal employees, you have folks working in the pit, you have technologists and geologists. All of these people have kind of complicated and overlapping responsibilities. And one of the big challenges, like one of the reasons why it takes 20 years to permit and build a new mine today, is it’s just a lot of different people to coordinate across with regulators, et cetera.
So they’ve come and said, okay, well, how do we use AI to make this problem better? How do we use AI to make sure that we have all the data we need when we need it? How do we use AI to make sure we’re quickly responding to all of the requests that are coming from regulators on various parts of the permitting process? How do we use AI to make sure all of these complicated contractor groups who have sort of overlapping needs are communicating effectively? And when something updates in one part of the plan, it sort of flows down through to every other part of the plan. All the way through to how do we use autonomous systems to actually do the mining? Like what are all the latest and greatest cutting edge technologies in robotics that we can actually apply to a mine site? And you know, we’re technologists, so we know how to do that. So, you know, whether it’s drill rigs or autonomous trucks or security systems, like what are all the latest and greatest in robotics that we can apply?
And then all the way down to the refining, like how do we use the latest in reinforcement learning and control systems to actually control the chemistry for how we’re extracting copper from the slush that comes out of a copper mine? And those are really expensive chemicals. That’s like mainly the driver of costs in mines. So if we can use AI to do all that better, you operate your mine more efficiently.
So I know that’s like a very long answer, but I think it just goes to show if you’re doing something complicated and physical, we’re talking about interjecting AI at basically every single stage of that business and turning it from something that’s like not economical to do into something that’s very economical to do, and doing it more safely, more cleanly, and just overall better.
Matt Perault (07:44)
So is that then the answer to why this time is different? Like the idea of reinvigorating America’s industrial base, reinvigorating American manufacturing, that’s not a new concept. We’ve been trying to do that for a long period of time. Why now? Why is this different?
Erin Price-Wright (07:59)
It’s a really good question. I think it’s a conflagration of both geopolitical considerations and tech advances. So the Ukraine war has been a wake-up call for the United States. And if you go on the ground and you see the amount of low-cost, attritable mass being used to conduct a large-scale warfare, I think that would...
Matt Perault (08:25)
What does that mean?
Erin Price-Wright (08:28)
The United States develops weapon systems that are millions and tens of millions and sometimes hundreds of millions of dollars. We are really good at developing these beautiful exquisite systems, you know, where we have this exquisite supply chain of mom and pop shops who have been making this specialty part for 50 years. It’s perfectly engineered. It goes into this, you know, beautiful missile or plane or system.
And we have a handful of them and they are very good at the specific things they do. Where we’ve been, like what we’ve been confronted with in Ukraine is, you know, a $500 drone or maybe a $40,000 drone, let’s say. And this is a system that is manufactured quickly with cheap parts. The design is changing rapidly based on feedback from the field.
You know, can, you can, it doesn’t matter if you never get it back. Like if you blow up the drone, who cares? Because it’s a cheap drone and you’ll just build a new one. And we just, I think slowly have realized that we actually don’t have the manufacturing base to support that type of warfare. And I say like, you don’t have a defense industrial base if you don’t have an industrial base. If you’re not building things, you can’t build a lot of military things. You’re just like, you can’t build full stop. You can’t just spin up a defense base without having this know-how and process knowledge in the economy.
And so I think that it’s been a real wake-up call. And it’s one of the stories that I’m sure people have heard before that came out a few years ago, Ukraine was buying their drones from China. So they were buying DJI drones, and obviously China and Russia are very aligned. So China was giving Russia this back door to be able to figure out where the drone controllers were located who were operating DJI drones. And so Ukrainian soldiers were getting killed because they were using Chinese drones.
And so I think that it was a big wake-up call for Ukraine to realize, wow, we actually have to be able to produce this stuff. We need to be able to build things at scale to be able to fight this conflict. And I think watching that from the United States and not even being in a strong position to be able to help and support because we don’t have the kind of attritable mass or the know-how for how to build attritable mass that they need was like a big wake-up call for us.
