Is an AI Energy Crisis Looming?

00:00:00:02 - 00:00:21:19

Mark Mills

I would say in the electric electricity business, just like any energy business. But what will happen in the future is what's happened in the past. Priority will be given to residences. It's no politician is going to let permanent condition occur where the industrial sector gets energy before residents as the citizens get it. It just won't. Is won't happen, at least not in the democracy.

00:00:21:19 - 00:00:43:05

Mark Mills

It could happen in Russia, not here. And even there they'd have trouble. So that won't be the trajectory this sector will be to figure out how to make enough electricity. And if the industrial sector can't get the electricity they need, they'll go elsewhere, which is which is what's happening right now, by the way, Germany. Germany is the industrializing, says the United Kingdom, because they're not building enough electric power plants that are reliable.

00:00:43:07 - 00:01:08:18

Mark Mills

So I'll make a couple predictions. First, the plans for shutting down coal and gas plants will stop. In fact, they'll probably reverse plans, as they already did in Germany. So I think you'll see, however, that however it's packaged politically, the temporary pause extension, whatever, whatever the languages. Well, we're going to stop shutting down reliable, reliable power plants.

00:01:08:20 - 00:01:19:05

Chris Keefer

Welcome back to decaf off. Today I'm joined by returning guest Mark P Mills of the Mark. It has been, I think, over two years since you've been on, which is, a crime.

00:01:19:05 - 00:01:21:08

Mark Mills

Really? It's an eternity. It's an eternity.

00:01:21:09 - 00:01:38:09

Chris Keefer

There's a lot. It's an eternity. In t couple terms. It is, it is. We just celebrated our fourth anniversary. back in May. so, yeah, we we keep on kicking about an episode a week, and, no, your two episodes were, very well received, and we've come a long way since then, so it is great to have you back in the saddle here.

00:01:38:11 - 00:01:57:05

Chris Keefer

I was just reviewing, you know, what you've been up to, and, I don't recognize half your affiliations anymore. executive director of the National Center for Energy Analytics. distinguished. Distinguished senior fellow at the Texas Public Policy Foundation. of course, you're still involved with the City Journal. I think it's a publication, the Manhattan Institute.

00:01:57:05 - 00:02:02:23

Chris Keefer

So, Marc, why don't you take a moment and, catches up with, all of your adventuring these days?

00:02:03:01 - 00:02:31:07

Mark Mills

Well, it's the, the exciting part of the grandchildren. Not not the, not the business side, but we'll we'll we'll stick to business. So earlier this year, I, launched with the support of the Texas Public Policy Foundation, which is a state based think tank. I think it's the biggest state based think tank in terms of people and resources to, put together a national think tank on energy, mainly because there really isn't one.

00:02:31:08 - 00:02:49:09

Mark Mills

I mean, if you think about the the energy domains outside of universities, there's most of the think most of the people who do scholarship in energy work inside of a think tank that does lots of other things which is fine. It's where I was at the Manhattan Institute and it was a nice home. I'm still, as you say, affiliated.

00:02:49:09 - 00:03:09:07

Mark Mills

I write, regular column, is a contributing editor for the City Journal. Great publication. lots of eyeballs. So it's a great place to tell, you know, tell a story because people pay attention. which is why people why we do. Why you and I do what we do. If people that pay any attention. Why? Why bother you?

00:03:09:12 - 00:03:30:23

Mark Mills

It's an echo chamber. It's. You know, it's like singing in the shower anyway. So we, you know, over the years, I've thought about this, and the colleagues have asked this question is, so, you know, where, in effect, where is there an energy think tank? The IEA is not a think tank. The EIA, the Energy Information Administration, the US is a quasi think tank.

00:03:30:23 - 00:03:56:00

Mark Mills

The data collect collecting entity. So bottom line is, with support from the TPF organization is back office with funding from their trustees and other people foundations that have supported me. We kicked off, announced in March, the National Center for Energy Analytics. We have a board of advisors of, in my opinion, very impressive people. Easy to find out about us at the magic website.

00:03:56:04 - 00:04:23:01

Mark Mills

So you just do energy analytics.org. You'll see our eight scholars. We're announcing a new scholar, I think tomorrow, and will be expanding to what I would hope will become a small army of energy scholars to weigh in. And, we have on every page of the website, I adopted a, a mission statement, if you like, which is I stole a quote from the Philip K Dick.

00:04:23:03 - 00:04:50:13

Mark Mills

And you, you may, if you're a science fiction fan, remember Philip K Dick, prolific writer, science fiction, the 60s. A lot of his stories, became movies like Minority Report, was his his book, his short story novella became a movie a lot of. Anyway, he said, I thought brilliantly sometime ago, wrote reality is that which, when you stop believing in it, is still there.

00:04:50:15 - 00:05:13:02

Mark Mills

So we're a reality based think tank focused on trying to preach the gospel of reality, which is what I've been focused on my my whole life. So that's why we did it. we'll see. we've got some projects that we're going to announce soon. other than the usual things, scholars doing studies and writing op eds, I have a piece on nuclear energy.

00:05:13:04 - 00:05:38:06

Mark Mills

You may, may or may not have seen in InFocus quarterly. slide to start writing about nuclear energy again, because now it matters. It has it has it mattered in two decades? Because a lot of reasons. so I began my energy career in the nuclear industry. So I thought would be a good time to way back in when I began my energy career in the Canadian nuclear industry, as you may or may not recall.

00:05:38:08 - 00:06:08:06

Mark Mills

So my introduction to the nuclear debates and the energy debates was in Canada when the, pre Three Mile Island, when Canada was debating whether or not they should continue or abandon its nuclear power program, over some degree, concerns of safety, but primarily, concerns about nonproliferation. That was the primary motives that Congress, Congress, Parliament ordered the port au Commission hearings to determine whether Canada should proceed.

00:06:08:08 - 00:06:33:05

Mark Mills

And, obviously, Canada decided to proceed, but that was I was involved with that at that time. So it's a long way of saying that, you know, the center will look at a lot of things like nuclear energy or gas exports, the stability of grids. the debate over the so-called energy transition. You know, my opinion on that is I there isn't one, but it's okay if people people want to ignore data.

00:06:33:05 - 00:06:59:12

Mark Mills

That's okay. You're left. You're allowed to ignore facts. And, well we'll see. I mean, we've got some, I think, surprising things we'll be announcing over the coming six months to help to reset the framing of how to think about and debate about energy. And there's lots of things that people can disagree about and for, you know, substantive reasons or political reasons, perfectly reasonable.

00:06:59:13 - 00:07:14:06

Mark Mills

But as you know and you know, I've talked about this many moons ago, I'm a huge fan of of trying to have those debates from a sort of fact based perspective and then let the opinions flow from the facts.

00:07:14:08 - 00:07:35:07

Chris Keefer

Right, right. That's that's a good, order of operations there. so science and nuclear, you mentioned nuclear matters again. and I'm guessing part of that has to do with, the AI revolution. you know, the word energy crisis was tossed around by Sam Altman not too long ago saying that AI is, going to cause or is causing a new energy crisis.

00:07:35:09 - 00:07:54:10

Chris Keefer

Projections of, it's just really skyrocketing demand. And at least in North America, I mean, obviously with the, you know, the global picture, there's been increasing additions to electricity grids in places like China and India and other emerging economies. But we've kind of been flatlining. in North America, compute has been going up. It's been getting more efficient.

00:07:54:10 - 00:08:15:18

Chris Keefer

Maybe it's been, you know, balancing itself out. but, it seems to be surprising. Maybe everybody but you. I know you've been writing on this topic for some time, that essentially, our, our appetite for information is not sustainable. so. Yeah, let's, let's, let's dive in and talk a little bit. You wrote a book. I think it's published in 2021.

00:08:15:20 - 00:08:26:21

Chris Keefer

the Cloud Revolution, convergence of new technologies will unleash the next economic boom and a roaring 2020s. That's a kind of starting. yeah. Boom, boom. Let's talk about the boom.

00:08:26:23 - 00:08:55:15

Mark Mills

Well, I, you know, the title, the book, was was deliberate in the sense that the cloud is the most interesting and most consequential new infrastructure in, in our economy. But the book's about more than the cloud. It's about the clouds role, well, synergistically, as a driving force and as an amplifying force in transportation, in basic research, in health care, education, entertainment, material science, all these things are interrelated.

00:08:55:17 - 00:09:15:09

Mark Mills

They've always been interrelated. But I but the book, the thread is the cloud. And, I wrote a lot about the energy appetite of the cloud in the book because I've been, as you said, I've been writing about this and studying it for a long time, a couple decades, in fact. So it's kind of there was a, Senate hearing for those in your audience in Canada.

00:09:15:10 - 00:09:36:18

Mark Mills

Doc. It's just, I don't know, the US system. It's very much like the parliamentary system is that you have hearings before committees, and, the chairman can call a hearing on a subject and the set the Senate chairman, the Energy Natural Resources Committee called in a quasi emergency hearing for the same reason. Sam Altman said what he said was an energy crisis.

