[00:00:00.000] 
Hi, David. Hi, Charles.

[00:00:01.640] 
It's the end of the week.

[00:00:02.630] 
Yes. How are you?

[00:00:03.600] 
I'm very good. Excited for the weekend. Perfect. Got some squash to play.

[00:00:09.340] 
Okay.

[00:00:10.190] 
And then, of course, some shopping.

[00:00:12.500] 
Okay. I'll be skiing with plenty of snow.

[00:00:14.750] 
At Bromo? Yes. Very nice. I went skiing at Sutton last week, and it was beautiful. Good. Spectacular, actually. But we're going to do a behind the headlines.

[00:00:26.380] 
Yeah.

[00:00:27.300] 
And not not obviously related to the supply chain, but we'll get there. But two things happened in the last week. Everybody probably knows about it. Number one, we have this Deep Seq AI that was released by a team out of China. That's shocking the world because their claim, not verified, is that it only costs about six and a half million dollars to build compared to the the hundreds of millions, if not billions, that the American teams have spent. I did read this morning that some people are saying that the 6. 5 million might have been just the training of the model portion. Not and not including all the other costs, which even if that's true, so what? It's still a huge step forward in the AI world.

[00:01:24.700] 
Yeah, and done with chips that are not as performing as those that the OpenAI and others have access to.

[00:01:36.640] 
Because these people would be operating under the restrictions where they have older generation chips to use. And so that's also very cool. And at the same time, we have a new administration down in the US. And one of the things that they've announced is a $500 billion investment Stargate. Stargate, which is Stargate like moonshot. It's a plan. I really don't know what... Tell me about what exactly are they going to do with that money?

[00:02:10.950] 
Well, so it's Japan SoftBank. Then you got all the big ones. Ultimately, it's to build the infrastructure, the AI infrastructure. Great. To support, ultimately, the ability to continue to grow and build and train models and have more capacity.

[00:02:28.970] 
Okay, so So as we say that, the infrastructure, and I'm sure that the average listeners, I know what the infrastructure is, but I think it's worth diving into it. And once you dive into it, you realize what the implications of half a trillion dollars of spending in infrastructure really means. I, up until last year, had never even I thought about the energy consumption of a Google search. Never entered my mind. And I know that in theory, if I was in a test, if they said true or It's not false. A Google search consumes energy. I know the answer is true. I would have thought of it as infinitesimely small, right?

[00:03:24.780] 
Yeah. Which a Google search, a single Google search is probably true, but given the volume.

[00:03:31.780] 
When you take billions of searches a day, that adds up to something. Yeah, correct. But AI, an equivalent search in AI, 10 10 times. 10 times more consumptive of energy. And the reason is that when you query an AI, you're engaging an entire model. Almost like the sum human knowledge is being engaged to answer that query. And maybe when you think of this deep seek, I know one of the innovations is they've partitioned, instead of having one massive expert system, they have partitioned into two layers of expertise, one layer that knows what expert to consult in their expert systems, and then another layer of individual expert systems, which is one the way they got around to make the model more efficient.

[00:04:33.270] 
But are you referring to the part where some claim that ultimately it use OpenAI to, let's say, add the final touches to its ability?

[00:04:47.060] 
I'm not making that claim. I have no idea what I'm talking about. This is really shooting from the hip. But what I understand is one of the efficiencies that they have created is there's two layers to the model where one layer takes your query, figures out what parts of the expert systems it needs to engage to get the answer, and therefore leaves the other parts of the... If your question's about the weather, it's not going to engage the part that's arts and culture, the experts. So it leaves that dormant while... So it's attacking a more limited set of knowledge with any question it answers. I don't know. That's what I read. Whether it's true or not, we're just warehouse designers. We don't know anything, right?

[00:05:37.850] 
Yes.

[00:05:38.440] 
But we do know this. Here's a fact that I astonished me. So a Google search, we just said, okay, it's basically almost no energy, but it is energy. Multiply that by a billion searches a day or whatever the number is, and that's quite a whack of energy. And then you just said, if If Google search went away and then we replace it with AI, that would 10X the energy. And as models grow, at least in the way we've constructed them so far, that will grow as the models grow. But maybe deep seek will give us strategies to reduce that. Generating an image, and I just did one yesterday just for fun. Pretty good, too. I went on and build a presentation and It created an image for me. And it was the first time I've ever done that. And then I find out today that I was just in my reading that that image generation consumes the same energy it takes to charge your smartphone.

[00:06:47.480] 
For a full charge. For a full charge. That's insane.