And that combined with the rising geopolitical tension around China and the realization that if you pull our supply chains back for essentially everything we use in the United States, all roads lead to China. So the idea of building a fully US manufactured drone with US manufactured PCB boards and controllers, US manufactured brushless motors, US manufactured cameras and optical systems, US manufactured sensors, or not even US manufactured, but just allied manufactured, the rare earth magnets, all of these components, once you start going up the supply chain, are all coming from China. So there’s this big moment and large appetite to figure out how to bring more of that back and reshore in the U.S. And frankly, there’s no universe where we can do that without heavily relying on AI.
Matt Perault (11:44)
So that makes a ton of sense in terms of the sense of urgency. It sounds like what you’re suggesting is like a really massive shift in how we produce and what we produce and how the government procures. Are you optimistic about the pace of that shift as well? Like what are you seeing in terms of how responsive the economy and the government is to this sense of need?
Erin Price-Wright (12:07)
Well, on the government side, I think there’s been a ton of progress. We’ve seen a lot of reform in how procurement works. There’s extreme appetite, very bipartisan appetite to figure out how to modernize and reinvigorate the industrial base and the defense industrial base. So I think things are moving. Obviously, there’s probably more we can do and I’m certain that we could move faster, but I’m actually optimistic about the way things are going there. And on the technology side, we have a huge technological advantage with how strong we are in AI. And I think that’s gonna be really critical to figuring out how to actually make this happen because economically, financially, there’s just no world where we can compete on the labor economics of a China.
So we’re gonna need to figure out how to do all this using technology and software, and the United States is very good at software and very good at AI. So it’s pretty critical that we figure out how to merge those two worlds together. So I am optimistic. There’s a lot of work to do. It’s not gonna be an easy problem to solve, but I’m heartened by the fact that so many founders and technologists are focused on this problem, which was like the unsexy stepchild for several decades. Like it was just very much not cool to go into these fields like physical manufacturing, even things like electrical engineering and mechanical engineering. It was just not cool to go work as a mechanical engineer. So I’m very heartened by the fact that American dynamism is sexy. I think it’s a really good sign.
Matt Perault (13:44)
So speaking of unsexy, I think some people would say there’s nothing less sexy than procurement reform.
Erin Price-Wright (13:51)
I disagree, disagree. I think it’s very sexy.
Matt Perault (13:53)
For us, it’s critical. And I think as I’ve worked more and more with our portfolio, I’ve really understood how much selling to the government is such a critical market for them. It’s critical beyond just the revenue. It’s important to build those relationships with the government and the policymakers. So when you talk about reform that is important for Little Tech, what are some of the main components that we’re really advocating for?
Erin Price-Wright (14:10)
Yes, I’m specifically talking about procurement within the Department of War. And like broadly speaking, we have to figure out how to level the playing field between the large primes who have these existing books of business with these very large programs of record and a very set way of how they’ve historically worked with the government, which is cost plus models where they get a very, very specific list of not requirements even, but features and specs for how a particular thing has to function. And then they go off and they engineer to those set of requirements and build this exquisite system with its completely own isolated supply chain, that usually comes in 10 years behind schedule and you know, however many tens of millions of dollars over budget or billions.
Matt Perault (15:12)
I don’t have any background in procurement reform, but I’ve learned a little bit about it. So cost plus wasn’t really intuitive to me before I started looking into it, but it incentivizes increasing costs.
Erin Price-Wright (15:24)
It incentivizes being slow and being inefficient because you essentially take the cost that it took you to build something and like you add a margin. And so what we’ve been really fighting for alongside many others in the kind of defense landscape is reforming that and making and incentivizing the US government to buy products, commercial products off the shelf that are ready, that fulfill the same requirements without actually going through that design build phase. So you actually incentivize companies to build things in a productized way for less money and faster. And anything that they charge on top of that is margin to the business. So you actually incentivize companies to build products that are cheaper for the government.
Matt Perault (16:05)
Which seems like a yes yes yes kind of pitch. Is that what you’re hearing? And like when you hear no, what is the argument in response?