00:09:36:20 - 00:10:09:10

Mark Mills

And the reason they called it was that, utilities around the United States were reporting staggering increases in forecast demand for electricity. No new requests for connection. not everywhere, but in a lot of a lot of regions in the northeast, the west, southwest and south. And, a lot of it was from data centers, fact, the majority of data centers, but a lot of it's also from new factories, because the United States wants to repatriate its manufacturing from a variety of reasons, most mostly good reasons, not not always.

00:10:09:12 - 00:10:32:16

Mark Mills

So most. I'm a huge fan of manufacturing. I worked in manufacturing. I love to repatriate manufacturing. And of course, electric vehicles by definition have to be plugged in. So those are three vectors that are going on causing, unexpected increase in electric demand in the near future. But the biggest one is data centers. And the the line has been it's entirely because of I.

00:10:32:18 - 00:11:02:22

Mark Mills

I've written about this a lot recently as well. AI is the most energy intensive use of silicon ever by a factor of roughly ten. that is the same amount of time spent in compute with, regular CPUs and GPUs. So a conventional task that we're doing now, if you if you amplify that task with, some machine learning and AI, typically you increase the amount of energy used by the computing by tenfold.

00:11:03:00 - 00:11:32:08

Mark Mills

It could be as little as a double. It could be as much as 100 fold. Depends on the task. But ten fold is not bad. So knowing that and this is not news, by the way, the AI people are on this for a long time. I mean, if you think about it, the difference between, it's computing chips that run, algorithms that are, GPU based or tensor based, you know, inference base as opposed to calculation based is essentially the difference between, an airplane and a car.

00:11:32:13 - 00:11:51:22

Mark Mills

And by that I don't mean in absolute terms of energy, but in terms of how you use the engine. An airplane engine is running constantly, and you want to minimize the amount of time it's on the ground. It's expensive asset, but that's there always operating power plants run 24 seven if you can. Cars don't you know the personal car you turn on and off.

00:11:51:22 - 00:12:09:10

Mark Mills

You use it when you need it most. Compute is like a car. It consumes energy when you're using it, but not in between. It is not running flat out all the time. Machine learning, both in the inference and in the inputs and the learning phase, but especially in the inference phase. But it's trying to infer what's going on.

00:12:09:12 - 00:12:28:12

Mark Mills

Whether it's drawing a picture or answering a question is running flat out. It's uses a lot more electricity per minute, and it's meaningful as you expand it. But it turns out if you look at the data, AI is not the real reason that this is movement. Data center demand a data center demand was going up already has been going up for two decades.

00:12:28:14 - 00:13:00:05

Mark Mills

What happened is, in short, our economy's moved along North America. We got we got we finally absorbed what you can do with computing. If you look at the the rise of the app economy, which is very recent, the explosion of number of apps, apps are, everybody knows what they are. It's an application specific program. But that's that that innovation by itself was much more consequential so far than I, because what that's done is increase the utility of computing.

00:13:00:07 - 00:13:20:22

Mark Mills

Apps are very useful. So what that did is cause an explosion in the need for data centers, these platforms for apps. Everything about social media is app based. Everything about the sharing economy is app based, Uber is app based. And the list is there are hundreds. In fact, there are thousands of apps, app platforms, and there are millions of apps now.

00:13:21:00 - 00:13:45:07

Mark Mills

So what that did is cause it. We finally got computing connectivity good enough, accurate enough, and this is all been improving the last two decades or incrementally behind the scenes compute in the cloud, cheap enough that we have this explosion of, of uses that are actually fun, interesting, productive. That's what's been driving it. Now I as amplifying and accelerating that without a question.

00:13:45:11 - 00:14:06:20

Mark Mills

But if you look at the absolute demand for electricity from data centers, it's what's driving it. The AI part of the data centers maybe added 10% last year to the total demand. So it wasn't as if I came along with ChatGPT and oh gee, we now we need lots of ChatGPT 10% on a big number is a big number, but it's 10%.

00:14:06:20 - 00:14:24:16

Mark Mills

And maybe next year it'll be 1,250% of it, you know, maybe 20%. It'll keep adding to the base. But right now, the fundamental discovery is that if you build lots of data centers to do lots of useful things that people want, it consumes lots of electricity. Who knew?

00:14:24:18 - 00:14:44:09

Chris Keefer

Right? Right. You mentioned that, you know, this construction of this nebulous cloud is actually a lot of physical infrastructure, perhaps our most recent great infrastructure build out. You know, once we built our highways and bridges and ports and airports and roads. but it's one that's, you know, invisible. And, I mean, I think we're becoming increasingly, you know, an energy and materials blind society.

00:14:44:09 - 00:15:02:06

Chris Keefer

We don't learn about these things. In school, I was watching, I was watching a great Ford mini documentary called A Car Is Born. And this was back in the day when advertising, you know, it ripped the Band-Aid off and it showed you it. They were proud of, you know, blowing up massive amounts of rock to extract the iron ore and palletizing it in their blast furnaces.

00:15:02:06 - 00:15:18:10

Chris Keefer

And, you know, this was the big vertically oriented beauty, and they wanted to brag about it. You know, the the world is fundamentally changed in terms of people's perceptions. And I'll confess myself to be quite blind to this infrastructure describing. So I, I love getting a sort of granular on the ground sense of what exactly it is we're talking about.

00:15:18:10 - 00:15:34:19

Chris Keefer

So maybe you can describe, some of these data centers. I've heard them maybe sort of maybe greenwash is the wrong word, but data campuses, data foundries, what's the term that you use? And what the hell do these things look like? And, and the infrastructure that ties them all together.

00:15:34:21 - 00:16:01:04

Mark Mills

Well, that's that's a good question because the, you know, the reason words matter. I mean, the things that we learned, in elementary schools, sticks and stones, you know, it break your bones. Words. Can't words hurt when people use words? It don't matter. They hurt. They really do hurt. And words also mislead. phrasing matters. Nomenclature matters. This is why advertising agencies work, and politicians work so hard at phraseology.

00:16:01:10 - 00:16:29:10

Mark Mills

Why they they bend over backwards to come up with clever, elliptical, Orwellian names for things like in the United States, our Inflation Reduction Act, which will cause a lot of inflation. Inflation creation act. So, data center doesn't mean anything. It's it's like it's like calling a, I don't know, a skyscraper or a, you know, I don't know, a vertical strip mall or, you know, a shop, call it.

00:16:29:12 - 00:16:54:18

Mark Mills

but you just think that we call it a skyscraper, something different than the building, because it's really different than a regular building. So tall, so many square feet under. And the word data center finds its roots right in the early computing era, where you had to walk to the built part of the building that had the computing, the computing center, and it had the IBM mainframe in it, and you'd walk through it with your to do your work.

00:16:54:18 - 00:17:21:04

Mark Mills

And maybe, if you're lucky, you had a remote terminal. I did when I was at Queen's University. We had a data center with a remote terminal computing center. So that word is stuck with us, but it doesn't evoke anything visually useful. Google engineers some time ago, about 15 years ago, coined the phrase warehouse scale computing. And by that, they thought we should think in terms of the data center.

00:17:21:04 - 00:17:57:07

Mark Mills

It's not evocative enough. Most people know what a warehouse is. It's a really big building. So if I use the phrase warehouse scale computer, then you're thinking, that's one honking big computer. It's exactly what a data center is. It's a warehouse sized building size of a Walmart, or really literally a warehouse. Massive building, 1,000,000ft² in a and it's by a large has very few humans, maybe a dozen, but it has millions of CPUs and GPUs, tens of thousands of servers, computer compute servers, storage, memory in a single giant building.

00:17:57:09 - 00:18:27:08

Mark Mills

And it's it's there because like a power plant, there are economies of scale to putting things together like that. You know, speed of light actually matters. Even though light seems fast, it's not fast enough for the kind of calculations that go on. And computing distances of feet matter. You can't do real time data duplication for financial transactions, which is required by law in the Western world, you have to duplicate financial transactions, but you can't do real time duplication.

00:18:27:08 - 00:18:58:16

Mark Mills

If you're a duplicate data center, it's more than a couple thousand kilometers away because the speed of light is not fast enough to get it in actual real time. If you're processing millions of transactions a second. So you if you you can do arithmetic here to figure how many milliseconds you can you can tolerate not many. So these are really big buildings and how many and this what the focus is on because that's where the platforms reside if you like.

00:18:58:18 - 00:19:22:23

Mark Mills

But the buildings themselves are only part of the story. Give context. They have as many square feet, typically on these enterprise scale data centers as a skyscraper. You could call a skyscraper an enterprise scale office building, right? When people say there's an enterprise scale data center, but they mean it's instead of a small room with a computer in your office, we have a big building.

00:19:23:00 - 00:19:45:00

Mark Mills

Well, it's like saying, I said, having a small office in a strip mall. I have an enterprise class office building vertical, let's call called a skyscraper. There's lots of them in Toronto, in New York. But the if you look at them in square feet and cow, the world has, you know, several hundred so maybe a thousand enterprise class office buildings, skyscrapers.

