WEBVTT
00:00:00.080 --> 00:00:04.160
If the demand doesn't materialize, Anthropic is still on the hook for the bill.
00:00:04.240 --> 00:00:10.960
It's less like a growth plan and more like a massive mortgage they have to service, regardless of whether the customers show up.
00:00:11.199 --> 00:00:13.759
I think now it's sixteen thousand dollars, right?
00:00:13.919 --> 00:00:17.199
The price of a car for RTX Pro 6000.
00:00:17.519 --> 00:00:23.280
What world are we living in here where a card for a computer that's supposed to run games costs more than a car?
00:00:23.600 --> 00:00:32.399
When you're paying sixteen thousand dollars a card and servicing a$500 billion compute bill, the pressure to choose subtraction is immense.
00:00:32.560 --> 00:00:42.079
The compute cost is so high that the only way to make the math work might be to cut the payroll, which turns the engine of growth into a machine for efficiency.
00:00:42.479 --> 00:00:44.479
All right, folks, we're back.
00:00:46.240 --> 00:00:46.560
Mr.
00:00:46.799 --> 00:00:48.240
Human in the loop here.
00:00:48.799 --> 00:00:51.920
So two moves for this one.
00:00:53.119 --> 00:01:05.840
One, you want to put your power bill from last year the same month to this year the same month.
00:01:06.799 --> 00:01:14.159
And you wanna ensure that it's actually the same, right?
00:01:14.640 --> 00:01:16.480
Now, could you have changed your behavior?
00:01:16.719 --> 00:01:17.200
Sure.
00:01:17.519 --> 00:01:31.760
But part of that bill potentially is based on the charges for upgrading the grid relative to AI and data center build out showing up on there.
00:01:31.920 --> 00:01:32.560
Yeah.
00:01:32.879 --> 00:01:56.560
And two, and this one matters, for the decision makers out there in the world, before you spend a few million on AI and then cut, you know, some millions in people's salary, ask yourself what could that few million build with those people instead?
00:01:58.000 --> 00:02:07.760
And just a few words on the elections that took place in the US a couple of days ago.
00:02:09.599 --> 00:02:11.680
Democrats, don't mess this up.
00:02:12.479 --> 00:02:13.120
Yeah?
00:02:14.080 --> 00:02:21.680
I already saw the governor of the great state of Texas trying to halt data center build out.
00:02:23.120 --> 00:02:24.719
So I guess we'll see.
00:02:26.159 --> 00:02:29.039
Now let's get into the headlines for this one.
00:02:29.360 --> 00:02:30.960
This is gonna be interesting.
00:02:31.199 --> 00:02:37.759
What I'm trying to argue against here is that the data center bust, right?
00:02:37.840 --> 00:02:44.879
If the data centers go bust, we as workers feel the pain.
00:02:45.520 --> 00:02:46.479
Let's get into this.
00:02:46.639 --> 00:02:47.199
Ainsley.
00:02:48.000 --> 00:02:57.280
Microsoft has walked away from an AI data center outside London because grid officials say the power may not arrive until the 2030s.
00:02:57.439 --> 00:02:59.120
That was October 19th.
00:02:59.360 --> 00:03:02.639
Microsoft says it relates to a single site.
00:03:03.520 --> 00:03:09.199
On October 26th, Anthropic, maker of the CLOD models, filed to go public.
00:03:09.680 --> 00:03:17.120
Its filing says it must pay for computing capacity whether or not it uses it, and leaves the total blank.
00:03:17.680 --> 00:03:25.199
On October 28th, Alphabet reported more than$51 billion spent on equipment in one quarter.
00:03:25.360 --> 00:03:27.280
Its shares then fell about 10%.
00:03:28.560 --> 00:03:31.360
The chips are ready, the power lines are not.
00:03:31.520 --> 00:03:34.479
So when the grid says wait, who pays?
00:03:34.719 --> 00:03:38.560
The companies, the workers, or us on our power bills.
00:03:39.919 --> 00:03:47.919
So here's where we get into the segment we're calling the Carlo gut check.
00:03:48.080 --> 00:03:48.719
Yeah.
00:03:49.039 --> 00:03:51.120
So I'm not gonna hold you folks.
00:03:51.520 --> 00:03:56.960
Um it's hard to kind of grapple with a lot of this stuff, right?
00:03:57.280 --> 00:04:03.039
Because ultimately it all funnels back to the human, it all f funnels back to the individual.
00:04:03.759 --> 00:04:04.319
Yeah.
00:04:04.800 --> 00:04:11.520
The individual votes, the individual uh contributes in the economy, they purchase things and so on.
00:04:11.759 --> 00:04:12.159
Yeah.
00:04:12.319 --> 00:04:15.919
It all fun funnels back to us, right?
00:04:16.480 --> 00:04:31.439
And what my gut is telling me is that if capital doesn't have a place to live, right, the humans that are, you know, individuals in this world is gonna take a haircut.
00:04:32.720 --> 00:04:33.199
Right?
00:04:33.759 --> 00:04:37.439
Because they wanna point capital at a certain place, right?
00:04:38.240 --> 00:04:47.920
Uh you see that happening with the AI build out, you see that happening with chips and purchases and data center build-out, all of it, right?
00:04:48.800 --> 00:05:01.040
And the problem is that the friction between the capital going to work, right, is causing problems, right?
00:05:01.120 --> 00:05:12.879
It's causing problems for the upside of what that is, which is the revenue f flowing back into the investors and the VCs and everyone, right?
00:05:12.959 --> 00:05:15.279
That's looking to capture the value from that.
00:05:15.759 --> 00:05:19.600
So that's the thing we're gonna argue about today.
00:05:19.839 --> 00:05:31.279
We're gonna argue about like whether or not this infrastructure build-out, if it gets halted in any shape or form, is that good or bad for the economy?
00:05:32.240 --> 00:05:36.240
Listen, I don't think we're gonna do anything about AI.
00:05:36.399 --> 00:05:36.879
Yeah?
00:05:37.199 --> 00:05:39.920
AI is inevitable.
00:05:40.639 --> 00:05:41.279
Yeah.
00:05:41.600 --> 00:05:48.000
I don't know if you ever got you you guys ever watched uh Star Trek programs from me yesterday, yeah.
00:05:48.319 --> 00:05:53.040
But um assimilation is futile, folks.
00:05:53.199 --> 00:05:53.680
Yeah?
00:05:53.839 --> 00:06:03.360
That's what uh Borg used to say to the normal folk is that it's futile to run against the program.
00:06:04.079 --> 00:06:16.800
This is happening, and the problem is that if we create friction and all of that, prices go up, revenue doesn't work, the system doesn't work the way it should, right?
00:06:17.439 --> 00:06:30.000
Um, I already see uh the governor, the new governor of the great state of Texas, suggesting that they want to halt data center build out to do studies and do all of these things.
00:06:30.240 --> 00:06:32.879
That sounds like jobs for a bunch of people that do studies.
00:06:33.040 --> 00:06:38.000
That sounds like jobs for a whole bunch of administrators, but those are temporary.
00:06:38.240 --> 00:06:41.360
Yeah, you don't do studies for a lifetime.
00:06:41.920 --> 00:06:44.480
Okay, you're standing in the way of progress.
00:06:44.560 --> 00:06:45.920
That's really what you're doing.
00:06:46.160 --> 00:06:53.600
And I don't want to come on here and sound like some other podcast that you guys never heard of, but the problem is is this right?
00:06:54.240 --> 00:06:59.439
If something is inevitable, yeah, somebody holding it back.
00:07:33.199 --> 00:07:35.120
It's already decided, Ainsley.
00:07:35.680 --> 00:07:36.160
Yeah?
00:07:36.720 --> 00:07:40.160
Humans in society get stuck with a bill.
00:07:40.399 --> 00:07:42.240
It's a trickle-down effect, right?
00:07:42.319 --> 00:07:44.480
That's what they mean by trickle down, okay?
00:07:44.800 --> 00:07:46.639
So let me frame this up for you.
00:07:47.519 --> 00:07:55.199
Um essentially, data centers get built out, uh, power bills go up.
00:07:56.000 --> 00:07:56.959
Who's paying?
00:07:57.360 --> 00:07:58.639
The individual.
00:07:58.959 --> 00:08:02.240
Um then they have services, okay?
00:08:02.480 --> 00:08:10.240
Services on top of that data center, whatever it is, uh open weight model, closed weight model, whatever is running on that thing, okay?
00:08:10.399 --> 00:08:15.680
And they point that at uh enterprises, and enterprises start using that thing, right?
00:08:15.839 --> 00:08:22.800
The cost of them using it gets passed on to whoever is demanding the services for what they're running in there.
00:08:23.199 --> 00:08:23.759
Right?
00:08:24.079 --> 00:08:30.240
And if you think about the services, the services is pretty vast, okay?
00:08:30.560 --> 00:08:35.759
Just imagine every organization using some sort of service that relates to AI on there.
00:08:36.080 --> 00:08:48.639
It doesn't matter, matter if it's Netflix, doesn't matter if it's uh TikTok, whatever the service that you can imagine could be running some processes on some data center somewhere.
00:08:48.960 --> 00:08:49.279
Okay?
00:08:49.679 --> 00:08:53.919
And then ultimately, what that turns into is pricing.
00:08:55.039 --> 00:09:12.320
The amount of money they pay for the compute power in there turns into the amount of money that they charge for someone to use that facility, which turns into what they have to charge you to utilize said service as an individual.
00:09:12.639 --> 00:09:16.080
So it all trickles down back to the individual who pays.
00:09:16.320 --> 00:09:19.039
And that's really what I'm here to argue about.
00:09:19.279 --> 00:09:24.399
I'm here to argue about the idea that someone ultimately has to pay.
00:09:24.720 --> 00:09:34.559
And so far as it goes with society, the people that's paying is the people that's consuming service and products from organizations.
00:09:34.879 --> 00:09:38.639
Organizations are set up specifically to do a thing.
00:09:38.879 --> 00:09:43.200
I don't know care if it's B2B or B2C, doesn't matter what it is.
00:09:43.440 --> 00:09:52.240
Ultimately, somebody at the end of the road, end of the chain, has to pay for for some sort of bill that comes due.
00:09:52.399 --> 00:09:54.720
And usually it's the individual.
00:09:55.039 --> 00:09:59.360
Your light bill, yeah, your eggs, okay?
00:09:59.600 --> 00:10:01.039
All of that stuff.