Erin Price-Wright (16:14)
You know, it’s a really big ship to turn. So I think there has been broad alignment by folks in the Department of War and by lawmakers that procurement has to change. There’s been broad alignment around this for, I’d say like well over a decade, since actually probably several decades. But it’s a really complicated machine to fix and it’s really hard to change those incentives. This is complicated legislation. So we have seen some great wins in the second half of last year on NDAA reform. There is strong cultural alignment within the Department of War at the top levels of leadership to figure out how to take this change in legislation and push it into how they reform policy. It’s gonna be a long road, but we’re definitely on it.
And it’s a very exciting moment because it’s also a time when we have all of these mission oriented founders who want to build things for the national interest, who have grown up at companies like SpaceX and Andrel and Palantir who have really helped pave the way for a lot of this reform and have incredible ideas about products that they could build that would serve the war fighter quickly with the latest and greatest in autonomy technology et cetera, that they could get to market really fast. And it finally feels like we’re starting to see a willing buyer on the other side of that who knows how to work with startups and is interested in, you know, bringing new tech into our armed forces.
Matt Perault (17:54)
So you’re in touch with these companies every day. And then a few weeks ago we had our American Dynamism Summit which is focused on trying to increase this connective tissue and create opportunities for our portfolio to sell the government. But you also said it’s a slow moving ship. Are you starting to see the space for opportunity expand.
Erin Price-Wright (18:13)
For sure, and I think you can just even see this based on… if you look at the sort of generations in our portfolio. I’ll start with Palantir as an example. So Palantir, I might get these specific numbers roughly wrong, but Palantir took over a decade to get its first $100 million program of record contract with the US military. Anduril, several of the founders from Anduril come out of Palantir, took about four years to do the same thing. And Saronic, which is founded by alumni from Anduril, took less than three years to reach that same milestone. So we’re seeing the timelines for getting real contract dollars and real programs compressed dramatically over the last decade, which is really exciting. And then what’s interesting that we’ve seen even just over the last several weeks as the conflict with Iran has intensified is a lot of even our earlier stage companies who were in dialogues with their buyers within the Department of War, they’re hearing their POCs come back to them and say, okay, I know we talked about this list of requirements, how much faster could you move if we cut that list by 75%? Like, how long would it take you to build 1,000? How long would it take you to build 100,000? So that level of urgency and intensity feels like it’s only increasing and will likely only increase for the foreseeable future as I think the geopolitical situation globally intensifies.
Matt Perault (19:55)
So as you’ve talked about the geopolitical situation and how it creates a sense of urgency, how it’s intensifying, you alluded a little bit to the changes that AI is bringing on the company side in terms of shifting the technology to respond to the moment. What are some examples of what you’re seeing there? How is AI really positioning companies differently to meet the moment?
Erin Price-Wright (20:16)
I think AI allows you to bring a sort of a factory mindset to really complicated problems. If you can even think about the need to build a factory of factories, like suddenly we’re in a position where it’s not like we have a decade to sort of build and plan one exquisite factory.
We’re talking about needing to build hundreds of factories, which, you know, that starts to sound like a factory itself. And AI is really uniquely positioned to be able to take a lot of the complexity and the sort of complex systems thinking that goes into how do you build a factory and really simplify it. So if you think about what a factory is, it’s, you take a complex system and you break it down into its component parts. And then within those component parts, you automate and then you iterate.
And that kind of deconstructing something down into its components and automating, that’s something AI is really good at. So we’re seeing companies both within the defense space and outside of the defense space, or maybe adjacent to the defense space, take a lot of lessons from AI and to like how they’re building and planning factories. So, you know, I’ll give you a couple of examples.
The process of site selection is kind of an art, like the sort of real estate deal making, talking with local politicians and how do you decide where you set up your factory is like a very complicated problem. And historically, you might hire a consultancy and you’d spend, you know, millions of dollars with them.