00:19:45:02 - 00:20:10:05

Mark Mills

The world has about 6000 enterprise class data centers already. an enterprise class data center, same square footage uses 100 times more power per square foot. We're building them at ten times the rate. Because we are skyscrapers. They cost the same to build. Interestingly. But if you're the owner, you can rent a square foot of a data center for about 500% more than you can rent a square foot of a skyscraper.

00:20:10:07 - 00:20:33:23

Mark Mills

Because of how much economic value there is in a square foot of a computing. You, warehouse scale computer. So put it then you have to connect data centers to people and other data centers and our devices and machines. So those networks, you know, the old phrase, the information superhighway is really apt. Here they are. There are information highways and they are wired and wireless.

00:20:33:23 - 00:20:55:20

Mark Mills

We have, you know, glass cables, copper cables still and wireless wireless are simply invisible cables. You're doing the same thing. And if you count the mileage of all the data centers connecting to all the devices, think of that as you know, it's a network. And compared to highways, global highways, you know, you can measure all the world's roads collectively.

00:20:55:22 - 00:21:18:18

Mark Mills

and, you know, barely a million plus miles. It's a lot of miles. But the network, data center networks are measured in billions of miles. I mean, it's a thousand times bigger. And, yeah, you know, you have it's a lighter, load on the network, but there's a lot more traffic. So these networks carry traffic that we measure and dollars and bytes.

00:21:18:20 - 00:21:43:12

Mark Mills

So to give a network the infrastructure context. So I try to think of it in terms of square feet or in square footage. It's much bigger than all the world skyscrapers by a factor of ten and growing at triple the rate in dollar terms. we're spending more money building infrastructure for the cloud globally, capital than all the electric utilities in the world are spending combined to build new power plants.

00:21:43:14 - 00:22:06:08

Mark Mills

So it's a lot it's a lot of hardware, a lot of money in energy consumption terms, because the infrastructure of the cloud uses energy, doesn't produce it. It uses roughly twice as much energy as a country of electricity, rather as a country of Japan. Or put in fuel terms, it's bigger than global aviation and growing faster than global aviation and energy consumption terms.

00:22:06:10 - 00:22:30:11

Mark Mills

I it's going to accelerate that because I the analogy that I use in the piece that you might have seen recently was that if you think about the era of computing, which is why I wrote my book, I think we are at the era where computing with a cloud is roughly equal to 1957 58 for aviation. At that point in aviation, the Boeing 707 was introduced with jet engines.

00:22:30:12 - 00:22:53:16

Mark Mills

The first commercial jet aircraft was made possible, but at that point had already been 35 years, three and a half decades of aviation. Hundreds of aviation companies air travel around the world. Overnight mail was already happened overnight. Took that because Postal Service was using airplanes already for 20 years to carry mail across country. Obviously used in war fighting.

00:22:53:21 - 00:23:32:12

Mark Mills

Aviation was a huge industry. But but in the 1957 58 timeframe, the addition of jet engines to the airframes caused an explosion in the use of aviation. That was the modern aviation era. I like adding jet engines to the airframe. We are we are at call it 1960, roughly, in terms of the growth now coming in compute compute infrastructure and compute utility, all the things that that aviation brought to the world, not just entertainment and convenience, you know, and, commerce, that kind of acceleration.

00:23:32:13 - 00:24:01:02

Mark Mills

And then more comes with the cloud because a cloud is a general purpose technology, aviation is not right. Aviation has a just a transportation technology. Cloud amplifies transportation, amplifies research, amplifies education. So the infrastructure. But every measure is astonishingly big energy, money miles, square feet. It's in its infancy. We're at the end of the beginning. That's beginning of the end, which is really crazy when you think about it.

00:24:01:04 - 00:24:20:05

Mark Mills

And then the energy implications of it are what caused the hearing. So, well, what are we going to do? How are we going to power all this stuff? It would it would be as if we held hearings in 19, you know, 64 early in the era of the explosion of global aviation, and people held a hearing to ask where they're going to get the fuel to fly airplanes.

00:24:20:07 - 00:24:46:20

Mark Mills

Well, make more fuel. The answer would be make more fuel, but you'd make you have to make sure cheap, because the cost of flying about a third of it's in the cost of the fuel. And we're not. We're now moving to an era where the cost of computing will become like flying. half, you know, anywhere from a quarter to a third of the cost of computing for it can can end up being associated with the cost fueling the compute, not not buying the chips.

00:24:46:22 - 00:25:16:15

Mark Mills

Chips get commoditized fast. Power plants are very hard to commoditized. This is sort of the asymmetry that I wrote about my book, The Physics of Producing Information and the demand for information. Our exotic and weird, the physics of producing energy are, well, no, not exotic. They're also weird, but they're highly limited. They're there aren't many. there's no Moore's law on energy production, whereas there's a moore's law, if you like, in an energy, in information production.

00:25:16:17 - 00:25:37:01

Chris Keefer

So I just wanted to, tug on a couple threads. you were talking about, you know, compute not being, the speed of the speed of light being too slow. And so inferring, at least for certain kinds of important transactions, financial ones, you're mentioning, the need for a distribution of data centers. I know, you know, I remember hearing ten, 15 years ago that Iceland was, you know, attracting a lot of data centers.

00:25:37:01 - 00:25:38:03

Chris Keefer

It's cold there.

00:25:38:05 - 00:25:39:21

Mark Mills

So Norway. Yeah. Yeah.

00:25:40:03 - 00:25:59:13

Chris Keefer

Plentiful. But but is there also a need to sort of just like cell towers, plug these things in a distributed enough network that they can move, move close to the speed of light and make those transactions happen? Or can you, can you sort of, you know, regionalize and, you know, another thing I've heard is that certain jurisdictions are saying we don't want to have a data center here.

00:25:59:13 - 00:26:09:01

Chris Keefer

We're going to use up all our power. You're going to create 12 jobs. you know, why? Why would I do this for my constituents? So to what degree is that? Is that distribution necessary?

00:26:09:03 - 00:26:38:05

Mark Mills

Well, the latter point for the latter point you make is what happens with any industry gets to scale. You end up with the classic Nimby problem, right? Even if it's not about power, it's it's, you know, big things. take up space, resources, land and energy. And some people don't like it. And so you have to you have to you have to deal in a democracy with, you know, compromises their movement, pay higher taxes, make it look pretty, bury it.

00:26:38:08 - 00:27:01:23

Mark Mills

Lots of solutions. But they all cost money. And that's sort of where the industry is, is now in many, many jurisdictions. Ireland, banned the connection to the grid of new data centers because a fifth of all Ireland's electricity is now used to power data centers. This is quite remarkable. And the complaint was the data centers were getting to use all the all the good electricity, wind and solar and the citizens couldn't use it.

00:27:02:01 - 00:27:25:03

Mark Mills

We're being, deprived of that opportunity. you know, this is and kind of infantile arguments because electricity gets is fungible and gets commoditized on grids anyway. But I get I get the concern. I understand that, but economic growth, what people are forgetting is that data centers epitomize the key feature of modern economies. Economic growth requires more energy.

00:27:25:05 - 00:27:47:05

Mark Mills

The fact you point out earlier that the electric demand growth has been slow, very slow in the last 15 years, and North America is not because we we, we shouldn't expect it to rise again. I've been predicting it would rise again because of a related phenomena. Our our economic growth was relatively slow slowing. That should have been for a variety of reasons, both both predictable and unpredicted.

00:27:47:08 - 00:28:11:04

Mark Mills

That is, you know, recessions and like, it also what I would call anti-growth policies, both governments episodically. I'm not talking necessarily about the ruling parties today, but episodically over the last 15 or 20 years, both Canada, the United States have implemented policies that are anti-growth, you know, excessive regulations that make it very difficult for, for, businesses to, to function here.

00:28:11:04 - 00:28:33:15

Mark Mills

So they go elsewhere, they go to China, they go to Indonesia, Vietnam. but some of that's reversing. But also, most importantly, and this is relevant to computing and everything else, all efficiency gains and reducing the amount of energy you need, electricity, you need to do lighting, to do air conditioning, to cook the heat. we've had tremendous technology efficiency gains across the board.

00:28:33:17 - 00:28:57:23

Mark Mills

And and the things that we use, in industries and at home and those efficiency gains. I've been really remarkable the last 20, 20 years. A part of that has been driven by regulations. Most estimates have been driven by normal technology progress. We we we find what engineers do in every, every domain. Is it always seek to compete by making whatever they built more efficient?

00:28:57:23 - 00:29:20:08

Mark Mills

That's what you do. But what happens is you run into the limits of what the efficiency gains can be, your physics limits. That's what's happening now. We've we've sort of wrung out all the low hanging fruit, so to speak, and, out of efficiency gains and most things, from air conditioners to heaters to lighting, to even to, to a great extent to many aspects of computing.