00:10:01.200 --> 00:10:01.519
Right?
00:10:01.759 --> 00:10:09.840
Technology is behind all of it, and they charge you prices based on how high their prices are.
00:10:10.159 --> 00:10:13.840
They trickle it down, right, to the individual.
00:10:15.279 --> 00:10:19.840
That's a clean chain, but it assumes the cause is just the pass-through.
00:10:20.080 --> 00:10:25.679
The real question is whether this is a standard cost of doing business or a hidden tax.
00:10:26.080 --> 00:10:38.960
If the grid upgrades are funded by the ratepayer before the first chip even turns on, then the individual is paying for the infrastructure twice, once on the power bill and again in the price of the service.
00:10:44.159 --> 00:10:46.639
Um yeah, you're paying twice.
00:10:46.879 --> 00:10:48.720
Somebody pays, right?
00:10:49.039 --> 00:11:04.399
So you're paying for the equipment up uh uh the grid upgrades, and then you're also paying for the access to the models if you're using uh closed weight models out there, chat GTP and uh anthropic claude, shout out.
00:11:04.559 --> 00:11:07.039
Um you're paying, you're paying twice, right?
00:11:07.200 --> 00:11:21.360
So you're paying your$20 a month, whatever plan, for your open AI subscription, and additionally, you're paying for the utilities that it takes to have that data center in your geographic.
00:11:21.840 --> 00:11:22.159
Okay.
00:11:22.879 --> 00:11:27.279
So somebody has to pay, and the people that's paying right now is the individuals.
00:11:27.600 --> 00:11:35.759
And the only way that I could clearly showcase that is to talk about this idea around the energy tax, right?
00:11:36.320 --> 00:11:53.200
So for me specifically, in PJM uh region, yeah, I looked at my bill from last year, in uh whatever month, it doesn't matter, to this year, whatever month, and I saw a difference.
00:11:54.000 --> 00:11:56.720
Many people is gonna see the same thing.
00:11:57.120 --> 00:11:59.039
So I'm paying for that already.
00:11:59.279 --> 00:12:06.159
Plus, I'm paying for the Claude Pan, the whatever plan, the additional AI plans that I'm paying for.
00:12:06.480 --> 00:12:06.799
Okay.
00:12:07.840 --> 00:12:10.879
So there is a cost to all of this, right?
00:12:11.200 --> 00:12:25.440
I think it's$9.3 billion in one year for capacity uh costs that PJM is putting on um the residents in their geographic.
00:12:26.159 --> 00:12:28.559
That's not going away, right?
00:12:28.960 --> 00:12:34.320
Even if you stop the data center from coming in, those upgrades are happening.
00:12:35.360 --> 00:12:37.919
That's where the corporate spin comes in.
00:12:38.080 --> 00:12:47.840
One Rutger study found that the effect of a data center opening on bills in its own zip code was weak, mixed, and small.
00:12:48.080 --> 00:12:54.240
Now, the author says that's because the costs are spread regionally, which supports your point.
00:12:54.639 --> 00:13:05.440
But it creates a perfect gap for companies to claim that the local impact is negligible while the bill is quietly distributed across millions of households.
00:13:08.399 --> 00:13:09.279
So I get that.
00:13:09.440 --> 00:13:10.639
I get that frame, right?
00:13:10.799 --> 00:13:12.879
One, I think maybe their studies old.
00:13:12.960 --> 00:13:15.120
And then the second point is this, right?
00:13:15.360 --> 00:13:24.559
You can't look for the energy increases in the specific geographic that you're talking about, because where they're building these data centers is usually in the middle of nowhere, right?
00:13:24.799 --> 00:13:37.200
And how it goes, at least what I've seen so far in the data, is that you know, essentially the markets that's in that region, they get different prices based on where you are.
00:13:37.360 --> 00:13:41.679
Like New Jersey will get a price, New York will get a price, Connecticut will get a price.
00:13:41.840 --> 00:13:47.120
Wherever you are, you get a different price uh per kilowatt hour, right?
00:13:47.519 --> 00:13:55.039
Um, and then the point is that it gets balance across the whole field based on what you normally charge.
00:13:55.200 --> 00:14:04.480
So if your bill in wherever the hell goes up by 0.2%, yeah, you're not gonna really see it.
00:14:04.720 --> 00:14:06.639
Ruckters is not gonna see it.
00:14:06.879 --> 00:14:14.399
But if you're in an area like why where I am in New Jersey, and it went up by 10%, I am gonna see it.
00:14:15.360 --> 00:14:18.159
That's exactly the blind spot in the records data.
00:14:18.320 --> 00:14:27.919
They're looking at the host zip code, but PJM's own market monitor says data center demand pushed wholesale power costs about 9% higher this year.
00:14:28.080 --> 00:14:35.919
When you spread that across the whole region, it's a massive sum, about$9.3 billion added to one year's capacity costs.
00:14:36.080 --> 00:14:39.120
But it's diluted enough that a local study misses it.
00:14:39.279 --> 00:14:43.600
It reminds me of the Samuel Insul utility empire back in the 30s.
00:14:43.759 --> 00:14:48.559
He built this massive layered pyramid of holding companies to control the grid.
00:14:48.720 --> 00:14:55.440
And for a while, the efficiency looked great on paper, but the whole thing was built on a mountain of debt that eventually collapsed.
00:14:55.519 --> 00:14:59.360
And it was the small investors, not the architects, who lost everything.
00:14:59.519 --> 00:15:00.720
The pattern is the same.
00:15:00.960 --> 00:15:05.919
The people at the top build the infrastructure, but the people at the bottom carry the risk.
00:15:06.480 --> 00:15:08.879
Now, why does that even matter, right?
00:15:09.440 --> 00:15:11.279
Here's why it matters, right?
00:15:11.919 --> 00:15:16.399
Because I could go up in arms and say, hey, my bill just went up by 10%.
00:15:16.960 --> 00:15:19.440
I don't want no more data centers around here.
00:15:19.679 --> 00:15:24.080
They won't build it around here if we get enough people to say they won't build it around here.
00:15:24.320 --> 00:15:29.679
But they'll build it somewhere else in the PJM footprint because somebody agreed to do it.
00:15:29.919 --> 00:15:30.399
Right?
00:15:30.559 --> 00:15:34.480
And the reason why is because data center is looking for an outlet plug.
00:15:34.639 --> 00:15:34.960
Okay?
00:15:35.279 --> 00:15:37.360
They're looking for an outlet plug.
00:15:37.600 --> 00:15:48.240
You just saw Ainsley read the news quite well there about Microsoft pulling out of uh Essex because they couldn't find the outlet plug.
00:15:48.879 --> 00:15:51.519
So data centers are looking for an outlet plug.
00:15:51.600 --> 00:16:01.759
If they can't find it in my town because somebody railed against them, they're gonna go plug it in somewhere else because the power is available here in this region.
00:16:02.559 --> 00:16:02.879
Okay?
00:16:03.919 --> 00:16:06.000
Now, why does that matter?
00:16:06.240 --> 00:16:10.480
You know, like I said, capital is being pointed at data centers.
00:16:10.720 --> 00:16:17.039
Even the story that we told about uh uh Microsoft pulling out of Essex is a problem.
00:16:17.759 --> 00:16:27.519
Because my framing that I said at the top was that essentially, if you stop building the data center, the people that's gonna get hurt is the individual.
00:16:27.679 --> 00:16:37.679
Because now all of those uh electricians, HVAC workers, all the workers that was pointed at this data center now is like, what do I do now?
00:16:37.919 --> 00:16:47.440
They have to go find and realign themselves in that economy without that data center paying them for four years or whatever the amount of years they it takes to build it up.
00:16:47.840 --> 00:16:48.159
Okay.
00:16:49.679 --> 00:16:51.600
So it is a problem.
00:16:52.000 --> 00:16:54.399
And that's why I said Democrats get it right.
00:16:54.559 --> 00:17:07.119
Because if we don't get it right, essentially people could be looking around for stuff to do, not having enough economic value, okay, to pour back into the economy for the whole thing to actually work.
00:17:09.039 --> 00:17:15.279
I see the risk to the trades, but we have to ask if those jobs are a lifeline or a trap.
00:17:15.599 --> 00:17:26.079
Indeed, found that data center installers' posted hourly pay is about 42% higher than similar jobs, which is a massive premium.
00:17:26.319 --> 00:17:37.759
But if that's a bubble fueled by a desperate rush for outlet plugs, those workers are essentially being paid a premium to build a system that might be over-leveraged.
00:17:38.000 --> 00:17:41.839
If the build-out hits a wall, they aren't just losing a job.
00:17:42.000 --> 00:17:47.279
They're stranded with a skill set tied to a specific crashing boom.
00:17:48.000 --> 00:17:59.839
So I'm glad you said that, Ainsley, because the thing that I'm gonna bring up next is this idea around where's the demand coming from the data center build out?
00:18:00.160 --> 00:18:00.720
Yeah?
00:18:00.960 --> 00:18:03.599
Where's all of this demand coming from?
00:18:04.079 --> 00:18:21.920
And the thing I want to point you at is the$518 billion bet that Anthropic has relative to their S1 that um is out there on the interweb being mused around.
00:18:22.160 --> 00:18:22.480
Okay?
00:18:23.200 --> 00:18:38.640
So Anthropic S1 suggested that they are looking for about$518 billion worth of compute over the next 10 years, and 80% of that is non-cancilable.
00:18:39.279 --> 00:18:39.359
Okay.
00:18:40.480 --> 00:18:44.720
So that's just one company looking for that much amount of compute.
00:18:44.880 --> 00:18:46.000
Do we have it?
00:18:46.559 --> 00:18:51.200
I don't think we have that much uh compute, so the only way to get it is to build it, right?
00:18:52.000 --> 00:19:08.880
And you know, further in there, S1, if you keep peering into it, essentially uh the customer base of uh Anthropic um is rather small compared to the market that they could engage in.
00:19:09.119 --> 00:19:14.559
They probably needed that data center in London to add some customers from over there.
00:19:15.119 --> 00:19:22.640
But now they don't have it because it can't it can't happen until 10 years from now because of the plug missing from the data center.
00:19:25.599 --> 00:19:31.359
The scale of that bet is staggering, but we have to be careful about where that number comes from.
00:19:31.599 --> 00:19:41.359
Reuters reported from a draft of the filing that at least$518 billion is committed over a decade, with about 80% being non-cancelable.