And they would go do this whole survey and they would engage all these political leaders and they would look at things like land availability and power availability and environmental regulation and all these other kinds of things and then come back with a proposal several years later. With AI, you can do things like that extremely quickly. So you can do like large scale surveys of power availability, land availability, regulatory regime and framework, permitting regimes in place and pretty quickly hone in on a very small number of sites that you can immediately engage local lawmakers to say like, what would it look like to build, you know, an energetics factory here? Or would you be interested in us building out a drone factory here?
That takes a multi-year and many millions of dollar process and lets you do that in a month, which is a pretty powerful tech advantage. Or, you know, how do you even go, how do you go through the process of permitting. Most regulations, local regulations are there for a reason. They’re probably reasonable and good regulations. Like we want them to exist. It’s important that regulations exist. But the process of going back and forth with permitting bodies and regulators as someone who wants to build a factory can be a true nightmare.
I look at one of our companies, Radiant, and they’re a nuclear company, so they’re going back and forth with various nuclear regulators. And every time they have to submit their paperwork, it’s like tens of thousands of pages. And like, if you have a comma in the wrong place, you know, it might just get kicked back to you. So part of that is an alignment problem. How do you make sure you’re on the same page with your regulators where everybody is incentivized to figure out how to make this work versus figure out how to say no?
And part of it is, you know, it’s a problem that’s really well solved with LLMs. Like paperwork is a great LLM problem. So all regulators should have AI tools and all companies that are dealing with permits and regulators should also have AI tools. Yeah, I mean, I could go on and on and on, but it’s just like every single tiny step of like, how do you build a factory is a problem that AI is actually really well suited to help speed up. So then you start thinking about, okay, like if it historically has taken five to 10 years to permit, build and commission a factory. Like in the new world of AI, that should take 12 to 18 months. And that’s what we’re seeing with our companies. They’re able to build a lot faster.
Matt Perault (24:39)
So you’ve talked to me a couple of times about a factory mindset. And when you say factory mindset and AI, keep thinking about the next frontier, which I think we think will be robotics. Are you starting to see exciting companies emerge in the robotics industry? And are they part of what you see as this new factory mindset?
Erin Price-Wright (24:55)
For sure. Most factories have lots and lots of robots. If you go into a factory today, the likelihood that you’ll see some sort of a smart conveyor belt system or like a robotic arm is very, very high. What’s exciting about the next frontier of robotics is how do you actually lower that cost of automation such that you can adopt automation in more parts of your assembly line or your production process. So if you think about a production process, like a true factory line is a series of modules. Like this was the Henry Ford realization. Like how do we deconstruct building a car into a hundred steps and every step is its own module in an assembly line. And then we’re gonna put one person on every single step and they’re gonna do that same step over and over again. So they get really good at that task, and then the next person gets really good at the next task. So that’s what a factory is, that’s what an assembly line is. That was like the big realization in the 1920s that Ford had.
And so now when you go to a factory, it’s still the same thing. You see a series of production steps that are kind of broken out by task. And then some of those tasks, the very expensive tasks, are usually already automated. So you’ll probably have, you know, for some very complicated tasks, you’re gonna have a robotic arm, you know, putting a part somewhere, screwing something in, doing some task, and there’s probably like an assembly line moving something to the next, or like a conveyor belt moving something to the next stage. So there’s already a lot of automation. The challenge is like, if you look at the cost of installing that arm, it’s really, really high. So the hardware itself, it’s like, you know, a hundred grand plus to buy one of those arms, one of those industrial arms. And then you have to actually program it and have to hire a system integrator. It’s like this really outdated control system that you have to very precisely program exactly the movement. And there’s usually reasonably low tolerance, you know, for anything different. So it knows exactly what it’s going to do. You program it to do exactly that and all in all to automate a step on a manufacturing line is probably close to $200,000. And so then that’s something you should think about annually. So then if you think about, or every time your production line changes. So if you think about the cost benefit analysis to automating a step on a factory line, it has to be a pretty expensive task in order to justify investing $200,000 a year to automate it. Because you think about the salary of the person that would do that task and you have to kind of offset that.
