Ep 341 - AI's Hidden Thirst with Erald Kolasi

Our friend, physicist and economist Erald Kolasi, stops by the Macro N Cheese clubhouse to talk with Steve about the profound effects of AI on the energy grid, water resources, and societal infrastructure. The discussion focuses primarily on large-scale corporate AI, such as generative AI.

Erald’s work bridges physics, economics, and ecology, revealing how AI’s rapid expansion is not just a technological phenomenon but a biophysical crisis – one that’s easy to overlook. Cloud is such a gentle word. Diaphanous. It sounds harmless. Lovely, even.

“When you're in front of your computer and you're just typing away and you're asking these systems to do all these magical things for you, it can seem like it comes out of nowhere. But no, in reality, all of this stuff takes enormous energy.” 

AI’s dematerialized facade obscures its physical infrastructure. It’s a classic capitalist contradiction where "progress" accelerates ecological breakdown. 

Erald and Steve talk about the race to the bottom, as states and municipalities trade public health for tax revenue. Regulatory enforcement is absent.  

While exploiting labor and plundering nature, the costs are socialized as these companies use public water and energy grids. Elon Musk’s xAI Colossus is based in Tennessee. (Remember the TVA, that impressive example of depression-era federal works? Help yourself, Elon.) It’s not just that they use public water and energy, it’s the vast and growing amounts of these resources, as Erald explains. 

The conversation also touches on the AI arms race, as the US competes with China, using “national security” as an excuse to justify resource wars.  

From energy consumption to water depletion, from labor displacement to geopolitical tensions, this episode exposes the contradictions of AI under a system that prioritizes profit over sustainability. 

Erald Kolasi is a writer and researcher focusing on the nexus between energy, technology, economics, complex systems, and ecological dynamics. His book, The Physics of Capitalism, came out from Monthly Review Press in February 2025. He received his PhD in Physics from George Mason University in 2016. You can find out more about Erald and his work at his website, www.eraldkolasi.com.  

Subscribe to his Substack: https://substack.com/@technodynamics 

Episode Transcript

(00:43):
All right, folks, this isSteve with Macro and Cheese. I talked
to this gentleman that I'mgoing to talk to today a few months
ago, it's actually probablythree or four months ago. And his
work in physics andunderstanding the physics of capitalism
was extremely eye opening tome. There's a lot of things I kind
of know as a layperson and assomebody who has spent a lot of time

(01:05):
in reading about modernmonetary theory in class and understanding
theory. And I've had a lot offolks come on and talk about the
environment and so forth. Butmy guest today, Errol Colassi, who
wrote the book the Physics ofCapitalism, he's a physicist and
an economist focusing on thenexus between energy, technology,

(01:27):
economics, complex systems andecological dynamics. His new book,
which is not so new, but isstill new to you if you haven't read
it yet. The Physics ofCapitalism, came out from the Monthly
Review Press in February of2025. He also received his PhD in
physics from George MasonUniversity and has been a guest on
this podcast. You will wantdefinitely to go back and listen

(01:50):
to that first episode that wedid together. It is absolutely amazing.
It's one of our most listenedto episodes out there, so please
do check it out. But before weget started, I want to tell you a
little bit about what we'regoing to talk about today. You know,
I've had gentlemen come on anddiscuss in great detail aspects of

(02:11):
artificial intelligence, AI.We've talked about the morality of
it. Does AI deserve its ownrights and responsibilities as sentient
beings? Well, these AI devicestake on human traits and we've discussed
a whole host of things as faras it goes with militarization of
AI and in the classroom even,okay, but today we're going to talk

(02:36):
about energy. We're going totalk about the interrelatedness and
interconnectedness of allthings. We're going to talk about
the displacement and thedestruction of the biosphere. And
we're going to talk about thereal resource, folks. The real resources
that modern monetary theorybases all of its token on is the
availability and mobilizationof real resources. And that is going

(03:00):
to come into play big time inour conversation.
Today.
So without further ado, let mebring on my guest, Errol Colassi.
Welcome to the show, sir.
Hey, Steve, thank you forhaving me back on. Very excited.
Absolutely. This subject, itis near and dear, I think, to many
people's hearts for differentreasons. Like, I think everybody

(03:20):
comes to concerns with AI froman employment perspective to how
this is going to be used tooversee them at work or even watching
what's happened in Gaza withthe militarization of it, which we've
talked about in severalepisodes in the past. But I don't
think we've talked at length,at least in a detailed fashion, on

(03:43):
what it does to the biosphere.And quite frankly, we did discuss
some of this in our previousconversation, but we didn't go into
the depth we're going to gointo today. So what I'd like to ask
you to do is kind of set thestage and explain what the AI world
is and what it is doing, andthen we can get into how it is funded