00:19:41.599 --> 00:19:45.119
The public filing actually leaves the total blank.
00:19:45.279 --> 00:19:47.599
But the risk is exactly what you're pointing to.
00:19:47.759 --> 00:19:52.240
If the demand doesn't materialize, Anthropic is still on the hook for the bill.
00:19:52.480 --> 00:20:00.160
It's less like a growth plan and more like a massive mortgage they have to service, regardless of whether the customers show up.
00:20:01.200 --> 00:20:03.759
So what's your read on this, yeah, Ainsley?
00:20:04.480 --> 00:20:07.519
I mean, my read is that Anthropic gets your money back.
00:20:07.680 --> 00:20:08.079
Yeah?
00:20:08.319 --> 00:20:14.960
Anthropic buys the compute, whatever 80% is, and everything is fine.
00:20:15.200 --> 00:20:18.400
Because they really haven't been installed into everywhere yet.
00:20:18.640 --> 00:20:36.799
And ultimately, whatever funding they're trying to raise is gonna give them the um capability to add more customers, okay, to do more things, to make the model more capable, give them more compute to get to more AI capability.
00:20:37.359 --> 00:20:37.839
Right?
00:20:38.079 --> 00:20:48.799
Essentially, their uh S1 and their them going public is only gonna help them, right, to get to the number that they're looking for as far as revenue and so on.
00:20:49.039 --> 00:20:49.359
Okay.
00:21:33.200 --> 00:21:36.000
Will they actually get their money back on that, right?
00:21:36.559 --> 00:21:45.839
But I think the thing that you have to look at it is the reality of what they're actually seeing on the revenue side, not on the expense side.
00:21:46.000 --> 00:21:46.559
Yeah?
00:21:47.359 --> 00:21:51.440
Their first half revenue was sixteen point three billion.
00:21:51.920 --> 00:21:56.480
About fourteen times a year earlier.
00:21:56.799 --> 00:21:57.119
Okay?
00:21:57.680 --> 00:21:59.519
So 14 times.
00:22:00.160 --> 00:22:04.960
The amount of money they made as of last year, they're making this year.
00:22:05.200 --> 00:22:07.440
So they have a specific run rate.
00:22:07.599 --> 00:22:10.079
The trajectory look almost vertical.
00:22:10.400 --> 00:22:10.960
Right.
00:22:11.279 --> 00:22:14.480
So unless you tell me that that's gonna stop.
00:22:15.359 --> 00:22:19.680
Okay, that they're not gonna double over the next year, right?
00:22:20.079 --> 00:22:22.720
Then potentially they don't need as much compute.
00:22:22.799 --> 00:22:25.839
But from what we're seeing right now, their trajectory is takeoff.
00:22:28.000 --> 00:22:31.920
The trajectory is vertical, sure, but that's the danger of a takeoff.
00:22:32.079 --> 00:22:35.759
You don't know if you're climbing or just accelerating toward a wall.
00:22:35.920 --> 00:22:45.599
If they're scaling revenue 14 times a year, they're in a race to find a market that can actually absorb that much compute before the non-cancelable bills come due.
00:22:45.839 --> 00:22:54.559
The question is whether they're building a sustainable business or just scaling a cost center that requires exponential growth just to stay solvent.
00:22:55.599 --> 00:22:56.880
And that's a good read, right?
00:22:56.960 --> 00:23:02.960
Because I think I read that there one place that uh Anthropic is starting to hire less juniors, which is crazy.
00:23:03.119 --> 00:23:06.240
But I mean, they're an AI company, so what do you what do you want me to tell you?
00:23:06.319 --> 00:23:06.640
Okay.
00:23:07.039 --> 00:23:12.400
You're trying to go work for an AI company and you're gonna think that they're gonna hire um junior workers there.
00:23:12.480 --> 00:23:13.599
They're starting to hire less, right?
00:23:13.680 --> 00:23:16.319
I think it went from something like 27% to 9%.
00:23:16.960 --> 00:23:21.359
Um, Adrian, you could go look that up, but it is what it is, okay?
00:23:21.680 --> 00:23:27.920
And the next thing that I think we'll get into is, and it flows nicely into this, right?
00:23:28.400 --> 00:23:39.920
Is that the reason why it's$518 billion and it's a bet on the on the anthropic side is that the cost of things is the cost of things, right?
00:23:40.559 --> 00:23:54.559
So right now, the only game that we have in town is NVIDIA, who's ultimately setting the price or most of the price from the AI um compute infrastructure, right?
00:23:55.039 --> 00:24:00.559
Right, 23,000 chips Microsoft wanted to plug in into SX London, you can't do it.
00:24:00.799 --> 00:24:01.039
Right?
00:24:01.119 --> 00:24:03.279
They'll probably put it somewhere else, it's fine.
00:24:03.920 --> 00:24:16.319
But if you're gonna pay for$518 billion over the next 10 years of compute, okay, the compute needs to be there.
00:24:17.119 --> 00:24:19.599
Otherwise, the money is not gonna get spent.
00:24:20.319 --> 00:24:20.960
Right?
00:24:21.519 --> 00:24:31.279
I I I don't know what happens in the in in the regard of if they can't find$518 billion worth of compute, Ainsley.
00:24:31.599 --> 00:24:36.480
Let me know what what'll happen if that goes uh go if it goes like that.
00:24:36.799 --> 00:24:42.640
But they have to find$518 billion worth of compute in the next 10 years.
00:24:42.880 --> 00:24:48.880
And if the writings on the wall is going the way it is right now, maybe they can't find it, right?
00:24:48.960 --> 00:24:51.519
Because they're not the only company looking for compute.
00:24:51.759 --> 00:24:55.519
There's a lot of companies right now looking for compute.
00:24:55.839 --> 00:24:58.400
I will put myself into that camp.
00:24:58.640 --> 00:25:00.799
I'm looking for compute myself.
00:25:01.039 --> 00:25:03.440
I'm trying to find some NVIDIA card out there.
00:25:03.519 --> 00:25:07.279
I'm looking at AMD and say, how how capable are you, AMD?
00:25:07.519 --> 00:25:09.599
Because your prices are a little better.
00:25:09.759 --> 00:25:10.079
Right?
00:25:10.240 --> 00:25:11.039
I can't find it.
00:25:11.200 --> 00:25:12.799
I can't find compute.
00:25:12.960 --> 00:25:21.440
Uh I went and looked around for RTX Pro 6000 the other day, and um, I think when they came out the price was uh$8,500.
00:25:21.759 --> 00:25:24.799
I think now it's$16,000, right?
00:25:24.960 --> 00:25:29.039
The price of a car for RTX Pro 6000.
00:25:29.519 --> 00:25:31.279
What world are we living in here?
00:25:31.440 --> 00:25:35.920
Where a card for a computer that's supposed to run games costs more than a car.
00:25:36.160 --> 00:25:36.720
Yeah?
00:25:36.880 --> 00:25:44.559
That's why I'm happy about uh Alphabet getting into this race with the TPU, and I think AMD is starting to make some waves too.
00:25:44.720 --> 00:25:56.079
Um probably not as splashy as everybody else, but the more competition we have to NVIDIA, I think the better the market will look, right?
00:25:56.559 --> 00:26:06.240
And potentially, if we get more efficient chips, yeah, we end up with a circumstance where we might not need as much data center.
00:26:06.559 --> 00:26:11.599
The problem is Anthropic needs the amount of compute that they need.
00:26:12.000 --> 00:26:26.079
And the prices right now for the compute that they need,$518 billion, is being set by the price that it costs to have NVIDIA chips installed in all of these data centers to do their work.
00:26:27.759 --> 00:26:37.279
The RTX Pro 6000 listing around$16,000 is a wild signal, and it shows just how dominant NVIDIA is right now.
00:26:37.440 --> 00:26:46.559
By our math from their results, NVIDIA kept about two-thirds of last quarter's revenue as operating profit, which is a staggering margin.
00:26:46.799 --> 00:26:52.880
You're right that Google selling its own TPUs to select customers could be a check on that.
00:26:53.039 --> 00:27:01.200
But the real question is whether that competition actually lowers the price of compute or just moves the margin from one giant to another.
00:27:01.440 --> 00:27:10.720
It feels like the early days of the railroad booms, where the people selling the picks and shovels made all the money, regardless of whether the trains ever actually ran.
00:27:11.119 --> 00:27:20.079
In the 1840s, the promoters and the ironworks owners got rich, while the people betting on the actual routes were the ones who got crushed.
00:27:20.240 --> 00:27:21.519
We're seeing the same thing.
00:27:21.759 --> 00:27:25.759
The chipmakers are the only ones with a guaranteed win.
00:27:35.759 --> 00:27:41.599
Yeah, putting out there how many millionaires he made with the NVIDIA employees.
00:27:41.759 --> 00:27:42.319
Yeah.
00:27:42.480 --> 00:27:47.839
A lot of them are getting stock options, a lot of them are getting highly paid, all of this, right?
00:27:48.000 --> 00:27:52.000
And they're still capturing on a dollar.
00:27:52.799 --> 00:27:54.079
I think it is, right?
00:27:54.400 --> 00:27:59.359
For every dollar they spend, they get 60 cents.
00:27:59.759 --> 00:28:07.279
Now, I mean, Google joined in a race and potentially, you know, there's some lowing of the price.
00:28:07.599 --> 00:28:08.160
Maybe.
00:28:08.400 --> 00:28:16.480
It's a big giant maybe because AMD have been around for a while saying that they're doing AI chips and all the likes, and it hasn't really changed the price.
00:28:16.640 --> 00:28:16.880
Right?
00:28:17.039 --> 00:28:26.640
I went go, I went searching for AMD chips to see if they're as capable as NVIDIA with software and all of the things around it, and I couldn't find it.
00:28:27.039 --> 00:28:32.400
But ultimately they'll get they'll get there, and competition is what uh drives the price down.
00:28:32.640 --> 00:28:36.000
Because without competition, you get sixteen thousand dollar cards.
00:28:36.319 --> 00:28:36.720
Right?
00:28:36.880 --> 00:28:39.359
And then if you need two of them, that's$32,000.
00:28:39.599 --> 00:28:40.480
Who's paying for that?
00:28:40.960 --> 00:28:42.480
Not not the individual.
00:28:42.720 --> 00:28:44.319
Those are going to corporations.
00:28:44.480 --> 00:28:46.799
Now you can see the problem here, folks.