So the exciting thing about AI and robotics and where we’re going is it really dramatically lowers the cost of automating a module. So you get lower cost, less precise hardware that has higher fault tolerance because instead of these extremely precise control systems. You are using robotic learning, you’re using vision models, you’re using other types of modern AI concepts to be able to learn how to do a task so that the cost of setting up isn’t $200,000. So maybe now today, the cost of automating a module on a factory line is $100,000. And sure, you need to hire some very strong techs to take care of your robotic fleet but you’re also changing the cost benefit analysis of automation on a factory line. And so I’m not talking about replacing workers who are on existing factory lines. That’s usually not how manufacturers are thinking about this. They’re usually thinking, we’re thinking about it in the sense of like, what things could we not build in the U.S. before that we could build now? So where are things where we actually can make a reasonable cost trade-off?
If a factory that historically would have required us to hire 500 people to staff, now we can hire 250 people to staff. And 100 of those jobs might be much higher paying jobs, because we’re talking about robotic techs, quality techs, like robotic programmers and robotic engineers, instead of like someone just screwing a bolt in. But overall, now suddenly something makes economic sense that didn’t make economic sense before. So that’s how I would think about autonomy and robotics like on factory floors and the real difference that it’s making.
Matt Perault (29:38)
You’re presenting a really clear vision for this future world that we might have. It sounds like there are lots of reasons that we’re like, that you’re explaining that we’re taking steps…
Erin Price-Wright (29:48)
It’s not the future world that we might have, like it’s happening now.
Matt Perault (29:52)
A world that is happening now, but the transition to it, does seem like, as you described, it’s like a slow moving ship. And so I assume there are some tech barriers, there are some economic barriers, there are also policy barriers to fully realizing the world that you’re articulating. When you talk to the companies that you work with, what are they saying in terms of policy barriers or also policy accelerators? Like, what are they seeing in the policy environment that is either making it harder for them to achieve this vision or making it easier for them?
Erin Price-Wright (30:22)
Yeah, I mean, it’s full stop, fundamentally too hard and slow to build a new factory in the U.S. So that’s one thing. It’s certainly getting better and timelines are compressing, but as it stands today, when you start looking at the kind of power availability and land availability it just takes a long time. How do you train talent? How do you find the worker base that’s actually gonna do this? How do you ramp up a factory?
How do you, the lead times for machines, if you think about a lot of the machines on the factory floor are, you know, we don’t manufacture those machines in the United States, so maybe some of them are manufactured in China, but a lot are manufactured in places like Korea and Taiwan and Germany, and that’s the importation of those machines is still a very kind of politically complicated. There’s still a lot of uncertainty and understanding how to actually bring those machines as quickly as possible to the U.S., how to get trained on them. And then if you look at the robotics and automation piece in particular, the tricky thing now is that most of the robotic components that you might use to bring autonomy to a factory line, are all made in China. Like 99.9 % of the robotic supply chain is China.
So, there is no company in the U.S. right now that’s making actuators at any kind of scale. And those are the kind of specialized motors that allow robotic arms to move. Maybe you could buy one from Germany for like 10 times the cost as you could buy one from China. And all of those components that go inside robotics are all fully Chinese supply chains. So we have to figure out how to jumpstart our US-based robotic supply chain.
That’s really important and we need industrial policy to do that. On the other hand, if we do it too quickly, like if suddenly tomorrow we said, you can’t buy any robotics components from China, like that would kill our robotics industry before it even really got started. So it’s sort of that, I’m not a policy expert. This is where like I need you to help define what these things should be, but you know, figuring out how do we start this ecosystem in the U.S. without killing it before it even happens…is something that we think about a lot.
Matt Perault (32:49)
Erin, what about AI and manufacturing specifically?
Erin Price-Wright (32:53)
Yeah, I mean, another example of where we’re seeing AI make a really big impact in manufacturing is actually in this sort of design for the manufacturing process. So China, for instance, has this benefit of having these very tight-knit ecosystems where in Shenzhen, you’re a design engineer, you sit a block away from the manufacturing engineer, you just go try stuff out. And because your friend tells you this is hard to build, the next time you design something, you don’t design it that way.
