EP 381: AI’s Energy Crisis - Can Quantum Save the Day?

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life. I think by this point, there's no denying the power that generative AI has for everything, for growing your business, growing your career. But speaking of power, that's an often overlooked aspect of generative AI because it is resource heavy. It needs so much energy to run. Actually, some. Some recent studies say that data center energy demand could double in just two years. So what do we do, right? All of these big companies, they're scrambling to figure out this energy problem, but maybe it's quantum, right? Could quantum computers, could quantum energy save the day when it comes to AI's energy crisis? All right, I'm excited to talk about that today on Everyday AI. What's going on, y'?

All?

My name is Jordan Wilson. I'm the host of Everyday AI and this is your daily live stream podcast and free daily newsletter helping us all learn and leverage what's going on in the world of generative AI so we can grow our companies and career. So if that sounds like you, you are 100 in the right place. If you have not already, make sure to go to your everydayai.com and sign up for our free daily newsletter. In there, we keep you updated with everything that you need to know in the world of AI, as well as we break down the insights each and every day from our podcast and from very knowledgeable guests like we have today. So before we dive in, let's start as we do every single day, by going over the AI news. All right, so first, let's talk about this. Perplexity has is in the process of unveiling a new financial analyst tool for its users. So Perplexity has just introduced a preview of its upcoming tool, Perplexity for Finance, which will allow users to search for detailed financial information about companies. The platform promises features such as real time stock quotes and historical earning reports and comparisons with industry peers, all presented through an engaging user interface. So this move comes amidst challenges for Perplexity, including allegations of plagiarism and copyright issues, which the company is addressing by launching a new revenue sharing model with news publishers. All right, next, Lenovo has partnered with Meta to unveil an ambitious new piece of AI hardware at Tech World 2024. So, Lenovo has just introduced Lenovo AI Now, a local AI agent designed to transform traditional PCs into personal assistants using a large language model based on Metas Llama 3.1, which enhances privacy and allows for real time interaction without relying on cloud processing. So Lenovo AI now will offer a natural language interface enabling users to easily adjust settings like like screen brightness or activate productivity tools with simple voice commands, making it user friendly and more efficient. The platform supports a variety of AI applications catering to professionals, students and creatives by providing tools for document summarization, content creation and more, ensuring a smooth user experience with minimal battery consumption. So, hey, maybe we'll have something smarter than an Alexa that can't even tell me the time in less than 60 seconds. All right, in our last piece of AI news kind of relevant for today's conversation, the Biden administration is considering capping AI chip exports to certain countries for national security reasons. So the Biden administration is exploring the possibility of imposing country specific caps on the sale of advanced AI chips from major American companies such as such as Nvidia as a measure to safeguard national security. So the discussions are still in the early stages, but the idea has gained momentum recently, building on new regulations that simplify the licensing process for AI chips chip exports to countries like the uae, the United Arab Emirates and Saudi Arabia. So the Commerce Department has already restricted AI chip shipments to over 40 countries, primarily in the Middle East, Africa and Asia, due to fears that these products could end up in China. So the National Security Council is considering how many nations might use advanced AI capabilities, particularly in relation to internal surveillance and the potential risk to U.S. intelligence. So after that news, Nvidia stock actually fell from an all time high and plunged 4.2% on news on this potential kind of scale back. All right, we're going to have a lot more on those stories and everything else you need to know in our newsletter. But you didn't tune in today to learn about AI news. You tune in to really talk about this AI energy crisis. And I'm excited to have an industry veteran on the show today, someone with many decades of experience. So I'm excited to welcome on the show. There we have him, Peter Chapman, the President and CEO of IonQ. Peter, thank you so much for joining the Everyday AI Show.

Oh, thanks Jordan, for having me today.

Hey, and, and very early, right? This is a live stream, unedited, unscripted. So joining us from the West Coast. So we should have some sort of award, Peter, that we give to people who wake up at 5am to do this.

But if you look behind me, the window is dark.

Yeah, yeah. If you're listening on the podcast, Peter's joining in. Pitch black. But Peter, maybe can you first explain a little Bit about what IONQ is.

Yep. So we're a manufacturer of quantum computers, a new kind of platform for doing calculations.

All right. And we could probably spend hours talking about quantum computers, and I would assume most of it would go over my head. But, Peter, maybe can you simplify? Right. What is, what are quantum computers? What is quantum power? What is this quantum word that seemingly very few people understand?