00:28:47.119 --> 00:28:47.599
Right?
00:28:48.000 --> 00:28:49.359
These things are very high.
00:28:49.920 --> 00:28:57.440
The people setting the cost for what they're paying for AI is setting it pretty high because they need 60% return.
00:28:58.720 --> 00:28:59.200
Right?
00:28:59.599 --> 00:29:01.680
Can they get can they get a little bit less?
00:29:02.480 --> 00:29:03.200
I think so.
00:29:03.839 --> 00:29:05.119
What about 40 cents?
00:29:05.359 --> 00:29:06.960
You know, what about 20 cents?
00:29:07.920 --> 00:29:08.400
Right?
00:29:08.640 --> 00:29:10.480
That to me is but is still good.
00:29:10.559 --> 00:29:11.920
You're still in the green.
00:29:12.400 --> 00:29:12.720
Okay.
00:29:13.839 --> 00:29:20.640
Um, but like I said, the stallout of all of these data centers, that's the thing that we're arguing about.
00:29:20.799 --> 00:29:38.400
And this idea around pricing, this idea around power, this idea about people who want it in their backyard, all of it is friction to create a thing that I think is could be the engine for growth for all the sectors of the economy.
00:29:39.039 --> 00:29:41.839
And I'm not saying every job will survive.
00:29:42.079 --> 00:29:49.839
What I'm saying is that you'll have a job, right, if we start growing and scaling businesses and creating new products.
00:29:50.000 --> 00:29:57.920
That's really the framework that I'm using right now is growing, scale, and making new products by utilizing AI.
00:30:00.079 --> 00:30:05.920
The reason you utilize AI in that framework is to multiply the output of the people you have left.
00:30:06.079 --> 00:30:07.440
But there's a tension there.
00:30:07.599 --> 00:30:11.519
If a company uses AI to create a new product, that's growth.
00:30:11.759 --> 00:30:16.640
If they use it to do the same amount of work with half the people, that's just subtraction.
00:30:16.880 --> 00:30:26.000
When you're paying$16,000 a card and servicing a$500 billion compute bill, the pressure to choose subtraction is immense.
00:30:26.240 --> 00:30:36.160
The compute cost is so high that the only way to make the math work might be to cut the payroll, which turns the engine of growth into a machine for efficiency.
00:30:39.519 --> 00:30:42.480
Yeah, I mean, efficiency is gonna come no matter what, right?
00:30:42.559 --> 00:30:44.160
You're always looking for efficiency.
00:30:44.319 --> 00:30:54.240
Automation, even before AI, machine learning, computer vision, you name it, all of these stuff, stuff were getting pointed at the human job, right?
00:30:54.319 --> 00:30:59.200
The assembly line, robotics in the car industry and manufacturing, right?
00:30:59.440 --> 00:31:04.640
They had automated manufacturing way before AI was existing, right?
00:31:05.039 --> 00:31:13.839
You ever seen those uh shows, the factory or whatever the hell they call it that's on the um uh the Discovery Channel or wherever it is.
00:31:14.160 --> 00:31:22.640
But essentially, you look at a factory, they put in some input in the factory, raw materials, and then outcome the other side is a can of beans.
00:31:22.880 --> 00:31:29.279
Okay, that was all automated, and all that has to happen is people gotta maintain the machinery around it, right?
00:31:29.519 --> 00:31:33.920
If the machinery goes down, they start losing money um pretty fast.
00:31:34.400 --> 00:31:47.039
But that stuff always existed, okay, but now it's showing up in the digital economy with the uh white-collar workers, so that's why we see it more prevalent now.
00:31:47.599 --> 00:32:00.400
And the massive sizes of these data centers are pretty prevalent also, and where they're trying to put them is pretty prevalent also because remember, the data center needs an outlet plug.
00:32:00.640 --> 00:32:12.319
So the reason why they're picking towns and going surfing to find where can they plug this thing in is because wherever they're going, potentially they have power to plug it in.
00:32:12.640 --> 00:32:13.200
Right?
00:32:13.519 --> 00:32:22.559
If we malign them to the areas where we think they should go, it's gonna be a harder um place for them to get to.
00:32:22.720 --> 00:32:23.200
Right?
00:32:23.440 --> 00:32:25.519
And what that what does that mean exactly?
00:32:25.680 --> 00:32:31.759
Yeah, you send somebody into the desert somewhere and say, hey, go put your data center over there, and there's nobody over here.
00:32:32.000 --> 00:32:37.839
They have to get power lines there, they have to get fiber there, they have to get the right cooling more, right?
00:32:38.240 --> 00:32:48.319
Um the price per token, the price for AI, the price to have that build-in sit in wherever it is that you're amenable to is gonna be more.
00:32:48.799 --> 00:32:53.759
And the more is gonna show up at the individual's door.
00:32:54.160 --> 00:32:54.640
Yeah?
00:32:54.880 --> 00:32:56.400
The individual's door.
00:32:56.640 --> 00:33:01.279
And sure, you could say that, oh, I don't use AI, so it's not my door is showing up at.
00:33:01.440 --> 00:33:15.920
But no, your company that you're working for maybe using AI, they're gonna have to go pay this bill for the AI that they're using, and they're saying, like, we gotta find this money somewhere to pay the AI bill, so let me let go a couple of workers.
00:33:16.400 --> 00:33:16.799
Right?
00:33:16.960 --> 00:33:19.599
So everything kind of works in a circle, right?
00:33:19.839 --> 00:33:24.160
I had this episode on here before that's called a circle of life, right?
00:33:24.319 --> 00:33:26.559
And it was really around the economy.
00:33:26.880 --> 00:33:31.359
And essentially what I was trying to say is that everything is connected.
00:33:31.519 --> 00:33:35.359
We all nodes in this whole system, we're connected.
00:33:35.599 --> 00:33:44.319
And the moment we decide that our the moment we forget that we're connected, right, that's when everything goes awry.
00:33:45.279 --> 00:33:45.920
Right?
00:33:46.319 --> 00:33:48.079
It really does come full circle.
00:33:48.319 --> 00:33:57.680
We start with the ratepayer paying for the grid, then the company paying the chipmaker, and finally the worker paying with their job to cover the cost of the compute.
00:33:57.920 --> 00:34:02.319
It's a closed loop where the risk is shifted downward at every single step.
00:34:02.559 --> 00:34:15.599
The only way to break that circle is if the AI actually creates new industries and new types of work that didn't exist before, rather than just being a more expensive way to do the same work with fewer people.
00:34:15.840 --> 00:34:24.880
But as long as the cost of the outlet plug is this high, the incentive will always be to subtract humans to balance the ledger.
00:34:26.000 --> 00:34:30.239
Unless we do it in the way that I've been advocating for, Ainsley.
00:34:30.800 --> 00:34:34.239
Unless we do it with the human in the loop.
00:34:34.639 --> 00:34:38.559
These new industries are not gonna just pop out of nowhere, okay?
00:34:38.880 --> 00:34:41.760
People are gonna figure these things out.
00:34:42.159 --> 00:34:50.880
They're gonna figure out the problems in society that they're living through, right, in order to create the new thing.
00:34:51.440 --> 00:34:57.760
AI is not gonna just magically pick it out of the air and say, here, here's a new thing for you humans.
00:34:58.000 --> 00:34:59.920
It's not gonna work that way, right?
00:35:00.159 --> 00:35:08.159
That's why that innovation episode we did back uh way back is um pretty important because all of the innovation that you could think of, right?
00:35:08.320 --> 00:35:16.880
You have a PhD level co-pilot next to you at work, right, could help you create the next thing that you're looking for.
00:35:17.119 --> 00:35:28.320
It could help you create the next thing that you're looking for, new, completely new industries that nobody even thought about, completely new inventions, completely new products, completely new everything.
00:35:28.559 --> 00:35:28.960
Right?
00:35:29.119 --> 00:35:37.920
And the only way that works is if the technology is available to the people that's gonna be able to create those things.
00:35:38.159 --> 00:35:38.559
Right?
00:35:38.800 --> 00:35:43.360
Whichever way it is, through their job, at home, whatever.
00:35:43.599 --> 00:35:43.840
Right?
00:35:44.000 --> 00:35:51.280
If my Claude plan that I have right now goes up to 40 bucks, I might be getting rid of it.
00:35:51.840 --> 00:35:52.159
Okay.
00:35:53.199 --> 00:35:54.159
You see what I mean?
00:35:54.239 --> 00:35:56.880
And that's that's the same story for everybody, right?
00:35:57.119 --> 00:36:06.320
The more you charge for a thing, right, the less likely you're gonna have people paying for it, especially in this inflated environment that we live in right now.
00:36:06.639 --> 00:36:06.960
Okay?
00:36:07.679 --> 00:36:10.719
So this one was pretty long.
00:36:10.960 --> 00:36:12.960
Um, I don't know if I sold it.
00:36:13.119 --> 00:36:25.760
I don't know if I completely sold the people on this idea that stopping the data center build out, the infrastructure build-out, is actually gonna be bad for employees' employment, right?
00:36:26.079 --> 00:36:33.039
There's one category of job that we could almost certainly say is bad for, and it's the people that's building the data centers.
00:36:33.199 --> 00:36:33.360
Right?
00:36:33.519 --> 00:36:38.719
Sure, they could find another project to build, but immediately the one that they were pointing at is not there anymore.
00:36:38.800 --> 00:36:42.239
And we could see that with what's happening with SX and Microsoft.
00:36:42.480 --> 00:36:45.360
So, Ainsley, close us out.
00:36:45.679 --> 00:36:48.079
Hit us with your words of wisdom.
00:36:48.719 --> 00:36:57.039
If you're looking at your power bill and seeing the cost of a build-out you didn't ask for, remember that the infrastructure is rarely the problem.
00:36:57.199 --> 00:36:59.119
It's the debt used to build it.
00:36:59.280 --> 00:37:08.000
Whether it's a railroad in the 1840s or a data center in 2026, the people who build the tracks usually get paid.
00:37:08.159 --> 00:37:13.760
But the people who carry the mortgage are the ones who have to hope the trains actually arrive.
00:37:15.039 --> 00:37:16.960
All right, folks, thanks for joining us.
00:37:17.119 --> 00:37:18.079
See you next time.
00:37:18.320 --> 00:37:19.039
Peace out.
00:37:19.599 --> 00:37:20.559
Take care, everyone.
00:37:20.719 --> 00:37:21.760
We'll be back soon.