00:29:20:10 - 00:29:43:18

Mark Mills

So now you're entering a new era. So what? So what do we do? Well, you know, we have to produce more electricity. That's what that's we don't have. We don't have a lot of options. or they'll go elsewhere. So this gets to your question about distributing the data centers. Where can I put them? So simplistically speaking, there's two kinds of tasks.

00:29:43:19 - 00:30:14:15

Mark Mills

Well, there's tasks that, can be done any time. For example, if I want to store your videos or your pictures that you take of your family and friends and kids and whatever, it doesn't really matter that much if the cloud storage is in Iceland or in Reykjavik or, Reykjavik, Iceland, or whether it's in the network, you know, Northwest Territories, it makes a matter because you can access it in a, in a half a second or so.

00:30:14:19 - 00:30:31:10

Mark Mills

And the latency is not offensive. It's not that the speed of light is pretty good. but if you really if you want to run video and real time streaming video, which is what people are, then you latency of a half a second gives you video jitter. So all that stuff has to be pushed to the edges of the network.

00:30:31:12 - 00:30:58:01

Mark Mills

And in local data centers so that they can stream to, you know, jitter with lots of data, high bandwidth stuff. So it depends on the task. A lot of if you try to do a calculation, you, you know, during machine learning, I want to teach a machine learning model how to recognize diseases better than an old model. I can have all the data in a data center anywhere on the planet, and let it run wherever it's cheapest.

00:30:58:05 - 00:31:20:05

Mark Mills

Once it's done, the learning task, wherever it's cheaper, is easier. I can then put that learned model where it can be used, so that can be done anywhere. But anything that has to be done in real time or latency matters, which is computing the not so anything that has to do with real, real world automation. It has to do with things like video, things like we're doing that with zoom.

00:31:20:07 - 00:31:48:05

Mark Mills

latency matters because you get all that jitter and freezing. And it's not only unacceptable as a like a consumer product, it can be lethal if it's automating a machine that is interacting with a person, whether it's automating it, a car remotely or an aircraft and I think the ultimate self-driving cars will end up being controlled not by computers in the car, but by computers on the edge of the network.

00:31:48:05 - 00:32:13:17

Mark Mills

Close to the car. Not in the not a long way away, a cloud nearby cloud, because then you can put lots of computing horsepower with low latency very close to the vehicles, and then you can have a functionally infinite amount of compute power. So the application will determine it. What will happen. This will be applications for everything, but the model for a network, including the cloud, is what some analysts have called the peach model.

00:32:13:19 - 00:32:37:06

Mark Mills

Right? The things on the edge or on the surface of the peach, the fuzz. And in the middle there's the meat. And the core is that it's the pit, it's the peach. Gross. Everything gets bigger, but geometrically the surface gets bigger faster. So there's far more applications for compute on the edge with, you know, light compute AI in your device, I and your headset, I in the thermometer.

00:32:37:06 - 00:33:03:05

Mark Mills

I had to refrigerator these, lightweight chips, so to speak. And energy terms can can perform rudimentary AI tasks on the edge. But anything that's a heavy lift, doing anything to do with health care, anything with basic research, anything analytical, the compute horsepower you need in general. And for AI is so huge that they, you know, you can't use a single AI chip on to do it.

00:33:03:10 - 00:33:31:00

Mark Mills

You're using hundreds of thousands of AI chips. Well, to do it's not hard to figure out if a single AI chip, a single GPU, and a card, consumes one and a half to two kilowatts. Typical small Canadian house is two kilowatts and means midsize house is, you know, four kilowatts. Peak big house got five. So if I have a card and, you know, a computing card, as you probably know, maybe your audience knows this.

00:33:31:02 - 00:33:55:18

Mark Mills

It's like this. Think of it as like a stripped away laptop without a case of it. Something like that. But if I have one of those at two kilowatts and that's but have four GPUs on it, but I need 10,000 GPUs to do the arithmetic here. I got a rack with a hundred of them. So now I'm at 100kW in something the size of a refrigerator in your kitchen, 100kW.

00:33:55:19 - 00:34:21:22

Mark Mills

And then I put a thousand. I'm in a building called a data center. Now I'm at I'm at a 100MW. These are this is this is why the nuclear debate has been reignited. We're building data centers, each one of which uses as much electricity as can be produced by small nuclear power plant. So now you have all the tech guys say, oh, gee, well, maybe we should buy nuclear power plants or on our data centers say, well, okay.

00:34:21:23 - 00:34:22:19

Mark Mills

All right.

00:34:22:21 - 00:34:30:08

Chris Keefer

And we're we're even hearing of, of gigawatt scale, data centers. Yeah. sure. There was interesting.

00:34:30:10 - 00:34:31:20

Mark Mills

Gigawatts of big a big number.

00:34:32:02 - 00:34:49:06

Chris Keefer

It has a big number. It has a big number. and so, yeah, you're saying nuclear is back? I mean, there's lots of talk now of, obviously the tech bros and Silicon Valley, you know, Sam Altman, imagining that fusion is going to be that, energy revolution that will power those data centers, talk of advanced nuclear.

00:34:49:08 - 00:35:10:07

Chris Keefer

it seems like, given the immaturity of that sector, and how long it takes inertia to, to, to develop it. we're looking at upgrades of existing facilities, looking at bringing Three Mile Island unit one, the one that didn't meltdown, back on line, that was shut down for economic reasons just five years ago. I believe in 2019, due to cheap natural gas prices.

00:35:10:09 - 00:35:31:18

Chris Keefer

it's, it is a pretty extraordinary turnaround. And I think one element of that is the power quality issues. PJM is requiring is retiring tens of gigawatts of coal and gas over the next decade. you know, I think our vertically integrated monopoly utilities, it probably built a pretty healthy capacity in there. and, you know, we've had pretty reliable power as a result.

00:35:31:19 - 00:35:51:12

Chris Keefer

we're saving that down. We're redlining, and reliable power baseload back, baby. That's what it sounds like. so what do you think the risks are of, you know, well-capitalized tech companies being able to sign above market GPAs and basically gobble up all of that reliable nuclear electricity and and leave some of the, you know, the lower quality power for the rest of us.

00:35:51:12 - 00:36:12:02

Mark Mills

Paroles let you, let you deal with the lights being out and the refrigerator not working or you're air conditioner not working in a heat wave. I would say in the electric electricity business, just like any energy business. Well, what will happen in the future is what's happened in the past. Priority will be given to residences, as will happen.

00:36:12:04 - 00:36:36:08

Mark Mills

It just will happen. It's no politician is going to let, a permanent condition occur where the industrial sector gets energy before race residences and citizens get it. It's just is won't happen, at least not in the democracy. It could happen in Russia, not here. And even there they'd have trouble. So that won't be the trajectory this sector will be to figure out how to make enough electricity.

00:36:36:10 - 00:36:56:22

Mark Mills

And if the industrial sector can't get the electricity they need, they'll go elsewhere, which is which is what's happening right now. By the way, Germany, Germany's de industrializing says United Kingdom because they're not building enough electric power plants are the reliable. So I'll make a couple predictions. First, the plans for shutting down coal and gas plants will stop.

00:36:57:00 - 00:37:28:13

Mark Mills

In fact, they'll probably reverse plans, as they already did in Germany. so I think you'll see. However, however it's packaged politically, the temporary pause extension, whatever, whatever the languages. Well, we're going to stop shutting down reliable, reliable power plants because the the modus operandi in North America so up the last 15 years has been shut down. Existing reliable power plants, which would be nuclear power, coal and gas and replace them with unreliable power plants when when the batteries will.

00:37:28:15 - 00:37:50:11

Mark Mills

And so what will happen going forward is we'll do both because if you have electric to market growth, you're going to build both. You're not going to do one or the other, which is the operation of the last decade, just not sustainable. So you're right, the data center can't run on episodic power. And if they could and they thought episodic windMills were just fine with batteries, they'd buy the batteries.

00:37:50:13 - 00:38:13:02

Mark Mills

It, you know, instead of instead of trying to convince governments to force utilities to buy big batteries to keep the grid lit, if if the Technorati really believe that wind and solar and batteries combined are cheaper than conventional grids, they can go ahead, disconnect from the grid and build those and run the data centers on that. And they're not.

00:38:13:04 - 00:38:35:06

Mark Mills

And what they're doing is they're buying, shell game, funding windMills in Europe elsewhere to get credits that they can say we're running our data center on green energy when they're there, are doing it legally. It's a legally allowed credit, but the data center is is not running on those wind turbines. Obviously, it's running on whatever grid it's connected to.

00:38:35:08 - 00:38:55:06

Mark Mills

And that grid typically certainly in North America, is 70% powered by hydrocarbons. If you're in Ontario, you got a lot more nukes. And if you're in different parts of the United States, you get more coal plants and more gas plants in other parts. So there's there's an it is an interesting, challenge. This is what this hearing is about that I mentioned.

00:38:55:08 - 00:39:20:08

Mark Mills

utilities are now in a position where they cannot meet the demands that are coming for reliable power. Reliable means is, you know, not just, second by second and minute by minute, but hour by hour, but not just daily, but when there are wind droughts and solar droughts, those large areas of the country are covered with clouds and there's no wind when that happens, you can't cycle back data, so you can you can turn them off.