(04:06):
and what the impacts of thatare on employment, the biosphere
and our lives in general.
Yeah, absolutely. Thank you.So when you talk about what AI is,
you know, I don't think you'regoing to find two people on planet
Earth who will agree on themeaning of artificial intelligence.
Right. I just recently talkedto somebody who doesn't even like
the term artificialintelligence. And there's people

(04:26):
like Noam Chomsky and otherintellectuals who will tell you these
generative AI systems likeChatGPT and Copilot and Gemini, they're
just pattern recognitionsystems, right? They're not intellectualizing,
they're not actually thinking,they're not doing anything. They've
just been taught to capturepatterns and that's all they're doing.
So that's a whole other sortof complicated mess. But what I mean

(04:46):
with AI, in the context ofthis discussion that you and I are
having, if we could justfocus, it is I'm talking about these
systems that are based on deeplearning and neural networks. So
that's some of the things thatI just mentioned. You know, Grok,
ChatGPT and Llama, things likethat, Right? So these large language
models and other similar kindsof models, you know, image generators,
things like that. Sopredominantly generative AI and some

(05:09):
of the things that areultimately going to be based on generative
AI, like agentic AI, right?These autonomous AI systems that
are supposed to accompany youin your daily life as you do tasks
at work or just normally inyour personal life. So that's what
I'm talking about when we'retalking about the rise of AI in the
context of this discussion,and what are the implications going
to be for our energy use andresource consumption, and what are

(05:32):
the constraints on that andwhat's going to be the impact that
the rise of these systems isgoing to have on society? Right.
And like you sort ofmentioned, there are many complex
dimensions to this issue. Sowith AI, you could talk about the
economics of it, the impact onjobs and job displacement and the
impact on the stock market,you could talk about its impact on

(05:54):
education, you can talk aboutits impact on our health and psychology.
And all of those arefundamentally important issues and
I think will probably overlapwith the discussion that we have
today. But what underlies allof those issues is the energy and
biophysical resources that arenecessary to make this world happen
in the first place. So at theroot of all that are these things

(06:16):
called data centers, and theseare these gigantic specialized facilities
that have these enormous rowsof server racks, which is basically
rows and rows of computersystems if you step inside one, especially
the hyperscale data centers,you know, the larger ones, although
there are many other kinds ofdata centers, there's even on site
ones. This is what companiesused to do back in the day. They

(06:38):
used to have on site smalldata centers where they basically
packed all of their serversand computers. Now, of course, they've
moved more and more to thecloud, quote, unquote. And the cloud
isn't some magical world upthere in the sky, it's right down
here. It's these massive datacenters where they're putting all
these computing resources. AndWhenever you use ChatGPT, whenever
you use these AI systems, yourqueries, your search results and

(07:00):
all of that, they're going tothese data centers. And there's all
these computer chips insidethese data centers. And for AI applications,
especially relevant ones areNvidia's GPUs, right? So graphics
processing units. So there'sall these computer chips, they do
a bunch of calculations tofigure out the next word that you're
going to get in your reply,right? And so you tell Chatgpt, you
know, give me an essay on 19thcentury Russia or something like

(07:23):
that. That's what's happening.The request is being sent to these
data centers and all of thesecomputer chips are doing their matrix
multiplications at rapidspeeds. And then eventually that
answer is sent back to yourhome, to your router, and that's
what you see right on thescreen. So that is sort of the underlying
biophysical dynamics of whatis happening when people interface
with These systems. And Ithink that's important to emphasize

(07:43):
because you can be lulled intonot a false sense of security, but
you can kind of forget aboutsome of the energy and biophysical
aspects of this. When you'rejust in front of your computer and
you're just typing away andyou're asking these systems to do
all these magical things foryou, it can seem like it comes out
of nowhere. But no, inreality, all of this stuff takes
enormous energy. So I justwant to talk about the growth of

(08:05):
data centers over the past fewdecades. So in the year 2000, Steve,
data centers comprised about0.1% of all the electricity consumption
in the United States. Nowthey're up to 4% in 2023, 2024. So
4,000% increase in their shareof national electricity consumption.