00:00:00.080 --> 00:00:04.160
If the demand doesn't materialize, Anthropic is still on the hook for the bill.
00:00:04.240 --> 00:00:10.960
It's less like a growth plan and more like a massive mortgage they have to service, regardless of whether the customers show up.
00:00:11.199 --> 00:00:13.759
I think now it's sixteen thousand dollars, right?
00:00:13.919 --> 00:00:17.199
The price of a car for RTX Pro 6000.
00:00:17.519 --> 00:00:23.280
What world are we living in here where a card for a computer that's supposed to run games costs more than a car?
00:00:23.600 --> 00:00:32.399
When you're paying sixteen thousand dollars a card and servicing a$500 billion compute bill, the pressure to choose subtraction is immense.
00:00:32.560 --> 00:00:42.079
The compute cost is so high that the only way to make the math work might be to cut the payroll, which turns the engine of growth into a machine for efficiency.
00:00:42.479 --> 00:00:44.479
All right, folks, we're back.
00:00:46.240 --> 00:00:46.560
Mr.
00:00:46.799 --> 00:00:48.240
Human in the loop here.
00:00:48.799 --> 00:00:51.920
So two moves for this one.
00:00:53.119 --> 00:01:05.840
One, you want to put your power bill from last year the same month to this year the same month.
00:01:06.799 --> 00:01:14.159
And you wanna ensure that it's actually the same, right?
00:01:14.640 --> 00:01:16.480
Now, could you have changed your behavior?
00:01:16.719 --> 00:01:17.200
Sure.
00:01:17.519 --> 00:01:31.760
But part of that bill potentially is based on the charges for upgrading the grid relative to AI and data center build out showing up on there.
00:01:31.920 --> 00:01:32.560
Yeah.
00:01:32.879 --> 00:01:56.560
And two, and this one matters, for the decision makers out there in the world, before you spend a few million on AI and then cut, you know, some millions in people's salary, ask yourself what could that few million build with those people instead?
00:01:58.000 --> 00:02:07.760
And just a few words on the elections that took place in the US a couple of days ago.
00:02:09.599 --> 00:02:11.680
Democrats, don't mess this up.
00:02:12.479 --> 00:02:13.120
Yeah?
00:02:14.080 --> 00:02:21.680
I already saw the governor of the great state of Texas trying to halt data center build out.
00:02:23.120 --> 00:02:24.719
So I guess we'll see.
00:02:26.159 --> 00:02:29.039
Now let's get into the headlines for this one.
00:02:29.360 --> 00:02:30.960
This is gonna be interesting.
00:02:31.199 --> 00:02:37.759
What I'm trying to argue against here is that the data center bust, right?
00:02:37.840 --> 00:02:44.879
If the data centers go bust, we as workers feel the pain.
00:02:45.520 --> 00:02:46.479
Let's get into this.
00:02:46.639 --> 00:02:47.199
Ainsley.
00:02:48.000 --> 00:02:57.280
Microsoft has walked away from an AI data center outside London because grid officials say the power may not arrive until the 2030s.
00:02:57.439 --> 00:02:59.120
That was October 19th.
00:02:59.360 --> 00:03:02.639
Microsoft says it relates to a single site.
00:03:03.520 --> 00:03:09.199
On October 26th, Anthropic, maker of the CLOD models, filed to go public.
00:03:09.680 --> 00:03:17.120
Its filing says it must pay for computing capacity whether or not it uses it, and leaves the total blank.
00:03:17.680 --> 00:03:25.199
On October 28th, Alphabet reported more than$51 billion spent on equipment in one quarter.
00:03:25.360 --> 00:03:27.280
Its shares then fell about 10%.
00:03:28.560 --> 00:03:31.360
The chips are ready, the power lines are not.
00:03:31.520 --> 00:03:34.479
So when the grid says wait, who pays?
00:03:34.719 --> 00:03:38.560
The companies, the workers, or us on our power bills.
00:03:39.919 --> 00:03:47.919
So here's where we get into the segment we're calling the Carlo gut check.
00:03:48.080 --> 00:03:48.719
Yeah.
00:03:49.039 --> 00:03:51.120
So I'm not gonna hold you folks.
00:03:51.520 --> 00:03:56.960
Um it's hard to kind of grapple with a lot of this stuff, right?
00:03:57.280 --> 00:04:03.039
Because ultimately it all funnels back to the human, it all f funnels back to the individual.
00:04:03.759 --> 00:04:04.319
Yeah.
00:04:04.800 --> 00:04:11.520
The individual votes, the individual uh contributes in the economy, they purchase things and so on.
00:04:11.759 --> 00:04:12.159
Yeah.
00:04:12.319 --> 00:04:15.919
It all fun funnels back to us, right?
00:04:16.480 --> 00:04:31.439
And what my gut is telling me is that if capital doesn't have a place to live, right, the humans that are, you know, individuals in this world is gonna take a haircut.
00:04:32.720 --> 00:04:33.199
Right?
00:04:33.759 --> 00:04:37.439
Because they wanna point capital at a certain place, right?
00:04:38.240 --> 00:04:47.920
Uh you see that happening with the AI build out, you see that happening with chips and purchases and data center build-out, all of it, right?
00:04:48.800 --> 00:05:01.040
And the problem is that the friction between the capital going to work, right, is causing problems, right?
00:05:01.120 --> 00:05:12.879
It's causing problems for the upside of what that is, which is the revenue f flowing back into the investors and the VCs and everyone, right?
00:05:12.959 --> 00:05:15.279
That's looking to capture the value from that.
00:05:15.759 --> 00:05:19.600
So that's the thing we're gonna argue about today.
00:05:19.839 --> 00:05:31.279
We're gonna argue about like whether or not this infrastructure build-out, if it gets halted in any shape or form, is that good or bad for the economy?
00:05:32.240 --> 00:05:36.240
Listen, I don't think we're gonna do anything about AI.
00:05:36.399 --> 00:05:36.879
Yeah?
00:05:37.199 --> 00:05:39.920
AI is inevitable.
00:05:40.639 --> 00:05:41.279
Yeah.
00:05:41.600 --> 00:05:48.000
I don't know if you ever got you you guys ever watched uh Star Trek programs from me yesterday, yeah.
00:05:48.319 --> 00:05:53.040
But um assimilation is futile, folks.
00:05:53.199 --> 00:05:53.680
Yeah?
00:05:53.839 --> 00:06:03.360
That's what uh Borg used to say to the normal folk is that it's futile to run against the program.
00:06:04.079 --> 00:06:16.800
This is happening, and the problem is that if we create friction and all of that, prices go up, revenue doesn't work, the system doesn't work the way it should, right?
00:06:17.439 --> 00:06:30.000
Um, I already see uh the governor, the new governor of the great state of Texas, suggesting that they want to halt data center build out to do studies and do all of these things.
00:06:30.240 --> 00:06:32.879
That sounds like jobs for a bunch of people that do studies.
00:06:33.040 --> 00:06:38.000
That sounds like jobs for a whole bunch of administrators, but those are temporary.
00:06:38.240 --> 00:06:41.360
Yeah, you don't do studies for a lifetime.
00:06:41.920 --> 00:06:44.480
Okay, you're standing in the way of progress.
00:06:44.560 --> 00:06:45.920
That's really what you're doing.
00:06:46.160 --> 00:06:53.600
And I don't want to come on here and sound like some other podcast that you guys never heard of, but the problem is is this right?
00:06:54.240 --> 00:06:59.439
If something is inevitable, yeah, somebody holding it back.
00:07:33.199 --> 00:07:35.120
It's already decided, Ainsley.
00:07:35.680 --> 00:07:36.160
Yeah?
00:07:36.720 --> 00:07:40.160
Humans in society get stuck with a bill.
00:07:40.399 --> 00:07:42.240
It's a trickle-down effect, right?
00:07:42.319 --> 00:07:44.480
That's what they mean by trickle down, okay?
00:07:44.800 --> 00:07:46.639
So let me frame this up for you.
00:07:47.519 --> 00:07:55.199
Um essentially, data centers get built out, uh, power bills go up.
00:07:56.000 --> 00:07:56.959
Who's paying?
00:07:57.360 --> 00:07:58.639
The individual.
00:07:58.959 --> 00:08:02.240
Um then they have services, okay?
00:08:02.480 --> 00:08:10.240
Services on top of that data center, whatever it is, uh open weight model, closed weight model, whatever is running on that thing, okay?
00:08:10.399 --> 00:08:15.680
And they point that at uh enterprises, and enterprises start using that thing, right?
00:08:15.839 --> 00:08:22.800
The cost of them using it gets passed on to whoever is demanding the services for what they're running in there.
00:08:23.199 --> 00:08:23.759
Right?
00:08:24.079 --> 00:08:30.240
And if you think about the services, the services is pretty vast, okay?
00:08:30.560 --> 00:08:35.759
Just imagine every organization using some sort of service that relates to AI on there.
00:08:36.080 --> 00:08:48.639
It doesn't matter, matter if it's Netflix, doesn't matter if it's uh TikTok, whatever the service that you can imagine could be running some processes on some data center somewhere.
00:08:48.960 --> 00:08:49.279
Okay?
00:08:49.679 --> 00:08:53.919
And then ultimately, what that turns into is pricing.
00:08:55.039 --> 00:09:12.320
The amount of money they pay for the compute power in there turns into the amount of money that they charge for someone to use that facility, which turns into what they have to charge you to utilize said service as an individual.
00:09:12.639 --> 00:09:16.080
So it all trickles down back to the individual who pays.
00:09:16.320 --> 00:09:19.039
And that's really what I'm here to argue about.
00:09:19.279 --> 00:09:24.399
I'm here to argue about the idea that someone ultimately has to pay.
00:09:24.720 --> 00:09:34.559
And so far as it goes with society, the people that's paying is the people that's consuming service and products from organizations.
00:09:34.879 --> 00:09:38.639
Organizations are set up specifically to do a thing.
00:09:38.879 --> 00:09:43.200
I don't know care if it's B2B or B2C, doesn't matter what it is.
00:09:43.440 --> 00:09:52.240
Ultimately, somebody at the end of the road, end of the chain, has to pay for for some sort of bill that comes due.
00:09:52.399 --> 00:09:54.720
And usually it's the individual.
00:09:55.039 --> 00:09:59.360
Your light bill, yeah, your eggs, okay?
00:09:59.600 --> 00:10:01.039
All of that stuff.