[00:06:50.000] 
Insane. Here's what's insane. Now I feel really guilty. It generated the first image. I didn't like it. I said, okay, I said, it's natural. I said, pretty good, but can you? And then didn't like second version. It was only the third version. I said, close enough, I'll work with that. So I've charged three smartphones in a matter of 90 seconds. Insane. Insane.

[00:07:17.650] 
Yeah. And there's a... It's nearly impossible to... Well, it's not nearly. It is impossible to, as we continue to leverage these advancement or day-to-day life or whatever, where it's being used to ask people to moderate themselves and be cognizant of that consumption. Because last year was the first year in my 20 years of searching, not necessarily Google, but before Google was Ask Cheeves and NetSuite.

[00:07:51.980] 
No, not NetSuite. Netsuite. Netsuite is much, much more sophisticated than NetScape. So we've We've been googling well before Google existed even and never thought about it until last year. So now, are we really going to stop ourselves from making images now that we can? Of course not. The applications for people like us, for every student, for the world of just being able to create images, it's an irresistible advance in our ability to communicate with one another. So it's not going to happen. We're going to be charging smartphones like crazy. Now, that's to sensitize us to the fact that our own individual actions in interacting with AI at its most basic, nascent, infant level, which is where it is today, consumes a whole lot of energy per task. I read another statistic that I find amazing. If you take a little, a next step and you go to the training of an AI model. They estimate, one analyst estimated, was the equivalent of watching Netflix nonstop streaming Netflix on your TV nonstop for 183 years.

[00:09:22.120] 
Yeah.

[00:09:23.090] 
Just imagine how many times you would have watched every episode of Breaking Bad if you're streaming NetSuite for 183 years nonstop, 24/7, not even allowed to sleep. And so that's another fun way of understanding what power hogs these things are and making the average citizen and our colleagues in the industry really aware of the end, not at the end, at the very up at the very beginning of the energy supply chain of AI, it's a massive beast. So I was reading another thing, if we go a step further. Okay, so we got training models is an energy hog, making images, even just doing little queries, all of this is consuming way more energy than we think. Right now, I think it's the International Energy Agency out of the UN calculated, they project by 2027, two years from now, the entire global consumption of energy just for AI, just AI, will be the equivalent to the state of Michigan's total energy consumption. I mean, it's crazy. The state of Michigan, 10 million people, D Detroit, cars, all that good. Amazing ice cream out of Grand Rapids, Michigan. But that's not a very energy-efficient plant, I would point out.

[00:11:11.210] 
But other than that, the state of Michigan, imagine every power plant in Michigan, and all it's doing is running AI applications for the world. And this brings us to, where is all this energy getting consumed?

[00:11:27.830] 
Well, yeah, in data data centers. Data centers, which this data centers exist for forever, it's just that now they need to-add AI applications. Well, their capacity has to augment in ways that are, I wouldn't say exponential, but accelerate the ability to treat all that.

[00:11:53.410] 
Well, again, so everyone knows what a data center is, but when we're talking about what What outside of AI are data centers doing?

[00:12:03.910] 
Well, they're used right now for hosting application, cloud-based. Any organization that needs to run applications on the cloud, well, that's the cloud ultimately.

[00:12:18.170] 
Like Netflix. We just said 183 years of Netflix watching, right? Where's all the Netflix content sitting? In fact, I nearly said NetSuite again. Where's all NetSuite's hosting happening. All hosting the internet is sitting in data centers. I mean, that's it. So here's a statistic. Right today, So data centers are, as we just said, it's basically the Internet, it's hosting, it's all that, plus now AI. And we said that the global consumption of AI, of energy for AI is about equivalent to the state of Michigan. The total energy consumption of the world's data centers currently is equal to...

[00:13:11.810] 
Canada's consumption.

[00:13:12.810] 
The energy consumption of a G7 top 10 economy country, Canada.

[00:13:18.060] 
Every piece, every what being consumed.

[00:13:21.510] 
And we, per capita, consume enormous amounts of energy in Canada, mostly the heater houses. But that's one of the And we're sparsely populated. So transportation, both personal and commercial and industrial, there are much longer trips involved. So we are relatively big energy consumers. Our entire nation's energy consumption is equal to all the world's data centers. That should- Is insane.

[00:13:55.560] 
Right.

[00:13:56.060] 
And it gets worse, David. It gets worse. It gets worse By 2027, we're expecting that global energy consumption for all the world's data centers to be equal not to Canada's energy consumption, but who?

[00:14:12.180] 
Japan's.

[00:14:12.910] 
With three times, well, maybe not three, but two and a half times our population. Again, the number three or four or five, a top five economy with more than 100 million people. And that is going to be two years from now.