(08:26):
Wow. Data centers in theUnited States last year consumed
about 200 terawatt hours ofenergy in their electricity consumption.
That's about half of theworld's total data center electricity
consumption. About half of itis coming in the United States. And
if terawatt hours doesn't meananything to you, let me put it this
way. New York city consumesabout 50 terawatt hours of electricity

(08:49):
every year. So the datacenters in the United States are
consuming about four New YorkCity's worth of electricity every
single year. Right. So there'sjust massive energy guzzling facilities.
And another thing I want toemphasize before I turn it over to
you is that data centers arenot the only aspect of this story.
Right. So that's often theaspect that is emphasized in the

(09:09):
media. Oh my God. Data centersneed so much electricity. Where are
we going to get all thiselectricity? Oh my God. Data centers
need to consume so much water.Right. Because all of this equipment
inside the data centers, itgets really, really hot. And so now
people are turning to liquidpowered cooling and they're using
water, basically cold waterrunning through these pipes across
the data center to cool downthe data center. Right. To make sure

(09:31):
that the equipment doesn't fryso it can continue operating. That's
why a lot of the focus, and Iwant to talk about some of the consequences
associated with all that, withthe electricity needs and the water
needs and all the problemsthere. But this story is not just
about data centers. Andsometimes I don't like it when we
just isolate it to datacenters and to electricity consumption.
Really, the rise of AI in datacenters is having a lot of complex

(09:53):
upstream effects, effects anddownstream effects across all of
our energy networks. And sothat's one of the things that I hope
to get into as well over thecourse of our discussion.
You know, one of the thingsthat concerns me most specifically
about the United States, thisdecentralized capitalist driven model,
is there is really no centralplanning. There is really no central

(10:14):
authority that understands themain problems. There is no real regulatory
body that's actually saying,hey, you know, when we do these things,
these externalities happen. Itseems like a mad dash based on, let's
be fair, capital driving morecapital to do capital things which
are always about moreaccumulation, more wealth. And that

(10:36):
usually involves layoffs. Thatusually involves a host of other
things that sound great whenyou're hearing about, hey, my 401k
went up because my investmentin AI went up. But it doesn't talk
about the other interrelatedcasualties of war, the collateral
damage, if you will. I waswaiting for you to bring up the water
thing and then you said it. Iwas like, dag on. Because that water

(10:58):
thing really is serious. Imean, we're talking about entire
regions that are water poor,that don't have water yet. At the
same time, though, we'rebleeding away all of our water, literally
snatching it up to cool thesemega centers down. I mean, Elon Musk
himself, I think it's inMissouri as some massive, I mean,
like a city's worth of datacenters. He's turned things up using

(11:22):
all kinds of differentstrategies that have really destroyed
the biosphere for the localcommunity. Anyway, take us through
this man. Like, take it fromwhere I just stated and keep going
with it.
Absolutely. This is one of thethings that concerns me the most.
Right. Its impact on our waterresources and especially in the dry
and arid regions of the UnitedStates where a lot of these things
are going up. Yeah, youmentioned Elon Musk. So Colossus

(11:43):
is in Memphis in Tennessee.Right. And that's the name of this
massive. The supercomputer iswhat they call it. But it's basically
a gigantic data center withover, you know, 100,000 hoppers.
Those are Nvidia's GPUs,right. Over a hundred thousand computer
chips consumes about 3 to 5million gallons of water every single
day. To give people an idea ofhow much that is, I mean, it's the

(12:04):
equivalent of over 10,000typical American households every
single day. So it's anenormous amount of water consumption
because like I said, it's soimportant to keep these data centers
cooled down. Otherwise they'rejust not going to work. Right. They're
just going to overheat. And sothere's enormous water consumption
happening, and it's reallyimpacting people's lives. You know,
I think there was a metadatacenter in Newton County, Georgia.

(12:26):
It's a fairly semi ruralexurban area, right? And it started
drawing water from the localunderground reserves, the aquifers.
And people noticed shortlyafter this data center was being
constructed, but also after wefinished last year, they noticed
their wells running dry orhaving weird funny colors. The water
that was coming out had like aweird brown color and everything.

(12:48):
And so the suspicion is thatthe construction of this data center
in Newton county essentiallypolluted the water, right? And it's
led to people delayingretirement because now they have
to spend vast amounts ofmoney, you know, cleaning their water
or finding a way to get water,and some of their bathrooms aren't
working. It really has amassive effect on these local communities.
And the crazy thing is thatthese data centers, when they plug

(13:11):
in for water, they'retypically plugging into the public
water supply, right? Thepublic water systems. And that's
another funny aspect of thisstory, Steve. You know how I often
make fun of capitalists foressentially relying on the state
and public resources in orderto grow, right? And Elon Musk is
obviously example number oneof that bloodsucker of them all.
And now, in fact, Colossus isplugging into tva, right? The Tennessee

(13:33):
Valley Authority. And as youknow, that's a federal company, it's
a government companyestablished during the Great Depression.
So now they're getting theirelectricity from a government company,
and these data centers areplugging into public water systems.
But a lot of what they'redoing, too, in many places where
these water systems don'texist, is they're also tapping into
these underground waterreserves, right? And aquifers. And