We have to figure out how to abstract that knowledge in a more, in like our American way to build like what is our American version of Shenzhen. I would posit that the American version of Shenzhen uses software and AI to accomplish a lot of what the like physical ecosystem of co-location in China has given them. So like, how do we use AI to design better things that are easier and cheaper to manufacture?
And AI is really good at solving anything that looks like a coding problem. You look at tools like Cursor and Claude, and, you know, these are some of the early success cases of AI where we’re seeing massive economic value be generated. But what’s cool is like any time you can represent something as structured logic with sort of limitations that you’re trying to, and constraints that you’re trying to optimize around, you can represent that as code essentially and AI is really good at it. So AI is very, very good at design problems. AI is very good at scientific design problems. AI can design chips better than humans can design chips. And AI can design manufactured goods better than humans can at this point. So that’s one of the use cases that we’re especially excited about.
And whether we’re talking about chips or PCB boards, or entire components or entire factories, like AI actually just being able to design a factory or a data center from the ground up. That’s like a super exciting use case where AI is actually producing designs that are better and easier to manufacture and cheaper to manufacture and faster to manufacture than people are.
Matt Perault (35:05)
There are a lot of reasons why I was excited about this conversation. One of the main ones is you bring so much expertise across so many different subject matter areas where there’s some kind of, there’s some set of assumptions in the policy conversation. And I think the business reality is potentially really different. Maybe the main one that I’m thinking of right now is the conversation about data centers. In the policy conversation now, data centers are increasingly becoming toxic for a variety of reasons. I think there’s a perception that the benefits of data centers accrue to the most wealthy companies, the most wealthy individuals, their concerns about their environmental impacts, their concerns about their impacts on communities, noise, et cetera. There are concerns about the amount of jobs they produce or don’t. So I’m curious about your sense of data centers from the vantage point that you have where you’re really tracking companies that are involved in energy generation.
Erin Price-Wright (36:05)
I love this question. I love data centers. It’s probably not a surprise. I think that as a society, we are so lucky that suddenly there is all this capital and attention and political will on behalf of tech companies to figure out how to build data centers because they represent something that I think were the lessons that we learn and the process knowledge that we learn will translate to economic growth that extends so far beyond AI, like it’s ridiculous. Data centers are attracting a lot of capital, so there’s a ton of investment going into them. And this is investment going into teaching us how to build mega projects again in the U.S., which is a really good thing.
We haven’t built, we haven’t had any large scale growth in this country, physical growth in a really long time. And so, you know, our wheels are very creaky. So when we’re talking about reindustrialization, building of data centers in the U.S. is such a boon to reindustrialization in the U.S. because we’re going to be teaching contractors and EPCs how to build like high tech sites. And sure, like maybe those jobs specifically on data centers are short-lived, but I guarantee you those regions that attract data centers and build data centers in their regions, they’re going to attract the next set of factories that are part of the re-industrialization movement in the U.S., because they’re going to have the workers who are trained and skilled in building these kind of high-tech centers and sites that are gonna make companies wanna build there. And they’re going to have the kind of regimes in place to be able to do that.
I think the other thing that’s really important about data centers is they’re putting into focus a problem that’s been able to live sort of under the surface for many decades, which is our aging electric grid. And they’re going to mean that we have kind of the capital and the will and the desire to figure out how to modernize something that we need to modernize anyway. Like our grid is an incredibly complex machine that hasn’t really grown in several decades. And we’ve been able to get away with it for a long time because sort of efficiency gains in things like our refrigerators and industrial motors and in other processes have sort of offset any sort of electric load growth that we’ve seen on the grid over the last 30 years. And that’s over. And that’s over not because of data centers, that’s just over because we’re seeing for the first time, AI aside, a large-scale growth in electricity demand in the United States for the first time in 30 years. And we’re going to need to solve that problem whether or not we build data centers. And the nice thing about building data centers is it puts a lot of capital behind solving that problem. Like figuring out how we improve delivery of electricity on the grid is something that I for one would really like Google to help pay for so that the consumer doesn’t have to bear that cost alone.