00:39:20:08 - 00:39:38:04

Mark Mills

A lot of a lot of products and services will stop working. So you have to have a grid that's reliable, and you're going to need a grid that it's expanding. So both will have to happen. Which means I think we're going to redound back to the reality that we have to do two things. One is short term. One is long term.

00:39:38:06 - 00:40:07:09

Mark Mills

In the short term, we have to stop shutting down power plants. Just you have to stop it and reactivate power plants that can be reactivated, whatever they are. And that's what's happened in Europe. They'll reactivate coal plants, nuclear plants, if they haven't been decommissioned and all over gas. But you need all the gas plants, and you'll need to accelerate the construction of gas turbines, because in the short term, in the next five years, which is when this stuff is being built, there is zero possibility.

00:40:07:11 - 00:40:34:14

Mark Mills

And this is a bet I'd take happy with anybody that will build another gigawatts of nuclear capacity, never mind 10 or 20 or 50GW in the next five years, or even ten years in the long term. We need to re reinvigorate the supply chain, the capacities to build nuclear power plants at scale. And by that I mean lots of little ones and lots of big ones, both the new generation of small plants and the older generation of big plants.

00:40:34:14 - 00:40:54:07

Mark Mills

We have to build both. We're going to need both. But roughly speaking, a decade to really animate the whole supply chain, to build at scale. it's certainly possible. In a couple of years, we'll see. The first of a kind of some new kinds of smaller nuclear power plants. That'll be exciting. They'll have to run them for a couple of years to see how they work.

00:40:54:09 - 00:41:17:20

Mark Mills

and then build a supply chain to start getting them up at scale. I'll take a few more years. So even on an accelerated basis, you're talking 5 to 7 years before you really begin to insert these power plants into the operation system, and then not at scales that matter. We're not talking tens of gigawatts. We need tens of gigawatts of new power plants, United States and Canada, huge quantities that'll have to come from.

00:41:18:01 - 00:41:53:07

Mark Mills

There's only one place. This is this isn't not rocket science. You can't build enough batteries to make that at any price. But even if you want to spend the money, that costs are astonishing combination of batteries and wind and solar to produce the reliable power is triple the cost of the nameplate price tag on these windMills because, as you know, I have to build 2 or 3 for each one in order to have the reliability plus enough extra electricity when they're operating to store for the non-operating times.

00:41:53:12 - 00:42:10:17

Mark Mills

And then I have to pay for batteries. When you do the arithmetic on that, you end up not with cheaper electricity. It is cheaper when the wind's blowing. But if you want electricity, when the wind's not blowing or the sun's not shining, it's not cheaper. We know for a fact it's more expensive by 2 to 300% sort of ballpark.

00:42:10:19 - 00:42:34:01

Mark Mills

So what do you do? Wow. Order nuclear plants. Great. I hope the tech industry orders up a ton of them. It'd be wonderful. But they're not going to. You're not going to see any of them because nobody can build them at scale. Yet what we know we can build rapidly. as either gas fired turbines or gas fired naturally aspirated engines, the so-called diesel engines at scale can be built very quickly.

00:42:34:04 - 00:43:00:00

Mark Mills

They're incredibly thermodynamic, efficient burn natural gas, not diesel oil. You can dual fuel them so you can arbitrage, fuel for on a price basis. But if one were taking a bet, it would be that we're not going to want to, we're not going to want to avoid having the economic opportunities associated with growth of whether it's EVs that people might want to buy.

00:43:00:02 - 00:43:23:10

Mark Mills

I think lots I think millions more of them will be sold. Yet new power, new, factories to build chips, chip fabs by themselves are incredibly electric intensive. One it one industry in Taiwan, the Taiwan Semiconductor Manufacturing Corporation uses one fifth of all of Taiwan's electricity. One one company. So we want we want to replicate that here. Great.

00:43:23:12 - 00:43:44:04

Mark Mills

You go girls, but you don't have to build power plants. This is true in Taiwan too. As so I think I think we're going to be forced back into a reality based energy posture where we'll have a more realistic view of the need for nuclear power. And it's where it's it's manageable. Risks are extraordinarily manageable with well-designed power plants, but that'll take time.

00:43:44:04 - 00:44:06:15

Mark Mills

We're going to we're going to stop closing coal plants and have a different timeline for that. We still think that they should be close. Change the timeline to build enough alternative power. And then we're looking to build a, is the Irish would say a shitload of gas turbines in the in the next five years? That's what that's, that's what's going to happen.

00:44:06:17 - 00:44:17:02

Chris Keefer

Is that starting to happen like in Texas in the Permian where natural gas prices have been negative? They're just trying to deal with that excess production. Okay. Yeah. Yeah. I mean, it's also happening. Right? I guess the fuel concern.

00:44:17:04 - 00:44:40:13

Mark Mills

In Germany, I mean, Germany has just passed a lot of to build a huge flight of gas turbines because they and this is, this is a contemporaneous with they used to get cheap gas from Russia. They still get a lot of gas from Russia. They have to switch their sources of gas. But they're not they're not rushing to build a massive new fleet of wind turbines.

00:44:40:15 - 00:44:42:10

Mark Mills

I mean, this message in this.

00:44:42:12 - 00:44:46:14

Chris Keefer

Don't forget, these are hydrogen ready gas turbines right now.

00:44:46:16 - 00:44:51:05

Mark Mills

You want to take that better? but, you.

00:44:51:05 - 00:44:52:00

Chris Keefer

Know, strictly.

00:44:52:00 - 00:45:18:07

Mark Mills

Speaking, strictly speaking, since words are elastic, they are burning hydrogen because, as you know, natural gas, methane is H four. And so there are more hydrogen atoms in the, not by weight but by count. So for every, every molecule, right, of gas you have, 80% is hydrogen. So technically speaking, as soon as they turn them on, they're burning hydrogen.

00:45:18:09 - 00:45:21:22

Mark Mills

So that's good.

00:45:22:00 - 00:45:31:12

Mark Mills

And that's what they're going through. Does there's no path to make hydrogen as cheap as natural gas. It's all PowerPoint presentations. This. So no one no one is going to want to be.

00:45:31:12 - 00:46:03:02

Chris Keefer

Even on even on Canada, even on Canada's east coast. We won't be able to do it. we've got, we've got a new report on that, on that. I'll send it along to you. you know, in terms of, I just kind of the esthetics of energy and, you know, a lot of the tech pros are talking about small modular reactors, having them physically on site, you know, maybe they're not the ones managing them, but, you know, I guess the temptation for a kind of a behind the meter approach and maybe that has to do with fears of, you know, pitchfork wielding, residents, or that kind of competition for grid

00:46:03:02 - 00:46:18:06

Chris Keefer

power. you know, I'm pretty skeptical about, smaller scale nuclear in general. But do you think that's what's driving some of those, those choices or preferences to, to get behind the, to get behind the meter? yeah. To avoid back.

00:46:18:07 - 00:46:42:16

Mark Mills

Behind the meters. Behind the meter. Stuff's been around for a long time in the industrial sector. And there's, there's sort of a dividing line as soon as, as soon as it's an industry where the, the power plants you need are utility scale. This is true in the refinery industry, chemical industry, then building power plants behind the fence. makes sense because you're just basically operating with a utility, and there's no reason you can't hire the same people utility hires.

00:46:42:18 - 00:47:10:11

Mark Mills

So if you if you need 100MW, 200MW or megawatts of power, and you happen to also need steam, which is true of a lot of industries, makes a lot of sense to be behind the fence, build your own power plant. It makes more sense not to have that hassle because it's another management vertical. So that's why the model you mentioned becomes more popular, where you you provide the site and they an external owner operator, operates the power plant on your site for your exclusive use.

00:47:10:13 - 00:47:30:11

Mark Mills

And that's the conversation that's going on around nukes. I think that'll happen, by the way. I think a, a nuclear owner operator will, successfully very likely, build nuclear power plants for data centers inside the fence line, for a data center. And maybe, you know, two of them when they need one and then sell power back to the grid.

00:47:30:11 - 00:47:52:02

Mark Mills

I mean, who knows when that surprised me? because it's going to. What's which is the right solution will depend totally on what the data center is. Right? Because it may be if you're near a robust grid and you have a partner that can put the power plants inside of what amounts to a distribution substation, why would you put them on your site?

00:47:52:04 - 00:48:19:18

Mark Mills

So it'd be there's gonna be all kinds of models going forward. This is a have been in the past, but I do think there'll be some some new nuclear power plants built, and I think they'll be financed by data centers. I think that's a good thing. We'll get experience. But overall, generally speaking, what we've learned after 100 years of building power plants and electric grids in the Western world, is that the central utility model is overall cheaper, overall more reliable.

