America Is 10x Behind China in AI Infrastructure — The CEO Building the Solution
Motley Fool Hidden Gems Investing
This episode of 'Motley Fool Hidden Gems Investing' explores the critical bottleneck in the AI infrastructure race: energy supply. CEO Hanan
Key takeaways
- The primary bottleneck in AI development is not software or chips but physical energy infrastructure.
- Data centers now require gigawatt-scale power—equivalent to powering a million households—and traditional grid capacity cannot keep up.
Main topics
- AI infrastructure power requirements
- U.S. vs China grid capacity gap
Notable quotes
"China adds 540 gigawatts of power to the grid per year, so more than 10 times our capacity."
"A gigawatt data center essentially is consuming the equivalent of 1 million US households worth of energy."
Conclusion
The future of AI depends not on algorithms or chips alone but on scalable, sustainable power
Transcript preview
Speaker 1 (0:02) China adds 540 gigawatts of power to the grid per year, so more than 10 times our capacity. So that means that China, if we consider China to be our rival in this AI dominance race, has 10 to 11 times more capability and capacity than we do to build infrastructure and bring that online. Speaker 5 (0:31) That was Hanan Happy, co-founder and CEO of Exalot, on why the AI race is less about algorithms and more about electricity and why the U.S. is dramatically behind where it needs to be. I'm Motley Fool analyst Rachel Warren. The conversation about AI almost always focuses on chips and models, but Hanan argues that the real bottleneck is something far more physical. He joined me to discuss why a single year of grid delays could cost a hyperscaler $12 billion in missed revenue, why communities across the country are pushing back hard on data center development, and what investors watching the AI build-out should actually be tracking as the capital flows in. We hope you enjoy. Speaker 5 (1:09) Hello, everyone, and welcome back to Motley Fool Conversations. I'm Motley Fool analyst Rachel Warren. Today, I'm joined by Hanan Happy, the co-founder and CEO of Exawatt. Hanan brings an incredible deep tech background to the table. He studied mechanical engineering at the Technical University of Munich, later attended the Stanford Graduate School of Business. And prior to launching Exawatt, he held key engineering and leadership roles at industrial and tech powerhouses like General Electric, Accenture, Tesla. He also co-founded Valanci, an autonomous hardware and logistics company. But now he's tackling one of the biggest challenges facing the tech sector right now. His company, Exawatt, which has captured the backing of elite investors like Sam Altman, Andreessen Horowitz, is purpose-built to power AI, delivering dedicated renewable energy to the engines of modern intelligence. Exawatt calls its new energy generation category on-site firm solar. It's designed to bypass the years-long utility grid delays that are stalling the modern AI build-out. Hanan, welcome to the show. Speaker 1 (2:07) Thanks for having me, Rachel. Speaker 5 (2:09) Absolutely. You know, the last few years when I think a lot of investors talk about AI, the conversation is very much focused on, you know, semiconductors, networking capabilities. But you've argued that the next massive phase of AI investment isn't actually happening in software. It's happening in that physical infrastructure. So walk us through that. Why has the biggest bottleneck in the AI boom suddenly shifted from the silicon to the raw physical infrastructure? Speaker 1 (2:34) Yeah, absolutely. So if you think about just a couple of years ago, the largest data center that we had in the United States was about 100 megawatts in capacity. And that was considered very large. And, you know, the rack density was, you know, in the tens of kilowatts. And a building block of a data center was maybe 10 to 20 megawatts. And fast forward just in the last couple of years, the building blocks for data centers have scaled up to in the order of 300 to 500 to even 700 megawatts, the building block. So you could argue the building block of a data center is now five times larger than the largest data center we had in the country a couple of years ago. And the data centers themselves are now on average about a gigawatt and there are some that are in the order of 10 gigawatts. And to put that into context, a gigawatt data center essentially is consuming the equivalent of 1 million US households worth of energy. So you're basically saying I am building a city from scratch for a million people. And I'm trying to do that as fast as possible because I want to stay in the AI race. I want to have the best model, you know, I want to be competitive. And this is where, you know, we're past the idea of like, how do I, you know, write the best LLM algorithms or models? How can we overcome the chip shortage? Because the chips have become better too and more powerful, even if they have become more energy efficient, but they're still consuming a lot more energy as a total. And now we have to build power infrastructure. And as a country, we haven't had to build... massive amounts of power infrastructure for the last couple of decades. We haven't had much of a load growth in the US and now we do. And data centers are eventually going to contribute or be taking about 9 % to 10 % of total. energy produced in the US, the grid capacity. So we are in a mad rush to build power infrastructure. Building power infrastructure is not trivial. It's not like software that you can write code and it just scales infinitely. You have to move metal, you have to move concrete, you have to move earth, you have to do all sorts of permitting and construction and move workers. And we just don't have those capabilities at scale in the country. And what we're trying to do at Exawad is come up with a formula that allows us to do something in a factory setting with tight control over cost, tight control over the speed of execution, and try to scale that as fast as possible to address this power gap. But there's a massive, massive power gap, and I think everyone's scrambling to figure out how to cover it. And yeah, that's kind of the reality that we're facing today. Speaker 5 (5:23) Well, I think... As Speaker 1 (5:25) you noted, it's become increasingly obvious that Speaker 5 (5:27) the primary constraint is about delivering reliable power where AI infrastructure is actually being built. So what are the solutions to that look like? I mean, obviously, I know that's what ExoWatt's all about, but I'd love to hear your thoughts as well broadly on how you're able to get past this bottleneck in the next 5, 10, 15 years if you're building out all these data centers as a tech company. Speaker 1 (5:50) Yeah, good question. Again, going back to the history of data center build-out, three, four years ago, when you built a data center, again, whether it was 25 megawatts or even 100 megawatts, you really didn't think of power as a constraint. You basically put the data center where the location was favorable from a real estate perspective, maybe from a fiber perspective, typically close to urban and suburban areas. And power was not really necessarily a key factor. plugged it into the grid and the grid was able to give you power. But fast forward today where we have these massive data centers, the grid has no capacity. They have to come up and scramble with solutions to build their own power. And, you know, this started out with the mad rush to go build gas, right? So over the last couple of years, especially the last two years, Every data center that has been announced is saying, well, I'm going to build some version of gas generation behind the meter gas, and eventually there'll be some grid. And gas generation has its own challenges. We have, yes, abundant amounts of gas in the country, but we have the lack of supply chain for turning that gas into electricity. So turbines have been backlogged. five to seven years in some cases. We have also now fuel volatility, right? Given the geopolitical things that are happening around the world, gas prices are also fluctuating. And then just by the simple fact of having a gas line doesn't mean you actually have access to that gas to turn it into electricity. And so a lot of people started saying like, oh, I have a gas pipeline here, I'll build a data center. It's not that simple.