So if we can figure out how to do that, if we can add transmission to the grid in the process of building data centers and then use that additional capacity to really drive reindustrialization of the U.S., I think that that’s an incredible outcome. And I completely understand the controversy.
I get it, like, people are worried about jobs and they don’t hear a positive narrative for AI and all they hear is that it’s going to increase their electric base. I think we really need to work hard to counteract that narrative with the general public because I think it’s not actually true.
Matt Perault (39:46)
Yeah, it sounds like you’re articulating a case for why the benefits of data centers outweigh the costs. But I would love to get your sense on the specifics of the allegations about costs. Will people see their electricity rates rise? And what about water usage? And what about noise and all of those kinds of issues?
Erin Price-Wright (40:04)
I’ll maybe focus on electricity in particular, just to go deep there. And I’d say there is a similar argument if you go down the line for a lot of the arguments against data centers. You know, once they’re on the grid and connected to the grid, it’s much cheaper for the utility to serve a data center customer than like you as a consumer in your home. Cause you think about the highest cost of power to serve is the low utilization user. And like you in your house, you turn on a light switch, you run your dishwasher, on average, you’re using like five to 10% of your peak power. And then suddenly maybe like in the afternoon, you might use 25% of your peak power. And then suddenly in the afternoon on a hot day, you’re using a hundred percent. So your average utilization as a consumer is extremely low, which means the cost to serve you by utility is very high because they have to essentially plan for your peaks, but you’re very rarely using that.
On the other hand, a data center customer is very high utilization. So data centers are like something like 80, 90 % utilization. They’re using a lot of power all the time, super predictable. They’re very, very cheap for a utility to serve. So when you add a lot of data center loads to the grid, assuming, you know, they help bring some of that generation online and help pay for that generation online, which all of the hyperscalers have indicated that they’re extremely willing to do and in fact are already doing. They’re bringing more cheap power onto the grid, which lowers the cost for the average consumer. So in an ideal world, long-term, data center loads plus other heavy high utilization industrial loads are really, really good for the grid and really good for the consumer. They will lower the cost of electricity for every single consumer on the grid, which feels a little bit counterintuitive.
And so my big worry for this moment that we’re in is that there’s gonna be such a backlash against AI and data centers that all of these data centers are gonna go build their own grids. They’re going to be completely off the public grid. They’re gonna build their own power generation. They’re gonna build their own storage and backup. And they’re gonna give none of that to the grid itself, because everybody rejects them and makes it too hard for them to do it. And then we actually, as consumers, get none of the benefit of that investment into lowering our cost of power. So that’s like one of my big worries with the current moment in AI.
I’m optimistic that we’ll solve it because I’m an optimistic person, but I think that that would probably be the worst outcome that we could get to for the cost of power for the American consumer and also sort of the ability for us to support a broad scale U.S. reindustrialization, which will be energy intensive kind of after this wave.
Matt Perault (42:50)
There are a lot of people who are concerned about what AI is going to mean for jobs. Erin, how do you see that issue?
Erin Price-Wright (42:55)
I fundamentally see AI as a net job creator, especially when it comes to re-industrialization and non-white-collar work. We’re talking about bringing factory jobs back to the U.S. that haven’t existed here in decades. And I’m not talking about factory jobs where you have someone lifting heavy boxes or lifting heavy parts and throwing out their back or twisting one single screw into a socket over and over again all day long and like destroying their wrist. Like those aren’t the type of manufacturing jobs I’m excited about. Those are gonna be done by robots, thank goodness. No I’m talking about high skilled, high tech jobs, doing quality engineering at scale, doing process engineering at scale, robotic technicians, I’m talking about electricians who are working on really complicated systems level problems on our factories. These don’t require college degrees, but as we build out more industrial capacity in the U.S. I think we’re going to see an increasing number of this type of high skilled blue collar work come back to the US.
Matt Perault (44:12)
Thanks so much.