00:10:01.200 --> 00:10:01.519
Right?
00:10:01.759 --> 00:10:09.840
Technology is behind all of it, and they charge you prices based on how high their prices are.
00:10:10.159 --> 00:10:13.840
They trickle it down, right, to the individual.
00:10:15.279 --> 00:10:19.840
That's a clean chain, but it assumes the cause is just the pass-through.
00:10:20.080 --> 00:10:25.679
The real question is whether this is a standard cost of doing business or a hidden tax.
00:10:26.080 --> 00:10:38.960
If the grid upgrades are funded by the ratepayer before the first chip even turns on, then the individual is paying for the infrastructure twice, once on the power bill and again in the price of the service.
00:10:44.159 --> 00:10:46.639
Um yeah, you're paying twice.
00:10:46.879 --> 00:10:48.720
Somebody pays, right?
00:10:49.039 --> 00:11:04.399
So you're paying for the equipment up uh uh the grid upgrades, and then you're also paying for the access to the models if you're using uh closed weight models out there, chat GTP and uh anthropic claude, shout out.
00:11:04.559 --> 00:11:07.039
Um you're paying, you're paying twice, right?
00:11:07.200 --> 00:11:21.360
So you're paying your$20 a month, whatever plan, for your open AI subscription, and additionally, you're paying for the utilities that it takes to have that data center in your geographic.
00:11:21.840 --> 00:11:22.159
Okay.
00:11:22.879 --> 00:11:27.279
So somebody has to pay, and the people that's paying right now is the individuals.
00:11:27.600 --> 00:11:35.759
And the only way that I could clearly showcase that is to talk about this idea around the energy tax, right?
00:11:36.320 --> 00:11:53.200
So for me specifically, in PJM uh region, yeah, I looked at my bill from last year, in uh whatever month, it doesn't matter, to this year, whatever month, and I saw a difference.
00:11:54.000 --> 00:11:56.720
Many people is gonna see the same thing.
00:11:57.120 --> 00:11:59.039
So I'm paying for that already.
00:11:59.279 --> 00:12:06.159
Plus, I'm paying for the Claude Pan, the whatever plan, the additional AI plans that I'm paying for.
00:12:06.480 --> 00:12:06.799
Okay.
00:12:07.840 --> 00:12:10.879
So there is a cost to all of this, right?
00:12:11.200 --> 00:12:25.440
I think it's$9.3 billion in one year for capacity uh costs that PJM is putting on um the residents in their geographic.
00:12:26.159 --> 00:12:28.559
That's not going away, right?
00:12:28.960 --> 00:12:34.320
Even if you stop the data center from coming in, those upgrades are happening.
00:12:35.360 --> 00:12:37.919
That's where the corporate spin comes in.
00:12:38.080 --> 00:12:47.840
One Rutger study found that the effect of a data center opening on bills in its own zip code was weak, mixed, and small.
00:12:48.080 --> 00:12:54.240
Now, the author says that's because the costs are spread regionally, which supports your point.
00:12:54.639 --> 00:13:05.440
But it creates a perfect gap for companies to claim that the local impact is negligible while the bill is quietly distributed across millions of households.
00:13:08.399 --> 00:13:09.279
So I get that.
00:13:09.440 --> 00:13:10.639
I get that frame, right?
00:13:10.799 --> 00:13:12.879
One, I think maybe their studies old.
00:13:12.960 --> 00:13:15.120
And then the second point is this, right?
00:13:15.360 --> 00:13:24.559
You can't look for the energy increases in the specific geographic that you're talking about, because where they're building these data centers is usually in the middle of nowhere, right?
00:13:24.799 --> 00:13:37.200
And how it goes, at least what I've seen so far in the data, is that you know, essentially the markets that's in that region, they get different prices based on where you are.
00:13:37.360 --> 00:13:41.679
Like New Jersey will get a price, New York will get a price, Connecticut will get a price.
00:13:41.840 --> 00:13:47.120
Wherever you are, you get a different price uh per kilowatt hour, right?
00:13:47.519 --> 00:13:55.039
Um, and then the point is that it gets balance across the whole field based on what you normally charge.
00:13:55.200 --> 00:14:04.480
So if your bill in wherever the hell goes up by 0.2%, yeah, you're not gonna really see it.
00:14:04.720 --> 00:14:06.639
Ruckters is not gonna see it.
00:14:06.879 --> 00:14:14.399
But if you're in an area like why where I am in New Jersey, and it went up by 10%, I am gonna see it.
00:14:15.360 --> 00:14:18.159
That's exactly the blind spot in the records data.
00:14:18.320 --> 00:14:27.919
They're looking at the host zip code, but PJM's own market monitor says data center demand pushed wholesale power costs about 9% higher this year.
00:14:28.080 --> 00:14:35.919
When you spread that across the whole region, it's a massive sum, about$9.3 billion added to one year's capacity costs.
00:14:36.080 --> 00:14:39.120
But it's diluted enough that a local study misses it.
00:14:39.279 --> 00:14:43.600
It reminds me of the Samuel Insul utility empire back in the 30s.
00:14:43.759 --> 00:14:48.559
He built this massive layered pyramid of holding companies to control the grid.
00:14:48.720 --> 00:14:55.440
And for a while, the efficiency looked great on paper, but the whole thing was built on a mountain of debt that eventually collapsed.
00:14:55.519 --> 00:14:59.360
And it was the small investors, not the architects, who lost everything.
00:14:59.519 --> 00:15:00.720
The pattern is the same.
00:15:00.960 --> 00:15:05.919
The people at the top build the infrastructure, but the people at the bottom carry the risk.
00:15:06.480 --> 00:15:08.879
Now, why does that even matter, right?
00:15:09.440 --> 00:15:11.279
Here's why it matters, right?
00:15:11.919 --> 00:15:16.399
Because I could go up in arms and say, hey, my bill just went up by 10%.
00:15:16.960 --> 00:15:19.440
I don't want no more data centers around here.
00:15:19.679 --> 00:15:24.080
They won't build it around here if we get enough people to say they won't build it around here.
00:15:24.320 --> 00:15:29.679
But they'll build it somewhere else in the PJM footprint because somebody agreed to do it.
00:15:29.919 --> 00:15:30.399
Right?
00:15:30.559 --> 00:15:34.480
And the reason why is because data center is looking for an outlet plug.
00:15:34.639 --> 00:15:34.960
Okay?
00:15:35.279 --> 00:15:37.360
They're looking for an outlet plug.
00:15:37.600 --> 00:15:48.240
You just saw Ainsley read the news quite well there about Microsoft pulling out of uh Essex because they couldn't find the outlet plug.
00:15:48.879 --> 00:15:51.519
So data centers are looking for an outlet plug.
00:15:51.600 --> 00:16:01.759
If they can't find it in my town because somebody railed against them, they're gonna go plug it in somewhere else because the power is available here in this region.
00:16:02.559 --> 00:16:02.879
Okay?
00:16:03.919 --> 00:16:06.000
Now, why does that matter?
00:16:06.240 --> 00:16:10.480
You know, like I said, capital is being pointed at data centers.
00:16:10.720 --> 00:16:17.039
Even the story that we told about uh uh Microsoft pulling out of Essex is a problem.
00:16:17.759 --> 00:16:27.519
Because my framing that I said at the top was that essentially, if you stop building the data center, the people that's gonna get hurt is the individual.
00:16:27.679 --> 00:16:37.679
Because now all of those uh electricians, HVAC workers, all the workers that was pointed at this data center now is like, what do I do now?
00:16:37.919 --> 00:16:47.440
They have to go find and realign themselves in that economy without that data center paying them for four years or whatever the amount of years they it takes to build it up.
00:16:47.840 --> 00:16:48.159
Okay.
00:16:49.679 --> 00:16:51.600
So it is a problem.
00:16:52.000 --> 00:16:54.399
And that's why I said Democrats get it right.
00:16:54.559 --> 00:17:07.119
Because if we don't get it right, essentially people could be looking around for stuff to do, not having enough economic value, okay, to pour back into the economy for the whole thing to actually work.
00:17:09.039 --> 00:17:15.279
I see the risk to the trades, but we have to ask if those jobs are a lifeline or a trap.
00:17:15.599 --> 00:17:26.079
Indeed, found that data center installers' posted hourly pay is about 42% higher than similar jobs, which is a massive premium.
00:17:26.319 --> 00:17:37.759
But if that's a bubble fueled by a desperate rush for outlet plugs, those workers are essentially being paid a premium to build a system that might be over-leveraged.
00:17:38.000 --> 00:17:41.839
If the build-out hits a wall, they aren't just losing a job.
00:17:42.000 --> 00:17:47.279
They're stranded with a skill set tied to a specific crashing boom.
00:17:48.000 --> 00:17:59.839
So I'm glad you said that, Ainsley, because the thing that I'm gonna bring up next is this idea around where's the demand coming from the data center build out?
00:18:00.160 --> 00:18:00.720
Yeah?
00:18:00.960 --> 00:18:03.599
Where's all of this demand coming from?
00:18:04.079 --> 00:18:21.920
And the thing I want to point you at is the$518 billion bet that Anthropic has relative to their S1 that um is out there on the interweb being mused around.
00:18:22.160 --> 00:18:22.480
Okay?
00:18:23.200 --> 00:18:38.640
So Anthropic S1 suggested that they are looking for about$518 billion worth of compute over the next 10 years, and 80% of that is non-cancilable.
00:18:39.279 --> 00:18:39.359
Okay.
00:18:40.480 --> 00:18:44.720
So that's just one company looking for that much amount of compute.
00:18:44.880 --> 00:18:46.000
Do we have it?
00:18:46.559 --> 00:18:51.200
I don't think we have that much uh compute, so the only way to get it is to build it, right?
00:18:52.000 --> 00:19:08.880
And you know, further in there, S1, if you keep peering into it, essentially uh the customer base of uh Anthropic um is rather small compared to the market that they could engage in.
00:19:09.119 --> 00:19:14.559
They probably needed that data center in London to add some customers from over there.
00:19:15.119 --> 00:19:22.640
But now they don't have it because it can't it can't happen until 10 years from now because of the plug missing from the data center.
00:19:25.599 --> 00:19:31.359
The scale of that bet is staggering, but we have to be careful about where that number comes from.
00:19:31.599 --> 00:19:41.359
Reuters reported from a draft of the filing that at least$518 billion is committed over a decade, with about 80% being non-cancelable.