00:48:19:20 - 00:48:41:14

Mark Mills

you don't get you don't necessarily get high reliability by distributing power plants. You can it's possible, but on average, not not for really big. We're you're talking about hundreds of gigawatts of power. It's not it's not a good solution. So it'll be a niche solution. You're, skepticism of small nuclear power plants. I used to share, but I changed my mind a little bit.

00:48:41:16 - 00:49:01:02

Mark Mills

But it depends on the timelines. If you want to build nuclear power plants. Now, if governments want more nuclear, that they need to order what we know how to build. Obviously we know how to build gigawatt scale power plants. You they'll have two one gigs and it's there's no mystery to it. China's building them right now. One was just finished.

00:49:01:02 - 00:49:17:21

Mark Mills

The two unit in Finland South Korea. The Russians are building them as you as you may or as you probably know, second biggest exporter of nuclear power plants, Russia, after China, French are going to build a bunch more. We could do the same in Canada, the United States, if you want to build them, put the order in.

00:49:17:21 - 00:49:48:08

Mark Mills

Today. There's designs we know they work. Start building them. It'll be great to get the small reactors to be competitive for their applications. And I think there's markets for thousands of, tens of thousands of small reactors ranging call it ten megawatts to, 100MW. So that's those scales, just as in aviation, the you know, if you look at the aviation industry as a model, it's similar to power plant industry because they build power plants.

00:49:48:10 - 00:50:07:14

Mark Mills

It's the same thing. Like the air derivative turbines on a jet aircraft are exactly the same as the gas turbines. There's our stationary. In principle, what you find is, is a dominant size for an aircraft power plant. it's based on the optimal infrastructure and logistics around that industry, which turns out to be a 25 to 50 megawatt power plant.

00:50:07:14 - 00:50:29:23

Mark Mills

That's what a typical engine is for a seven, three seven. But we make lots of bigger engines for a smaller number of aircraft and lots of smaller engines for smaller aircraft. you're going to, I think in the in the nuclear power world, the optimal I'm going to guess is going to be something like a 500 megawatt unit, 300 or 500 megawatt configured as gigawatt scale reactors.

00:50:30:01 - 00:50:53:07

Mark Mills

and you're going to see that dominate like the seven, three, seven A320s dominate aviation, but there's still a huge market for lots of other applications, and there are no nuclear power plants to speak of that are filling those market niches. So I think that will happen. It'll just take longer because we have to establish the supply chain capabilities to manufacture those things at scale, and no one has manufactured any of them at scale ever.

00:50:53:09 - 00:51:14:17

Mark Mills

It's never does. You talk to engineers, they'll tell you how to do it. Okay, go. But you need money. You need to build a factory. Yes, you need qualified welders, qualified electrician, you know all. But the whole supply chain doesn't exist yet to build at scale. I think it'll come. And I think you're going to see mobile transportable reactors.

00:51:14:19 - 00:51:30:15

Mark Mills

You know, they've been around for a long time. It's not a not a new idea space program needs them. By the way. NASA has a, as a the Da Vinci reactor, which they want to put on Mars. Mars is too far away for solar power to be useful at scale. So you're going to want nukes.

00:51:30:15 - 00:51:52:11

Mark Mills

You're gonna need nuclear power plants to push rockets to Mars fast enough. Chemical rockets aren't going to work. you've got, use in the subsurface and and all oil gas industry, there are certainly lots of applications where you would like to put data centers. To your point, where we going to store data on processed data where it is, it'll be nice and cool, but does not have electricity.

00:51:52:13 - 00:52:13:15

Mark Mills

Be nice to have a power plant that you can turn on and refuel at every ten years. And it's these are very nice features. So I think all that happens. But if you look at the supply chain, the timelines for engineering qualifications, you know, decade plus, but the world two decades from now will look very different in terms of the number of kinds of nuclear reactors.

00:52:13:17 - 00:52:33:01

Mark Mills

But my, my, my admonition would be for the nuclear balls. And you see this a lot in the United States that were some states that banned new nuclear power plants, are passing legislation to allow construction of new modular small reactors. Okay. They they're passing laws allow you to build, but they don't know how to build that worthless law.

00:52:33:03 - 00:52:35:00

Chris Keefer

And then when you build the big stuff in Illinois.

00:52:35:02 - 00:52:36:10

Mark Mills

Yeah, right. It's crazy.

00:52:36:12 - 00:52:55:11

Chris Keefer

Yeah. absolutely. I wanted to chat a little bit about, I guess, some of the geostrategic implications of the cloud and of of AI. you know, there was talk of a missile gap in the, in the 60s and 70s. I don't know how long that went on for, but is there, like, the potential of a compute gap if, we can't keep up in terms of powering or building this, this infrastructure?

00:52:55:13 - 00:52:58:05

Chris Keefer

Or is that not something we need to be so worried about?

00:52:58:07 - 00:53:17:02

Mark Mills

Part of the trope that has been run out is that if we don't build the data centers, AI data centers and power them, that China's willing to power theirs because they're building lots of coal plants and they'll. And there, as you know, they don't have built a lot. They're building a lot. So there's going to be lots of cheap electricity in China for a long time.

00:53:17:04 - 00:53:40:05

Mark Mills

So it's not going to be hard for them to build AI based data centers and power them inexpensive. Li. But you know, it's it's kind of it's a silly argument. We're not going to not build power plants because because of the climate change movement. And so as so wrapped rather axle over building, building power plants that are useful, that is the power data centers, 24 seven.

00:53:40:07 - 00:54:01:05

Mark Mills

We're not going to abandon it. I just don't believe it. It's just not going to happen to me. It's a lot of debate, over the implications of set aside the debate over climate change. I'm just saying the motivation for opposing power plants is climate change. Obviously, people want to avoid building power plants that burn hydrocarbons, okay? You know, build them.

00:54:01:05 - 00:54:19:03

Mark Mills

You're not gonna have enough electricity. I mean, it's just the fact you either build them and have enough electricity or you won't. I don't think United States and Canada are going to, decide a future with not enough electricity is acceptable. It's just not going to happen. It might take a few blackouts in Canada, in the United States to get citizens unhappy about it.

00:54:19:05 - 00:54:45:16

Mark Mills

And then politicians will pay attention quickly, and we'll bill power plants again. So I think if we actually push it to that point, it'll get resolved very quickly. I think it gets resolved before we get the blackouts. I think people don't are afraid enough of blackouts that when grid operators come to a governor or a or a premier and tell them you're going to have blackouts, if you shut that down, the they'll back off because it's not a political statement.

00:54:45:16 - 00:55:13:23

Mark Mills

These are engineers, which is you you're going to you're going to go dark next time as a heat wave. We're going to we're going to be cycling like happened in Alberta recently. Right. It's not going to be, it's not going to be pleasant. So I don't think that's the issue. The different issue is, is there a a race because I said existential threats or benefits between China and the United States to get the AI supremacy and that that language could could China beat us to AI supremacy?

00:55:14:01 - 00:55:35:15

Mark Mills

You know, Chinese engineers are really smart. I went to maybe 12 or 14 cities in China over 2 or 3 visits, not recently, but, you know, about 5 or 6 years ago. Yeah, there's no question. There's the great engineers and they build they build impressive things that not just, you know, commodity stuff. They, they, you know, rockets, their rackets work, the GPS works, they're they're ships work.

00:55:35:15 - 00:56:01:09

Mark Mills

They're airplanes work. They're building Boeing competing airplanes soon. So I think, they're serious competition things, just generally speaking. And this is not meant as a cultural statement. It's just an operational reality. New stuff like AI, when compute is new and technologies are new and different in order to wring out of what its utility is and have industries know how to build those things, nobody knows exactly what to do with it yet.

00:56:01:09 - 00:56:22:19

Mark Mills

It's pretty clear, right? Nobody knows exactly what should be built. And it's not going to just be an Nvidia. There's gonna be lots of there's at least 2000 nonpublic, GPU makers that are making GPUs claim to be probably are better than Nvidia's GPUs. Impressive things that are coming. Almost all of them are here in North America, most of them in the United States.

00:56:22:21 - 00:56:51:08

Mark Mills

not in China. There's a few there, but most of them are here. So we're sort of at, the 1980 and compute. And I say that because in 1980, after President Reagan was elected, the Japanese were going to launch a massive effort to leapfrog the United States to compute now, computers at that time, 1980, we, you know, for those of a certain age, will remember this is four years before the Apple Mac, when the Apple two E had been out.

00:56:51:10 - 00:57:14:00

Mark Mills

so personal computers were, you know, toys and cute. But by and large, a compute world was dominated by mainframes and Japan was going to leapfrog us. And, that's not what happened. What happened is the era of personal computing exploded in the United States because nobody nobody ordered it directed. It wasn't a government program. It was just as a Boolean industry of capital and entrepreneurs.

00:57:14:02 - 00:57:30:05

Mark Mills

AI is going to be exactly the same way what we do with it, what kind of chips, what kind of applications. So we're sort of like, maybe it's not 1980, maybe it's 1984, right? When the first Mac, which looks really clunky in hindsight, if you go back and look at it, but then.