[00:14:29.350] 
And that That increase is driven by the AI.

[00:14:32.700] 
Right. So if the entire country of Japan floated away and resettled on Mars, we would still need a Japan-like production powerhouse to keep our data center is running. And then if we go 2035.

[00:14:51.040] 
Ten years.

[00:14:51.630] 
10 years, what are they projecting?

[00:14:54.550] 
Well, the US.

[00:14:55.960] 
Half the US.

[00:14:57.070] 
Oh, I'm sorry. Half the US.

[00:14:59.490] 
Which is a mind boggling number. Imagine, half the US. So the implications are pretty significant. The energy consumption part of it is straightforward because if we haven't made the point by now, you're not listening. We hammered it on your heads. But there's going to be a desperate need to increase energy consumption and therefore energy production to fuel all the growing need for data centers.

[00:15:33.120] 
And just a note, as we were talking before this, when you think about energy requirements to run data centers, a significant part of it, or we believe a significant part of it is to cool those data centers because these generate vast amounts of heat.

[00:15:50.700] 
Cooling is a huge part.

[00:15:52.450] 
With Jeremy making some research, as much as half-half the energy in a data center is just cooling the data center. Yeah. And then so why are all data centers not located in Canada or Minnesota? Or in the Antarctica. Yeah, exactly. Is because there's a proximity element to it as well.

[00:16:11.740] 
So there is a need for some data center applications require proximity to the user or the consumer of that data center. Yeah. Well, I mean, it's crazy. So I guess, I'll tell you a little anecdote. One of our clients was looking to expand a freezer, an industrial freezer, and cannot get the power, can't get the power. The city will not commit to providing them the power they need to expand their freezer. Imagine that company's business is hugely and directly immediately impacted by the fact that all levels of government are prioritizing energy production to go to data centers and not to other industrial activities. And my worry, I think the other headlines recently have demonstrated, as we're all caught flat-footed on this, no one has been... And I'm not saying that critically. I had no idea, so who am I But we're caught flat-footed about how much energy AI requires and how much of a competitor against other industrial uses of energy it is becoming, and that every country, United States, Canada, the North American continent, has to have well thought out and not too slowly implemented plans for increasing power supply across all sources. I'll give you a statistic, but I will admit, I asked Gemini AI for this, so who knows?

[00:18:07.490] 
Sometimes it does get questions wrong. But we have something like three to five times more hydroelectric power potential in Canada than we are currently exploiting.

[00:18:21.510] 
Well, yeah, that's amazing. And I heard also on a podcast that we both enjoy is it's not so much about a race to green energy. It's a race to- All energy.

[00:18:36.460] 
All energy. Yeah. With a preference for renewable energy. But there's also green or carbonless non-renewable energy, such as nuclear. Microsoft, if you think about Three Mile Island, which is famous for a partial meltdown that happened in 1978, '79, something like that, which scared the whole world away from nuclear power. I mean, Chernobyl and Fukushima were actually even worse. But Three Mile Island was the first scare. And that happened. There were two units, and Unit One- It melted. Had a partial meltdown.

[00:19:14.840] 
And unit two kept running, and they shut it down in 2019.

[00:19:21.000] 
You can tell me- Well, so then Microsoft has now made an offer to buy this a nuclear plant in order to use it fully and that the entire energy coming produce out of it would be used for its own data centers.

[00:19:39.440] 
Yeah, purpose.

[00:19:40.110] 
And just to put in perspective, that unit one produces the equivalent of of the quid of energy. Power, 800,000 homes every year. That's what Microsoft said, Oh, I'll sign that up. I'll make a 20-year commitment to buy all the power generated to power data centers. So that's a example of how tech companies are thinking about this and they're pushing their government. I mean, it's no accident that inauguration day, you've got all those techies. And before you think conspiracy or anything like that, sometimes the most obvious answer is the answer, which is there is a desperate need to get the energy problem solved in order to maintain a position of, if not dominance, at least prominence in the AI world because the technology is so transformative for everyone. So the energy is one of the issues. The other thing that I think people don't... Well, We talk about it sometimes in news, but for supply chain, it's very important, is all of the data centers of the world and all of the end applications where AI will find itself in are all We're all going to need some industrial metals that are fairly widely distributed, like copper, nickel, aluminum, and silicon, some precious platinum group metals, which are, again, maybe not...