And maybe even as to why we need quantum computers and then we can get to what they are. Love it. If you go Back to the 1980s, Nobel physicist Richard Feynman, who was on the Manhattan Project, realized at the time he had access to early computers, I'm sure he was doing simulations back then, realized that these classical approaches to modeling Mother Nature would, even if we allowed Moore's law to continue for a million years, meaning we doubled their computational power every year, they still wouldn't be powerful enough to be able to solve some of these problems that mankind has. In particular, chemistry was the early one. And it's interesting now we're coming up to AI, which also is very power hungry. And so it's kind of that same analogy that he realized back then is, you know, even if you converted all the matter in the universe into transistors, it still wouldn't be enough. So he said, we need a different kind for these class of problems. We need a different kind of computer. And so his idea at that time was a quantum computer that uses quantum information to be able to model Mother Nature. And, you know, what is it? What is it we're trying to do with strong AI, model Mother Nature's human brain. Right. So, so it's a very similar kind of problem. So what is quantum underneath computing? We kind of first start with different representations of information. If you grew up in the 50s and 60s, we used analog, if you remember, from an oscilloscope, a sine wave, we used analog signals to do video and records. It was really good for that. For computers, we use discrete information, zeros and ones. The great thing about zeros and ones is that it was agnostic of noise. The problem with the analog approach was that a little bit of noise would come in. That would be the hiss that you would hear on a record or maybe the static that you would see on the screen. They were susceptible to noise, but digital, the great thing about them was, is between 0 and 5 volts and 5 volts was 1 and 0 was 0, but 4.9 volts is also 1. So you could have a little bit of error and it was still okay. So we Built different computing devices based on the signal type. The signal type in quantum is actually superposition. Sometimes we say that if we look at the representation, you know, analog would be a continuous value. Digital would be zeros and ones discrete. And if you look in quantum, it would be qubits. And sometimes we say it's 0, 1, or both or everything in between. That part's the thing that's really confusing because we don't have that in our natural world in the way that we experience it. Because down at the quantum world, down at the. At the size of atoms, if you remember from your high school, we have to go all the way back to your high school textbook in either chemistry or physics, is that you have a nucleus and you have an electron that's running around it. And the way we often think about that is the same as our planetary system. That the sun is the nucleus and planets are the electrons. But that's actually a very incorrect analogy for the atomic world. And actually the electrons are in all positions at the same time. And it's not until you measure them that suddenly they collapse down into a known position. So what you're getting is a series of probabilities that an electron is at a particular location. And so it's a really different world. So quantum computing is all about probabilities. It's being able to calculate probabilities. And what you do is you chain them together in massive parallel computations to be able to get out the most probable answer. I'll explain a little bit there. Again, the difference between classical and quantum. Microsoft gave a good explanation here. Said in the classical world, we do everything sequentially. We do them one at a time. If we wanted to solve the problem of finding the path through a maze, in classical computing, we would go down every little corridor in the maze and look at them one at a time. But in quantum computing, what we would do would look at all the paths at once in a single instruction and find the best path in a single instruction. It's that massive amount of parallelization that has the potential for quantum to be really interesting. And that's where the energy savings is. So instead of going through and looking at a trillion different possibilities, we're going to go through and look at all of them at the same time. The quantum computer and a classical computer are fundamentally different. Quantum will not replace classical computers. The analogy would be in your kitchen, you have a microwave, you have a stove, and you have an oven. They all do different things. And, you know, you use the microwave when you want to heat Something up that has water. It's really good at that, but you're probably not going to bake a cake in it. And so that same kind of analogy applies to quantum and classical as well.

I think that analogy is, is very helpful. Right. Not having to sequentially go through and check out all these mazes, but instead doing it all at once. Maybe. Could you, before we dive into Quantum's potential impact on AI, could you maybe explain a little bit more historically kind of this, you know, ongoing path toward quantum? Right, because it's not new. It's been going on for decades, and all the biggest companies in the world are, are investing time, money and resources into quantum computing. But maybe could you tell us a little bit about where we are today and kind of maybe what hurdles. Right. Still need to be cleared? I know it's, you know, again, that, that could maybe take hours to talk about, but quickly kind of bring us up to where we are today with Quantum and what that could change.