00:57:30:07 - 00:57:31:04

Chris Keefer

Oh yeah, I mean.

00:57:31:06 - 00:57:52:00

Mark Mills

People were blown away. I mean, a little box on your desk, I said it was giant stuff. So we're it's hard to get your head around how credibly exciting that was in 1980, in 1984. So we're excited about ChatGPT and what AI is doing. It's 1984. I mean, we haven't even begun to wring out and innovate in fun.

00:57:52:00 - 00:58:11:03

Mark Mills

These things. Chat is not good at those things. It's good at a lot of things. But that's not it's it's business culture. It's political culture is is not as good as the United States. And I don't mean this is a jingoistic sense or as good as Canada as me. BlackBerry came out of Canada. It's just not as good.

00:58:11:03 - 00:58:40:17

Mark Mills

And it's regressed with the political environment. And in China right now, they've had a regression pulling away from, free market incentives. There. And you really have to function in a sort of quasi chaotic, you know, capitalist risk taking environment to figure out what's going to be next. Something is remarkably different, is I that's a long way of saying I'm not worried at all about losing the AI, supremacy war to China.

00:58:40:19 - 00:58:59:05

Mark Mills

I think that's the only way that will happen is if we Soviet ties our economies here. If we do complete command and control, it's not I mean, it's possible. I mean, I, I don't know, your prime minister kind of kind of sort of seems like he wants to do a lot of command and control. Maybe, maybe Canada will participate.

00:58:59:07 - 00:59:24:05

Mark Mills

And our president is sort of these more towards command and control. I don't know. I think it's unlikely. And I'll think this is where Canada, the United States are very similar. I mean, I worked as a scientist and engineer in both countries, and culturally. People do this kind of innovating in an environment a lot more like we have in North America than exists in China, and even more than less exist in a lot of Europe right now.

00:59:24:05 - 01:00:03:02

Mark Mills

Europe, Europe is is sort of slipping. They have extraordinary engineers in Europe, no question. But, they're, I think making epic mistakes at the moment. They're reversible. They might reverse. I'm not sure what social mistakes and what epic industrial planning mistakes and the more, more, more we drift into industrial planning and the more and the firmer the gets, in the sense that the governments are deciding the path for innovation, the path for investment, the less likely you get real innovation because the people who are doing the planning are, by definition, bureaucrats.

01:00:03:04 - 01:00:12:13

Mark Mills

And I don't mean to insult bureaucrats. I know lots of very capable, bright bureaucrats. But that's not renovation happens. It just isn't. And there's never it's never been the case.

01:00:12:15 - 01:00:39:10

Chris Keefer

The suck in the time we have left. you you're comparing us to sort of being where we were in the 80s, inferring that there is a just an enormous amount of growth potential in the sector. you said before that's, you know, appetite for information is essentially insatiable, unlike the appetite for calories, for instance, you had an interesting, anecdote in one of your recent, papers just walking through, the prefixes of numbers, you know, starting with the kilo in 1795, the mega, 1873.

01:00:39:10 - 01:00:53:14

Chris Keefer

And we're up in the era of the zetta. just give us a sense of the sheer size that we're talking about. And then I really want to chat a little bit about In Silico, talking about some of the novel applications that, this degree of compute and I were going to open up for us.

01:00:53:16 - 01:01:13:03

Mark Mills

You know, you're write a book and an hour and one recent article, I sort of one way to think about how we we we have to measure everything we do. We just humans have we count what we do, right? We count miles of count pounds. We count the count counts as we. We count and we we've had numbers for a long time.

01:01:13:05 - 01:01:37:03

Mark Mills

But it's it was interesting to me when I was doing research for my book, to find that the creation of names for big numbers, because we have to count things, sort of tracks technology progress. And for obvious reasons, you don't need the, prefix million. If you don't make or count a million of anything, if everything you're counting is in thousands for practical reasons, you don't need a bigger name.

01:01:37:05 - 01:02:03:01

Mark Mills

So when I looked across the history of names for big numbers, there was there's a long gap before we started naming really big numbers, numbers for not just billions but trillions and, you know, quadrillions. And of course, the prefixes are these odd, you know, people, no gigabytes, but they have two gigabyte, you know, terabyte, terabyte. And you've got to get them a petabyte and they exabyte, but each increment is a thousand fold, as you know.

01:02:03:03 - 01:02:29:08

Mark Mills

And then after exabyte we that's where we get the zettabyte. And that's a big number. So all the data trackers I throw around that this word we're in the Zettabyte era. by that they mean is around 50 zettabytes of data, moving around the cloud. Collectively, it's managed move the shuffle about and makes it sound small because it's 50, you know, 50, it's 50 zettabytes.

01:02:29:10 - 01:02:48:12

Mark Mills

And we used to be in an era where, you know, you and I remember when people would brag about going from a memory device sort kilobytes to I have a megabyte, you know, then you got a gigabyte storage, like, whoa, that's a billion bytes. Now, you could buy a terabyte flash drives. It's crazy, you know, trillion. But each instrument is big.

01:02:48:12 - 01:03:09:09

Mark Mills

And and the reason I it was interesting to me is that, of course, in the physics of energy, we know a lot about how much energy is associated with each byte. Every byte involves an energy transaction. It's just the law of nature. There's a transaction and energy transaction to to create the byte, and then there's an energy transaction to move it just like people.

01:03:09:11 - 01:03:35:14

Mark Mills

And then there's an energy transaction to store it. But a byte is infinitesimal. You know, you could store a byte can be literally a countable number of electrons. Utterly crazy is tiny tiny volume, tiny amount of energy. But you moving lots of them. So I just did arithmetic with a the obvious. The calculator said, how would I visualize a zettabyte or a 50 zettabyte economy in data terms?

01:03:35:16 - 01:03:56:20

Mark Mills

And of course, the answer is, you know, if you see a stack of bills, dollar bills, you stack up them till it you two, you got, a zettabyte worth it would go to the sun and back, which is 93 million miles away, 700,000 times. So it's a big it's like, wow, I mean, this such a big number.

01:03:56:20 - 01:04:17:21

Mark Mills

How do you visualize a number? And now we're talking about to your point, which, you know, I've said over and over again in my writings, the only product humans are ever created with infinite demand potential is data, because there's no limit to how much we want to know about something or how much we want to count, how much we want to measure, how much, how granular we want to measure it, how much precision we need, the velocity.

01:04:17:23 - 01:04:33:03

Mark Mills

So just no limit is this. It's just purely based on what you can imagine, whereas everything else has a limit. There's only so many pounds of food you can eat. It doesn't matter how what you imagine is, is only so fast you can fly big. You build a house of Prussian. So everything has those limits but information.

01:04:33:05 - 01:05:02:18

Mark Mills

Weirdly, the the actual quantity. So the next number after Zeta thousand Zetas, which obviously is a Yahoo biota. So we're going to be the data byte era before this decade's out for sure. I think we'll get there way before the decade's out. So the international, the Bureau of International Measurements and Names and Measurements came up with new names for a number after that, because what's happened over the centuries is the bureaucrats come up with a name in advance of us needing it, which is pretty useful, actually.

01:05:02:18 - 01:05:28:19

Mark Mills

Have some utility. And after after zettabyte is is a data byte, and after you got a byte will be the random byte ROI na, which is really, As you may recall, I wrote in my book, I was like, my book came out. They hadn't chosen the name yet, and all the computer guys wanted it to be called a bronto byte after the Brontosaurus, because it would sound cool, but the the, the humorless French bureaucrats wouldn't do that.

01:05:28:19 - 01:05:54:03

Mark Mills

So they named it Erano Byte because, you know, it's just the prefix means something in Greek and whatever. I just I was disappointed, but it is fascinating that we just we have to invent numbers, names for numbers to accommodate the magnitude of demand for data is is crazy and numbers that are so big. One other context 50 zettabytes.

01:05:54:05 - 01:06:14:02

Mark Mills

All the molecules of the atmosphere of the earth weigh, I think, three zeta grams o atmosphere of the planet Earth. If you measure it, Zetas waste three zeta grams. I mean, is that is a really big number. So this is why would I say the cloud is the biggest infrastructure ever built by humanity? Yeah, think about it.

01:06:14:02 - 01:06:36:00

Mark Mills

We're we built an infrastructure to move a quantity of of things that are real, that bytes are real things, maybe tiny, but they're real. They have physical existence. The virtual world has physical existence and machinery and and and electrons and charges and we're doing this at the zettabyte level, and we're on our way to a yacht, a byte level.

01:06:36:04 - 01:06:58:22

Mark Mills

This is crazy. I mean, it's it has all kinds of fascinating applications for infrastructure and power and, you know, I'm not talking about, you know, what do you do with it? That's what the book's a lot about. More what do you what do you do with this power? Right. This is this is the part that's fascinating, as far as you know, trying to predict you get that much horsepower in compute sense.

01:06:58:22 - 01:07:17:21

Mark Mills

So what are people going to do with it? I mean, it's it's not like you can predict easily. There'll be all kinds of no one predicted the quote sharing economy in, the you got mail days. Everybody's slobbering about how exciting it was to do email that you could go to a website and do email. Wow, you got mail.