[00:21:19.480] 
They're precious. That doesn't mean they're totally rare. They're fairly available. But then there are these strategic commodities like cobalt and gallium and lithium and the rare earths that we absolutely are in a scramble to secure stable supply and create supply chains where the relative, it's called the the spheres of AI power, they want their own supply chain. So you go from a Google search, which you'd never thought about the energy consumption. You go to AI, suddenly you You realize we're almost way behind where we need to be for the energy and data center infrastructure. And then you realize, in order for all this to happen, I need more dams and more nuclear power plants. That's one thing. I also need More chips. More chips. And the chips need very strategic commodities to make.

[00:22:27.080] 
Because you refer to the data centers, but also it's... Because the AI will not be only... What am I trying to say? That not only data centers will be the area, the place where- Where those chips need to sit? Yeah, or that the intelligence need to run. Bringing the ability to compute some form of information will be brought to a very local piece, whether it's your fridge or your phone or your whatever part of- Any place where you consider automating is a place where you're going to need these strategic minerals.

[00:23:20.180] 
It's not just about queries and searches. Exactly. We go forever on where AI goes next and how you deploy that power. But then this podcast would take three hours. Neither of us are Joe Rogan.

[00:23:38.660] 
Or capable of. Yeah, exactly.

[00:23:41.110] 
It's very funny, though. Anyway, sometimes I will see with Joe, and it's always about aliens and pyramids. How anyone could be so fast, grown adults.

[00:23:53.240] 
Obsessed with.

[00:23:55.070] 
Aliens. Anyway, or sorry, this is a podcast about artificial intelligence, and he was just doing a podcast on non-human intelligence. We don't say aliens. We now call them non-human intelligence. It makes people think it's a more credible thing. I can't wait for two hours of Mermeade talk. So I guess for me, when I look at it, the knock-on effects of this scramble for resources, the scramble for energy is a complete... I'm lost for words.

[00:24:37.600] 
Not reinvention of supply chain, but putting in place the ability, the capacity to have access to these minerals and quantities that have not been levels that we haven't seen yet.

[00:24:53.830] 
Yeah. And the imperative and the importance and the The level of concern that different governments are putting on this scramble for a secure AI supply chain is upending trade relationships all over the world. And if you don't think they have in certain corners of the world, it's coming.

[00:25:27.380] 
Yeah. And not to turn this podcast into a political one, but it can be seen as an arms race, right?

[00:25:36.830] 
Absolutely. Yeah, I think it's absolutely an arms race.

[00:25:41.190] 
Because China has not held back on their ability to build energy.

[00:25:50.330] 
In AI infrastructure, they are well ahead of everyone. They're building coal plants. I know it's It's fashion. We'll say, well, they build all these solar panels, so they're green. They're building more coal power production than the rest of the world combined. And they know they can do another great leap forward using coal power to power their AI. And hopefully we won't go down quite that path. And that we'll find other solutions. It's nice to know Canada has the third largest uranium reserves in the world. And I think I did it. I don't know. There's a calculation. I did. Well, I did. Now I'm going to say I do it, but really, I get Gemini to do it for me. But I wrote the question, so I did the calculation. But just about how much energy the known reserves of uranium in Canada represents. And it's quite a happy number. It's an impressive number.

[00:27:02.770] 
I'll just go back to the, call it the headline, which you started talking about the Deep Seek, because that's what has been the headline this week. And with Deep Seq, I read somewhere it says, The US innovates, China imitates, and Europe or the EU regulates. But here with Deep Seat, what was not shocking, but what made the industry I think lose something like trillions of dollars, which some of it came back was, well, wait, there's innovation here, regardless how it was done, and regardless of the cost as well, you could think, well, China has the advantage on the energy, US has the advantage on the innovation, but China has demonstrated that it is highly capable as well.

[00:27:54.640] 
Anyway, it's very exciting stuff. It's very interesting stuff. I I hope we can solve the supply chain issues around making AI happen, the physical real infrastructure without putting in jeopardy the rest of our economy, which does include freezers and factories and all sorts of other uses of energy. Because if we can solve this, we have all the resources. North America has all the resources it needs to do it and do it clean and beautifully. So let's just...

[00:28:31.770] 
Which makes me think my own... I wasn't using Gemini. I think I was actually using Deep Seq to talk about, to inquire about this topic. And We see there's going to be a huge, as we've discussed, plenty about the need for energy, but the technology also will improve in itself in terms of efficiency, right? Yeah.

[00:28:58.680] 
It'll get more efficient.

[00:28:59.830] 
It It will get more efficient. Training models and how it's where data lies and where the processing is happening will also make it... There are real initiatives so that these models do not consume us on a per unit basis as much.

[00:29:21.370] 
How do some of those folks in Silicon Valley say, technology acts like a gas? It fills. Right. If I make it more efficient, I'm just going to use more of it.