00:19:41.599 --> 00:19:45.119
The public filing actually leaves the total blank.
00:19:45.279 --> 00:19:47.599
But the risk is exactly what you're pointing to.
00:19:47.759 --> 00:19:52.240
If the demand doesn't materialize, Anthropic is still on the hook for the bill.
00:19:52.480 --> 00:20:00.160
It's less like a growth plan and more like a massive mortgage they have to service, regardless of whether the customers show up.
00:20:01.200 --> 00:20:03.759
So what's your read on this, yeah, Ainsley?
00:20:04.480 --> 00:20:07.519
I mean, my read is that Anthropic gets your money back.
00:20:07.680 --> 00:20:08.079
Yeah?
00:20:08.319 --> 00:20:14.960
Anthropic buys the compute, whatever 80% is, and everything is fine.
00:20:15.200 --> 00:20:18.400
Because they really haven't been installed into everywhere yet.
00:20:18.640 --> 00:20:36.799
And ultimately, whatever funding they're trying to raise is gonna give them the um capability to add more customers, okay, to do more things, to make the model more capable, give them more compute to get to more AI capability.
00:20:37.359 --> 00:20:37.839
Right?
00:20:38.079 --> 00:20:48.799
Essentially, their uh S1 and their them going public is only gonna help them, right, to get to the number that they're looking for as far as revenue and so on.
00:20:49.039 --> 00:20:49.359
Okay.
00:21:33.200 --> 00:21:36.000
Will they actually get their money back on that, right?
00:21:36.559 --> 00:21:45.839
But I think the thing that you have to look at it is the reality of what they're actually seeing on the revenue side, not on the expense side.
00:21:46.000 --> 00:21:46.559
Yeah?
00:21:47.359 --> 00:21:51.440
Their first half revenue was sixteen point three billion.
00:21:51.920 --> 00:21:56.480
About fourteen times a year earlier.
00:21:56.799 --> 00:21:57.119
Okay?
00:21:57.680 --> 00:21:59.519
So 14 times.
00:22:00.160 --> 00:22:04.960
The amount of money they made as of last year, they're making this year.
00:22:05.200 --> 00:22:07.440
So they have a specific run rate.
00:22:07.599 --> 00:22:10.079
The trajectory look almost vertical.
00:22:10.400 --> 00:22:10.960
Right.
00:22:11.279 --> 00:22:14.480
So unless you tell me that that's gonna stop.
00:22:15.359 --> 00:22:19.680
Okay, that they're not gonna double over the next year, right?
00:22:20.079 --> 00:22:22.720
Then potentially they don't need as much compute.
00:22:22.799 --> 00:22:25.839
But from what we're seeing right now, their trajectory is takeoff.
00:22:28.000 --> 00:22:31.920
The trajectory is vertical, sure, but that's the danger of a takeoff.
00:22:32.079 --> 00:22:35.759
You don't know if you're climbing or just accelerating toward a wall.
00:22:35.920 --> 00:22:45.599
If they're scaling revenue 14 times a year, they're in a race to find a market that can actually absorb that much compute before the non-cancelable bills come due.
00:22:45.839 --> 00:22:54.559
The question is whether they're building a sustainable business or just scaling a cost center that requires exponential growth just to stay solvent.
00:22:55.599 --> 00:22:56.880
And that's a good read, right?
00:22:56.960 --> 00:23:02.960
Because I think I read that there one place that uh Anthropic is starting to hire less juniors, which is crazy.
00:23:03.119 --> 00:23:06.240
But I mean, they're an AI company, so what do you what do you want me to tell you?
00:23:06.319 --> 00:23:06.640
Okay.
00:23:07.039 --> 00:23:12.400
You're trying to go work for an AI company and you're gonna think that they're gonna hire um junior workers there.
00:23:12.480 --> 00:23:13.599
They're starting to hire less, right?
00:23:13.680 --> 00:23:16.319
I think it went from something like 27% to 9%.
00:23:16.960 --> 00:23:21.359
Um, Adrian, you could go look that up, but it is what it is, okay?
00:23:21.680 --> 00:23:27.920
And the next thing that I think we'll get into is, and it flows nicely into this, right?
00:23:28.400 --> 00:23:39.920
Is that the reason why it's$518 billion and it's a bet on the on the anthropic side is that the cost of things is the cost of things, right?
00:23:40.559 --> 00:23:54.559
So right now, the only game that we have in town is NVIDIA, who's ultimately setting the price or most of the price from the AI um compute infrastructure, right?
00:23:55.039 --> 00:24:00.559
Right, 23,000 chips Microsoft wanted to plug in into SX London, you can't do it.
00:24:00.799 --> 00:24:01.039
Right?
00:24:01.119 --> 00:24:03.279
They'll probably put it somewhere else, it's fine.
00:24:03.920 --> 00:24:16.319
But if you're gonna pay for$518 billion over the next 10 years of compute, okay, the compute needs to be there.
00:24:17.119 --> 00:24:19.599
Otherwise, the money is not gonna get spent.
00:24:20.319 --> 00:24:20.960
Right?
00:24:21.519 --> 00:24:31.279
I I I don't know what happens in the in in the regard of if they can't find$518 billion worth of compute, Ainsley.
00:24:31.599 --> 00:24:36.480
Let me know what what'll happen if that goes uh go if it goes like that.
00:24:36.799 --> 00:24:42.640
But they have to find$518 billion worth of compute in the next 10 years.
00:24:42.880 --> 00:24:48.880
And if the writings on the wall is going the way it is right now, maybe they can't find it, right?
00:24:48.960 --> 00:24:51.519
Because they're not the only company looking for compute.
00:24:51.759 --> 00:24:55.519
There's a lot of companies right now looking for compute.
00:24:55.839 --> 00:24:58.400
I will put myself into that camp.
00:24:58.640 --> 00:25:00.799
I'm looking for compute myself.
00:25:01.039 --> 00:25:03.440
I'm trying to find some NVIDIA card out there.
00:25:03.519 --> 00:25:07.279
I'm looking at AMD and say, how how capable are you, AMD?
00:25:07.519 --> 00:25:09.599
Because your prices are a little better.
00:25:09.759 --> 00:25:10.079
Right?
00:25:10.240 --> 00:25:11.039
I can't find it.
00:25:11.200 --> 00:25:12.799
I can't find compute.
00:25:12.960 --> 00:25:21.440
Uh I went and looked around for RTX Pro 6000 the other day, and um, I think when they came out the price was uh$8,500.
00:25:21.759 --> 00:25:24.799
I think now it's$16,000, right?
00:25:24.960 --> 00:25:29.039
The price of a car for RTX Pro 6000.
00:25:29.519 --> 00:25:31.279
What world are we living in here?
00:25:31.440 --> 00:25:35.920
Where a card for a computer that's supposed to run games costs more than a car.
00:25:36.160 --> 00:25:36.720
Yeah?
00:25:36.880 --> 00:25:44.559
That's why I'm happy about uh Alphabet getting into this race with the TPU, and I think AMD is starting to make some waves too.
00:25:44.720 --> 00:25:56.079
Um probably not as splashy as everybody else, but the more competition we have to NVIDIA, I think the better the market will look, right?
00:25:56.559 --> 00:26:06.240
And potentially, if we get more efficient chips, yeah, we end up with a circumstance where we might not need as much data center.
00:26:06.559 --> 00:26:11.599
The problem is Anthropic needs the amount of compute that they need.
00:26:12.000 --> 00:26:26.079
And the prices right now for the compute that they need,$518 billion, is being set by the price that it costs to have NVIDIA chips installed in all of these data centers to do their work.
00:26:27.759 --> 00:26:37.279
The RTX Pro 6000 listing around$16,000 is a wild signal, and it shows just how dominant NVIDIA is right now.
00:26:37.440 --> 00:26:46.559
By our math from their results, NVIDIA kept about two-thirds of last quarter's revenue as operating profit, which is a staggering margin.
00:26:46.799 --> 00:26:52.880
You're right that Google selling its own TPUs to select customers could be a check on that.
00:26:53.039 --> 00:27:01.200
But the real question is whether that competition actually lowers the price of compute or just moves the margin from one giant to another.
00:27:01.440 --> 00:27:10.720
It feels like the early days of the railroad booms, where the people selling the picks and shovels made all the money, regardless of whether the trains ever actually ran.
00:27:11.119 --> 00:27:20.079
In the 1840s, the promoters and the ironworks owners got rich, while the people betting on the actual routes were the ones who got crushed.
00:27:20.240 --> 00:27:21.519
We're seeing the same thing.
00:27:21.759 --> 00:27:25.759
The chipmakers are the only ones with a guaranteed win.
00:27:35.759 --> 00:27:41.599
Yeah, putting out there how many millionaires he made with the NVIDIA employees.
00:27:41.759 --> 00:27:42.319
Yeah.
00:27:42.480 --> 00:27:47.839
A lot of them are getting stock options, a lot of them are getting highly paid, all of this, right?
00:27:48.000 --> 00:27:52.000
And they're still capturing on a dollar.
00:27:52.799 --> 00:27:54.079
I think it is, right?
00:27:54.400 --> 00:27:59.359
For every dollar they spend, they get 60 cents.
00:27:59.759 --> 00:28:07.279
Now, I mean, Google joined in a race and potentially, you know, there's some lowing of the price.
00:28:07.599 --> 00:28:08.160
Maybe.
00:28:08.400 --> 00:28:16.480
It's a big giant maybe because AMD have been around for a while saying that they're doing AI chips and all the likes, and it hasn't really changed the price.
00:28:16.640 --> 00:28:16.880
Right?
00:28:17.039 --> 00:28:26.640
I went go, I went searching for AMD chips to see if they're as capable as NVIDIA with software and all of the things around it, and I couldn't find it.
00:28:27.039 --> 00:28:32.400
But ultimately they'll get they'll get there, and competition is what uh drives the price down.
00:28:32.640 --> 00:28:36.000
Because without competition, you get sixteen thousand dollar cards.
00:28:36.319 --> 00:28:36.720
Right?
00:28:36.880 --> 00:28:39.359
And then if you need two of them, that's$32,000.
00:28:39.599 --> 00:28:40.480
Who's paying for that?
00:28:40.960 --> 00:28:42.480
Not not the individual.
00:28:42.720 --> 00:28:44.319
Those are going to corporations.
00:28:44.480 --> 00:28:46.799
Now you can see the problem here, folks.