01:07:17:21 - 01:07:39:01

Mark Mills

You got a movie, you got mail. No. And then said, we're going to have the Uber economy and sharing economy and Rover where you can disintermediate service functions and create all kinds of new business models. And once a lot of no one thought about that, the equivalent to that kind of surprise will surely come as we expand the cloud and amplify it with it.

01:07:39:05 - 01:07:56:14

Chris Keefer

They I so so tell us about In Silico. we've heard about in vitro research. you know, this sounds like a novel application that's only unleashed by this. Sure. sheer amount of compute power. I just caught my eye as a as a medical professional. So you've done a bit more research on this. I'd love to hear about it.

01:07:56:16 - 01:08:23:23

Mark Mills

Yeah, I know, as you know, and I was I said this in my book, and I still believe it's true. We're a long, long way away from computing, replicating the the the artistic skill that comes with being a physician. There's because we, we don't know how the human brain works. It's not a computer. It's computer models. It doesn't work like, I in fact, I isn't.

01:08:24:01 - 01:08:46:22

Mark Mills

Artificial intelligence isn't intelligent. It's it's just very it's a very different neural network class of compute that emulates inference. But it's not it's not a human brain. It doesn't think like a brain. Does it work? But doesn't matter. The point, though, is some functions of matter and some features. are reducible to compute questions. This is what true.

01:08:46:22 - 01:09:05:17

Mark Mills

Of course, for some diagnostics, especially, one of the most difficult things, you know, as a physician is if you do things that are repetitive, a lot human beings get into cognitive dissonance just happens, which is why you have a consult with somebody else. You look at this x ray too many times, or MRI, and you have somebody else look at the same thing and they'll say, you missed that.

01:09:05:18 - 01:09:31:01

Mark Mills

And you go, how did that happen? Well, because humans have cognitive dissonance. So machines can help on that. They can actually help you see things that you miss just as a, as a coach consult can, but only if they're astonishingly powerful. I mean, really astonishing the amount of data sets. And they have to operate more like AI because you're not looking for compute, you're looking for inference, you're not looking for one plus one equals two.

01:09:31:03 - 01:10:04:13

Mark Mills

But those two things don't look like they're close to two. They look like they're you know what a it's how automated steering works. It's not on a track. It's this is about right. So as you get a greater ability to gather data. So let's use the human body as the example where we were seeing phenomenal advances in, the size and the precision of sensors to gather data about biological functions, human body functions, not just MRI and Cat scans merge, but a whole constellation of that problem.

01:10:04:13 - 01:10:25:17

Mark Mills

The problem is it's a data store. You're looking for patterns, and it's a data set different than you as a human are used to. I in principle. And the principle is not crazy. You could imagine just AI and compute capacity to hold a massive data set of what we know about you. You're my own body now, I'm not going to know everything about my body, but I can.

01:10:25:17 - 01:10:45:06

Mark Mills

I can begin to imagine collecting not just terabytes, but petabytes worth of data about my personal body. This is not right now. It's too expensive. I want to store petabyte. I get it to a cost that most people would not find acceptable. You store terabytes, you and I constantly because that's what's, you know you have on a flash drive.

01:10:45:07 - 01:11:06:05

Mark Mills

But remember Tera, it's a tiny number when you, you got to go to a thousand fold. So if something has a dollar, $90,000, it's the kind of. But that cost will come down. So as the cost of memory and storage and compute comes down and it's coming down at the same rate, now, as it has been for a decade or two decades or three decades.

01:11:06:07 - 01:11:29:10

Mark Mills

So now I can begin to think about a computer that has mass, a quantity of data, not just about my body, from I've gone to a diagnostic test in a hospital, but I'm collecting real time data. I mean, you have this is a physician. You know, we both subject ourselves to our doctors and you're not feeling well. And the question is, well, you eat anything different in the last few weeks or what did you you know, what did you get?

01:11:29:15 - 01:11:56:18

Mark Mills

We get questions about the past that you can't answer, but it's not crazy to collect that kind of real time data about yourself stored in a secure cloud environment, about yourself, continuous like so that you can have a cloud twin, your body, your information, a sense it's it's not a true twin. But they talk about this as twins so that all the data, not just about a diagnostic test was taken a year ago or ten years ago, or one I took yesterday.

01:11:56:18 - 01:12:20:22

Mark Mills

But all all of the biological data that I'm collecting about myself, including my walking and my steps, my respiration, my sleep, you know, body movements, maxi matter, that is, you know, there's been some very good AI based research that's shown that if I use a motion sensor, I just use the motion sensors, and I watch the change in how you use your arms.

01:12:21:00 - 01:12:39:16

Mark Mills

it's a decent indicator of a certain types of disease onset. And they're very subtle changes in your motor neural functions. But you you don't see them, but they're subtle enough that over time and I can pick them up all those kinds of things going into a computer in silico in silicon means that I can do two things in the future.

01:12:39:19 - 01:12:56:03

Mark Mills

One, sooner than later. One of them would be, well, the answer to the question, who was your doctor? You use your password and so you can take a look at my in silico digital twin and get answers to lots of questions. My blood sugar level two weeks ago. Because you could, you could imagine swallowing a pill every day.

01:12:56:03 - 01:13:17:00

Mark Mills

Your vitamin will have in it a smart sensor that's biodegradable, that will upload data about your GI system daily to your smartphone. Why wouldn't you do that if it was back a day? And it's not that kind of technology? No, this is not speculative fiction. My whole book is about the future is predictable for what's already invented, but not commercialized.

01:13:17:06 - 01:13:45:13

Mark Mills

What I just described already exists. Biodegradable silicon that can do sensing and measuring already exists, already being put to work. So as those things advance, I now have an in silico virtual, digital twin that can answer a lot of the questions about what? About my my state of health in the past that a physician could look at, not in combination with an AI, because the AI is going to look for patterns, changes and patterns and advise the doctor.

01:13:45:14 - 01:14:10:20

Mark Mills

So I, I noticed this did you what? And then you look at the data sets I well it's you know, I hallucinate because it's irrelevant. Or maybe I didn't notice that further down the pike is of course the the holy grail of in silico is to do full drug testing in silico. Because I have I have enough precision about my my bio, my personal, my personal physiology, my personal history.

01:14:10:22 - 01:14:36:02

Mark Mills

You as a physician know that all the off label uses for drugs, are discovered ex post facto. They're discovered because they have beneficial effects nobody thought about. In some cases, they have toxic effects on certain people, but therapeutic effects on others. We don't typically know who it's toxic for, who it's therapeutic for. With sufficient amount of data, one believes.

01:14:36:06 - 01:14:56:01

Mark Mills

And it's not a crazy belief because we have reasonable science on this, that I would be able to tell if a specific therapeutic was toxic or beneficial to you, whether it's for cancer or for a for a virus, and that kind of in silico testing, because the how that virus behaves, how the therapeutic behaves, it can be modeled.

01:14:56:03 - 01:15:21:07

Mark Mills

And because GPUs and AI can model real dynamic things, not static things, we can we can come close to modeling interactions of drugs. And a lot of drug discovery now is amplified by those kinds of modeling. So they're using supercomputers to do drug discovery and modeling in silico before they then go to animal trials. Once you know that's happening and it's expensive today, it's in a supercomputer.

01:15:21:09 - 01:15:48:13

Mark Mills

It doesn't take any any imagination to say, well, okay, that can be democratized. It might take ten years, but it'll get democratized because it will come cheap. Once it's cheap, then we're into a whole different era of, of what? The precision medicine, which has always been, the Holy Grail, but it's really about hyper specificity is, you know, this is what you, you deal with every day is it is what I'm doing right for that patient.

01:15:48:15 - 01:16:11:03

Mark Mills

And if wouldn't it be great if some in silico AI to at least give you a heads up on its risks for that person that you might not have noticed? this, I think, profoundly improve, to put it in called economic terms, the productivity of medicine. What they mean to are bad outcomes. Faster comes. Yeah. But I think that's I think I.

01:16:11:05 - 01:16:31:01

Chris Keefer

Think that's what I, yeah, yeah, yeah. No, I, I talk to my physician colleagues sometimes and they always assume I have a medical podcast. I think this is the most medicine I've spoken in the four years of making this podcast. So I appreciate that. And we're gonna have to park it here. definitely want to, have you back much sooner, than than previous hopefully the next month or two.

01:16:31:03 - 01:16:46:11

Chris Keefer

you're doing a lot of great writing on EVs. there's a lot of talk about, battery overcapacity right now and the falling price of batteries. we did an episode before batteries and energy. Lysenkoism. We've had a lot of fun, Mark. same today. And, looking forward to having you back again soon.

01:16:46:17 - 01:16:49:21

Mark Mills

Be great to come back. We can talk batteries next time. Maybe a lot of fun.

01:16:49:23 - 01:16:50:10

Chris Keefer

Beautiful.