Yes. So we're at the, at the, you know, very early period. What I would say is kind of, if you will, the 1970s kind of, you know, compared to intel, where the quantum computers themselves are finally getting to the point where they're powerful enough to start to run applications that we will not be able to do classically. So we're just at that really interesting cusp. We're also at a place where it's so early that we're still finding applications. Famously, at intel, when they did the first microprocessor, they thought that the only thing that would be good for was a calculator. They didn't imagine a spreadsheet and a word processor and the Internet and all the rest of that. We're a little bit like that too, in the sense that we, we're building these processors. But my guess is 20 years, 40 years from now, people will sit down and be using Quantum in every. In lots of ways we couldn't even imagine today. And so, you know, more than likely, you know, somebody will come back to this podcast in the Wayback Machine and say, Peter just lacked imagination. But, you know, the, the things that we believe that these are going to be good at, we've already shown machine learning. Machine learning is an interesting. Where the model we can create is better than what you can do classically. In addition, with very limited data, you can produce a better model. That's a limitation sometimes that you have to have large amounts of data to even create the model. Those are two aspects of it. Certain math problems. It's and you know, quantum is a conundrum. It's not really good at entering one plus one, but it's probably pretty good at doing, solving linear equations and differential equations. And so how can both of those things be true at the same time? Chemistry, the natural world, that was really the original thing from Feynman was the problem was with chemistry is every time you add an electron is that electron wants to interact with every other electron. So it's an exponential problem, it's a power of two problem. When you go from 30 electrons to 31, you just increase the computational needs by two. And so the idea to be able to simulate Mother Nature in terms of chemistry is a. Classically is just very, very limited. And so, you know, what would you like to, what material would you like to create for new batteries or, you know, transparent aluminum from Star Trek? Well, you're probably going to need a quantum computer to figure that out.

Yeah, and, and Peter, I want to maybe just frame this, this a little bit now about where we are today with, with generative AI and how quantum can, can factor into this equation. Because we've seen over the last, you know, two years now, kind of since this quote unquote, you know, chat GPT moment of generative AI. Even though arguments can be made, you know, it was before that, but now all of a sudden there's seemingly unprecedented demand on. We need more energy, you know, to power these data centers, to, you know, train all of these models. It takes, you know, a crazy amount of power, you know, all of these GPUs. Right. How can know quantum help solve that, that energy demand? Because it seems like, you know, one thing we can be certain about right now is the demand for generative AI is not going anywhere. And generative AI is much more energy hungry than traditional computers. So how does quantum play into that equation?

You're 100% right. I believe Ireland has said now that if you want to build a new data center there, you first have to build a power plant. So they're just not going to allow you to build more data centers. So it's a. So where does quantum come in? So it's this massive parallelization and very efficient. So if you look at our next generation chip, which has 64 qubits on it, 64 qubits. To simulate that using GPUs, you would need two and a half billion GPUs. So it just kind of says that there's, you know, you in our, that's it's a single chip and it would plug into Two standard wall sockets. So, you know, what is the energy requirement for two and a half billion GPUs? I don't know, but it's obviously huge. And so using the right tool for the right problem, obviously you can have huge savings. And so it's that comparison in energy use where the question is, Instead of using GPUs, can we use a quantum computer and have that same kind of savings? By the way, at 64, every time you add a qubit, it doubles its computational power. Okay, so at 64 qubits, it can look at a computational space of 18 quintillion. Just to put this in, that's a number that everyone doesn't use on.

Yeah, I'm like, I don't even know that number.

Yeah, it's a huge number. So to put it in context, at Frontier, the world's largest supercomputer at Oak Ridge, it can do 1.2 quintillion floating point operations per second. That's 1.2 quintillion per second. This is 18 quintillion in a fraction of a second. The next chip after that that we're building is 256 qubits. So that's two to the 256. So now we're getting into numbers where mankind doesn't have names for them. But I'll just to show you the computational power at 120, we're going to go to 256. But 120, the computational states it can consider is equal to the number of atoms in the known universe, all 14 billion light years across. And but we're going to 256, so that's 2 to the 140, more than 146, more than the number of atoms. So it just kind of shows even if you were to convert all the matter in the universe into transistors, it would have tough time keep competing with just one quantum processor. And so now the question is, how can we apply that to this, to AI? And obviously LLMs is one aspect of it, and then maybe a completely different approach to strong AI that doesn't use an LLM.

Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Genai. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands or just need help building your front end AI strategy, you can partner with us too. Just like some of the biggest companies in the world do. Go to your everyday AI.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on gen. So let's talk about that. So, you know, it seems like right now there is this huge focus on using GPUs, right? For large language models, right? We need a lot of them to train the next model for inference to make these models better that we all use every single day. Right now there's this, this kind of, this shift toward models that can reason, right? Like open, open AI's 01 model. Google is apparently working on a reasoning model. And then we have agentic AI, right? Where in theory you could have millions or billions of AI agents out there working in the cloud on solving business problems. So is there even a better way than the current setup that we have right now? And how can Quantum maybe change it? You know, are we not looking at large language models in GPUs in the future, at least when it comes to AI?

Well, there's multiple answers here. We're looking at IonQ at different part of the layer inside an LLM and see whether or not we can move that workload over to a quantum processor. So one of the things, for instance there's linear algebra. This is one of the computational, heavy computational needs inside training and inference for LLMs. And that's one of the things Quantum is good at. So we're looking at now replacing that with a QPU instead of a GPU or a cpu. And so it's early days on that, but we've managed so far to duplicate the results that you can find classically. But give us another six to nine months and we'll see if we can make it better. The issue isn't even just the energy usage. The question is, can the end result actually be a better LLM? Can it capture the signal and the data better? And can you do it with less data? And lots of our machine learning examples we have shown that we can produce a better model but with far less data. Obviously if you can get an order of magnitude or several orders of magnitude less data, then that's less power that you need because that's less number crunching that you have to run through.

Yeah, and I think maybe it's a good time, Peter, to address people like Michael out there. So Michael's saying this interview is a little bit over my head. Hey, don't worry, don't worry. We're gonna hopefully simplify this there. But he's asking, it seems like quantum computing is powerful, but does it require the same amount of power? And then also I, he's saying, I thought quantum computing was only theoretical, so maybe could you take, take that question there in, in two parts there, Peter?

Yeah, so it is not theoretical. We actually produce working quantum computers that are working on solving all sorts of different business problems right now. In fact, actually for your listeners, you could go out to any of the major cloud vendors and spend 10 bucks and you could write your first quantum program in the next 20 minutes. If you had a credit card and an account with the cloud guy, you can give it a try. And there's a bunch of jupyter notebooks to be able to run your first quantum 101 algorithm. So, yeah, they are here today. They're available to everyone. And so they're, they're not theoretical anymore. They're now real. As to the amount of power, as we talked about, well, on one side, you need two and a half billion GPUs versus one of our QPUs. And so, you know, it's, it's a huge savings in energy if you can map your problem from a GPU to a qpu where that makes sense.

And is that, is that maybe why, you know, quantum is not as widespread or is not as popularized now? Maybe because that process. Right. Maybe everyone hasn't yet realized their own use case or hasn't quite figured out their own use case for running something on a quantum computer in the cloud versus a traditional GPU or cpu. Is that kind of why it's not more widely used?

We're just coming up to the cusp of where the quantum computer's power is actually more powerful than the world's largest supercomputers for certain problems. So, you know, if you were to go back the. Interestingly, today we have a 35 qubit system. We're about to do a 64. Well, remember, every time you add a qubit, it doubles its computational power. So the difference between a 35 qubit and a 64 is huge. So the 35 was not more powerful than the world's largest supercomputer. But this next generation will be, and the generation after that kind of leaves classical behind forever. And so we're just at that really interesting point. The other side is kind of like in the 1970s is people are just starting to create these applications. And so, you know, we're, we're in the quantum world. We're looking for the first killer application. You know, the, in the classical world, that was the spreadsheet. The spreadsheet was kind of the first aha. Application where everyone said I had to have it. Word processing was a fast follow. As soon as everyone saw those applications, then suddenly it took off like a banshee. So we're just coming up to that point. People are now working on their first applications that will really show disruptive potential of quantum computing.

So it seems like recently, Peter, a lot of the big companies, Microsoft and Google in particularly just in the last month, news has come out that they're making significant investments into nuclear power, specifically for AI for their data centers. Is this it? Can this be a one, two combination? Right? Is this a good pairing kind of nuclear energy and then also quantum computing? And yeah, can you maybe explain, do those two things in theory exist at the same time?