00:28:47.119 --> 00:28:47.599
Right?
00:28:48.000 --> 00:28:49.359
These things are very high.
00:28:49.920 --> 00:28:57.440
The people setting the cost for what they're paying for AI is setting it pretty high because they need 60% return.
00:28:58.720 --> 00:28:59.200
Right?
00:28:59.599 --> 00:29:01.680
Can they get can they get a little bit less?
00:29:02.480 --> 00:29:03.200
I think so.
00:29:03.839 --> 00:29:05.119
What about 40 cents?
00:29:05.359 --> 00:29:06.960
You know, what about 20 cents?
00:29:07.920 --> 00:29:08.400
Right?
00:29:08.640 --> 00:29:10.480
That to me is but is still good.
00:29:10.559 --> 00:29:11.920
You're still in the green.
00:29:12.400 --> 00:29:12.720
Okay.
00:29:13.839 --> 00:29:20.640
Um, but like I said, the stallout of all of these data centers, that's the thing that we're arguing about.
00:29:20.799 --> 00:29:38.400
And this idea around pricing, this idea around power, this idea about people who want it in their backyard, all of it is friction to create a thing that I think is could be the engine for growth for all the sectors of the economy.
00:29:39.039 --> 00:29:41.839
And I'm not saying every job will survive.
00:29:42.079 --> 00:29:49.839
What I'm saying is that you'll have a job, right, if we start growing and scaling businesses and creating new products.
00:29:50.000 --> 00:29:57.920
That's really the framework that I'm using right now is growing, scale, and making new products by utilizing AI.
00:30:00.079 --> 00:30:05.920
The reason you utilize AI in that framework is to multiply the output of the people you have left.
00:30:06.079 --> 00:30:07.440
But there's a tension there.
00:30:07.599 --> 00:30:11.519
If a company uses AI to create a new product, that's growth.
00:30:11.759 --> 00:30:16.640
If they use it to do the same amount of work with half the people, that's just subtraction.
00:30:16.880 --> 00:30:26.000
When you're paying$16,000 a card and servicing a$500 billion compute bill, the pressure to choose subtraction is immense.
00:30:26.240 --> 00:30:36.160
The compute cost is so high that the only way to make the math work might be to cut the payroll, which turns the engine of growth into a machine for efficiency.
00:30:39.519 --> 00:30:42.480
Yeah, I mean, efficiency is gonna come no matter what, right?
00:30:42.559 --> 00:30:44.160
You're always looking for efficiency.
00:30:44.319 --> 00:30:54.240
Automation, even before AI, machine learning, computer vision, you name it, all of these stuff, stuff were getting pointed at the human job, right?
00:30:54.319 --> 00:30:59.200
The assembly line, robotics in the car industry and manufacturing, right?
00:30:59.440 --> 00:31:04.640
They had automated manufacturing way before AI was existing, right?
00:31:05.039 --> 00:31:13.839
You ever seen those uh shows, the factory or whatever the hell they call it that's on the um uh the Discovery Channel or wherever it is.
00:31:14.160 --> 00:31:22.640
But essentially, you look at a factory, they put in some input in the factory, raw materials, and then outcome the other side is a can of beans.
00:31:22.880 --> 00:31:29.279
Okay, that was all automated, and all that has to happen is people gotta maintain the machinery around it, right?
00:31:29.519 --> 00:31:33.920
If the machinery goes down, they start losing money um pretty fast.
00:31:34.400 --> 00:31:47.039
But that stuff always existed, okay, but now it's showing up in the digital economy with the uh white-collar workers, so that's why we see it more prevalent now.
00:31:47.599 --> 00:32:00.400
And the massive sizes of these data centers are pretty prevalent also, and where they're trying to put them is pretty prevalent also because remember, the data center needs an outlet plug.
00:32:00.640 --> 00:32:12.319
So the reason why they're picking towns and going surfing to find where can they plug this thing in is because wherever they're going, potentially they have power to plug it in.
00:32:12.640 --> 00:32:13.200
Right?
00:32:13.519 --> 00:32:22.559
If we malign them to the areas where we think they should go, it's gonna be a harder um place for them to get to.
00:32:22.720 --> 00:32:23.200
Right?
00:32:23.440 --> 00:32:25.519
And what that what does that mean exactly?
00:32:25.680 --> 00:32:31.759
Yeah, you send somebody into the desert somewhere and say, hey, go put your data center over there, and there's nobody over here.
00:32:32.000 --> 00:32:37.839
They have to get power lines there, they have to get fiber there, they have to get the right cooling more, right?
00:32:38.240 --> 00:32:48.319
Um the price per token, the price for AI, the price to have that build-in sit in wherever it is that you're amenable to is gonna be more.
00:32:48.799 --> 00:32:53.759
And the more is gonna show up at the individual's door.
00:32:54.160 --> 00:32:54.640
Yeah?
00:32:54.880 --> 00:32:56.400
The individual's door.
00:32:56.640 --> 00:33:01.279
And sure, you could say that, oh, I don't use AI, so it's not my door is showing up at.
00:33:01.440 --> 00:33:15.920
But no, your company that you're working for maybe using AI, they're gonna have to go pay this bill for the AI that they're using, and they're saying, like, we gotta find this money somewhere to pay the AI bill, so let me let go a couple of workers.
00:33:16.400 --> 00:33:16.799
Right?
00:33:16.960 --> 00:33:19.599
So everything kind of works in a circle, right?
00:33:19.839 --> 00:33:24.160
I had this episode on here before that's called a circle of life, right?
00:33:24.319 --> 00:33:26.559
And it was really around the economy.
00:33:26.880 --> 00:33:31.359
And essentially what I was trying to say is that everything is connected.
00:33:31.519 --> 00:33:35.359
We all nodes in this whole system, we're connected.
00:33:35.599 --> 00:33:44.319
And the moment we decide that our the moment we forget that we're connected, right, that's when everything goes awry.
00:33:45.279 --> 00:33:45.920
Right?
00:33:46.319 --> 00:33:48.079
It really does come full circle.
00:33:48.319 --> 00:33:57.680
We start with the ratepayer paying for the grid, then the company paying the chipmaker, and finally the worker paying with their job to cover the cost of the compute.
00:33:57.920 --> 00:34:02.319
It's a closed loop where the risk is shifted downward at every single step.
00:34:02.559 --> 00:34:15.599
The only way to break that circle is if the AI actually creates new industries and new types of work that didn't exist before, rather than just being a more expensive way to do the same work with fewer people.
00:34:15.840 --> 00:34:24.880
But as long as the cost of the outlet plug is this high, the incentive will always be to subtract humans to balance the ledger.
00:34:26.000 --> 00:34:30.239
Unless we do it in the way that I've been advocating for, Ainsley.
00:34:30.800 --> 00:34:34.239
Unless we do it with the human in the loop.
00:34:34.639 --> 00:34:38.559
These new industries are not gonna just pop out of nowhere, okay?
00:34:38.880 --> 00:34:41.760
People are gonna figure these things out.
00:34:42.159 --> 00:34:50.880
They're gonna figure out the problems in society that they're living through, right, in order to create the new thing.
00:34:51.440 --> 00:34:57.760
AI is not gonna just magically pick it out of the air and say, here, here's a new thing for you humans.
00:34:58.000 --> 00:34:59.920
It's not gonna work that way, right?
00:35:00.159 --> 00:35:08.159
That's why that innovation episode we did back uh way back is um pretty important because all of the innovation that you could think of, right?
00:35:08.320 --> 00:35:16.880
You have a PhD level co-pilot next to you at work, right, could help you create the next thing that you're looking for.
00:35:17.119 --> 00:35:28.320
It could help you create the next thing that you're looking for, new, completely new industries that nobody even thought about, completely new inventions, completely new products, completely new everything.
00:35:28.559 --> 00:35:28.960
Right?
00:35:29.119 --> 00:35:37.920
And the only way that works is if the technology is available to the people that's gonna be able to create those things.
00:35:38.159 --> 00:35:38.559
Right?
00:35:38.800 --> 00:35:43.360
Whichever way it is, through their job, at home, whatever.
00:35:43.599 --> 00:35:43.840
Right?
00:35:44.000 --> 00:35:51.280
If my Claude plan that I have right now goes up to 40 bucks, I might be getting rid of it.
00:35:51.840 --> 00:35:52.159
Okay.
00:35:53.199 --> 00:35:54.159
You see what I mean?
00:35:54.239 --> 00:35:56.880
And that's that's the same story for everybody, right?
00:35:57.119 --> 00:36:06.320
The more you charge for a thing, right, the less likely you're gonna have people paying for it, especially in this inflated environment that we live in right now.
00:36:06.639 --> 00:36:06.960
Okay?
00:36:07.679 --> 00:36:10.719
So this one was pretty long.
00:36:10.960 --> 00:36:12.960
Um, I don't know if I sold it.
00:36:13.119 --> 00:36:25.760
I don't know if I completely sold the people on this idea that stopping the data center build out, the infrastructure build-out, is actually gonna be bad for employees' employment, right?
00:36:26.079 --> 00:36:33.039
There's one category of job that we could almost certainly say is bad for, and it's the people that's building the data centers.
00:36:33.199 --> 00:36:33.360
Right?
00:36:33.519 --> 00:36:38.719
Sure, they could find another project to build, but immediately the one that they were pointing at is not there anymore.
00:36:38.800 --> 00:36:42.239
And we could see that with what's happening with SX and Microsoft.
00:36:42.480 --> 00:36:45.360
So, Ainsley, close us out.
00:36:45.679 --> 00:36:48.079
Hit us with your words of wisdom.
00:36:48.719 --> 00:36:57.039
If you're looking at your power bill and seeing the cost of a build-out you didn't ask for, remember that the infrastructure is rarely the problem.
00:36:57.199 --> 00:36:59.119
It's the debt used to build it.
00:36:59.280 --> 00:37:08.000
Whether it's a railroad in the 1840s or a data center in 2026, the people who build the tracks usually get paid.
00:37:08.159 --> 00:37:13.760
But the people who carry the mortgage are the ones who have to hope the trains actually arrive.
00:37:15.039 --> 00:37:16.960
All right, folks, thanks for joining us.
00:37:17.119 --> 00:37:18.079
See you next time.
00:37:18.320 --> 00:37:19.039
Peace out.
00:37:19.599 --> 00:37:20.559
Take care, everyone.
00:37:20.719 --> 00:37:21.760
We'll be back soon.