You know, it's, it's clearly, you know, as a species, we need to reduce our, our energy demands. Right. We're trying to get to, to go solve climate change. So building a lot of new power plants is probably not a really great idea, just kind of going into the future. And in particular, the classical approach to LLMs is driving energy demand exponentially. It has a, you know, when, when will we be happy with, you know, it's just, it's always going to be wanting more. There's never something where it's going through and wanting less. So I don't know, it's kind of maybe in a, a very basic analogy, which is if you're using the wrong tool for it, if you're trying to hammer a nail with a banana, you're going to need a lot of bananas. So it's just the wrong tool. So, you know, I would say that a lot of companies, their very simplistic answer is, let's build a lot more power plants. But maybe we just have the wrong tool for the particular problem that we're going after. And so maybe if we looked at it from a quantum computing perspective, that we could get the same results but with, you know, 1% of the energy usage. So I would be careful to build a lot of power plants Those might be sitting idle if, if they're not careful sometime in the near future.

So what's, maybe what's next? Right? So you kind of talked earlier that we were kind of almost at this inflection point and you know, even helping people realize, like myself, like, hey, you can go right now with a couple of bucks and you know, start taking advantage of, of quantum, you know, quantum computing if you have something that it can be used for. But you know, what's next, right? Like is, is quantum computing going to become the standard in 5, 10, 20 years? And you know, if so, what is that? Maybe big hurdle that has to be cleared in order for, you know, that to happen. And then in theory that can bring down kind of this, you know, over reliance on maybe, you know, power sources that aren't the best.

Right? To, if you look today, chemistry and optimization problems drive almost the majority of the demand in supercomputing and by definition, data center usage. Those are the two applications in particular that quantum is really good at. If we can move those workloads from CPUs and GPUs to QPUs, then the energy needs for those resource requirements would be dramatically less. And so that is, and of course now we have a new usage which is now AI, and it is also consuming a tremendous. It's a little speculative to say it's still early in that exploration, but it looks like that Quantum would be good at that as well. And so those would be the three largest applications that demand tremendous amount of compute where there seems to be clear advantages in quantum computing. So as soon as you see the first application, like in optimization or chemistry, where a company sits down and says, I have decided that I'm going to do drug discovery on a QPU because it's better and cheaper than using the cloud or my classical resources, that's when this will really start to take off.

All right, so Peter, we've covered a lot in today's conversation and in a 25 to 30 minute talk, we can't solve this. But as we wrap up here, because we've talked about everything from quantum versus classical quantum computing basics, we've talked into problem solving capabilities, we've tackled this from a lot of different angles. But maybe what is your one most important takeaway, you know, when we talk about AI's energy crisis, can quantum actually save the day?

Well, you know, it's, it's, it's an interesting question. Is human intelligence naturally quantum? And so is there another way to actually solve strong AI that doesn't even involve A large language model. Maybe there's other ways to do it. Just to make the point for your child, we don't teach it your child. With millions of repetitions, you only have to tell your child a couple of times a particular fact, and they managed to learn it. Clearly, the way that humans learn is very different than the way large language models learn. And so, and you know, we yet to understand that, but maybe that's a quantum process that's very different. And if you kind of look at it from that point of view, what is the energy required to teach a, you know, a third grader mathematics? Well, that for some strange reason doesn't require a power plant to teach that child. So it's just clearly different. So the, the question in the future, we're kind of, everyone's excited, right? You know, for good reason about large language models, but it's clearly not what humans are doing. And so maybe there's another way, you know, and so, and maybe that is, and this is very speculative because we don't know, but maybe that's actually a quantum process.

Oh, love to hear it. You know, I think, I think all of us in, in a very short period of time got so much smarter now about the future of computing and how quantum computing can play a huge role in that. So thank you very much, you know, Peter Chapman, president and CEO of IonQ, for joining the Everyday AI show. We appreciate your time and insights.

Thanks, Jordan. It was a pleasure talking with you.

All right, y', all, that was a lot of great information. Don't worry, we're going to be recapping it all in today's newsletter. So if you have not already, please go to your everydayai.com if this was helpful, please share this. If you're listening on a podcast, please make sure to follow the show, leave us a rating and subscribe. And make sure to join us tomorrow and every day for more Everyday AI. Thanks, y'. All. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit your everyday AI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.