(13:53):
this is especially morerelevant in, like, drier rural areas
where a lot of these datacenters are going because the power
might be cheaper. So theyreally love cheap power. So they're
going into all these randomplaces now across the United States,
like Newton County, Georgia,and they're really putting a lot
of stress on available waterresources and on water quality. And

(14:13):
people's water bills are goingup, right? So if a data center is
going to move next to you,your water bills are going to skyrocket.
It's gotten so bad that manycommunities in the United States
are actually fighting backnow. So in Becker, Minnesota, they
recently rejected a proposeddata center from Amazon. They said,
nope, this is going to consumetoo much. It's going to break us.
We can't handle it. And you'restarting to see fights like that

(14:35):
break out across the UnitedStates. It's a major issue here in
Virginia, where I live,because we have the largest cluster
of data centers in the world.Right. It's called Data Center Alley.
Just dozens and dozens of datacenters here. So it's leading to
a lot of political fights. Youknow, Chile said no to some data
centers from Google, I think,or from whoever it was. Uruguay is
having some fights over it. Soit's becoming a major political issue

(14:56):
because people are now wakingup to the energy and biophysical
demands of these massive datacenters and the possible consequences
on their communities. And wejust talked about water, Steve, but
I also want to talk about theelectricity angle. So as you can
imagine, when these massivedata centers come in, they put a
lot of new demands on the gridbecause a lot of them do connect
to the regular utility grid.Now many of them have their own backup

(15:18):
generation units in case powerfails, right. So they have a lot
of like diesel generators,natural gas generators, and some
of them have active on siteindependent power units. And they
might be largely disconnectedfrom the grid, but most right now
are still very much connectedto the grid. And if the grid fails
or if they have to consumeless from the grid, they talk to
utilities basically, and theyplan then they might turn on their

(15:41):
independent power units.Right. But it's leading to a situation
where because they're puttingmore and more stress on the grid,
you're starting to see morepower outages and brownouts in the
United States, especially nearcommunities where these places are
located. So in NorthernVirginia, for example, where there's
a lot of data centers, poweroutages are becoming more and more
common. Now officially it ishard to causally attribute them to

(16:03):
data centers because thecompanies themselves release very
little data. The politiciansdon't want to talk about it much.
Right. Because they justfinance the data center so they can
get the tax revenue. And thetax revenue is good for these things.
But they have all of thesecomplicated knock on effects right
now. They're putting a lot ofpressure on the grids. And especially
if you're talking about thesummer, you've already got people

(16:25):
cranking out their AC unitsbecause it's really hot. So that's
overloading the grid. And nowyou're adding data centers to an
already stressed out grid andit's not a good combination. Right.
And so it's leading to a lotof brownouts and power outages across
these communities. In theUnited States, utilities, because
they're investing in morecapacity, you know, more Capex and
more capacity, they're passingalong those bills to you, right?

(16:47):
So the more and more datacenters that are coming in, your
electricity bills, this hasalready happened in the United States,
right? Our electric bills arejust skyrocketing. That's going to
continue happening. And I alsowanted to talk about the health aspect
of it a little bit, becausewhat a lot of people don't realize
is that when data centers haveto move to, like, diesel generators
and natural gas generators,either because they want their own

(17:08):
independent power units or asa backup source if power fails, those
things emit a lot of harmfulpollutants. Right? So the natural
gas generators, for example,that Elon Musk installed at Colossus
in Memphis, 35 natural gasgenerators, those things are pumping
out benzene, which is acarcinogen. They're pumping out nitrogen
oxides, which are a criticalcomponent in smog, right? So that

(17:30):
dirty, heavy, polluted air,and this is particulate matter, right?
So small tiny particles thatcan get in your lungs. And so these
are things that aggravatethings like asthma and heart disease
and lung disease and all ofthese conditions which people are
struggling with. And in thecase of Elon Musk and Colossus in
Memphis, it was also placed ina generally, you know, near a poor

(17:51):
area, right. An area that'salready struggling with a lot of
the history of environmentalinjustice over there in South Memphis.
That's where this facility wasput. And they brought in these natural
gas generators, all themethane they're pumping out. Right.
Which is a dangerousgreenhouse gas, as you know very
well. We'll get to that partlater. But right now I'm just talking
about the public healthaspects of it, right? So if you live

(18:11):
near these massive datacenters, don't be surprised if five
to 10 years from now or beyondthat, we start getting stories in
the media about, oh, peoplewho live here have higher rates of
cancer or higher rates of allthese other issues, Right? And so
this is, I think, anunderappreciated aspect of the rise
of AI and data centers. It'snot just the economic and job displacement,

(18:32):
not just the pressure on ourenergy and biophysical resources.
It's also the public healthaspect of it as well. What's going
to be the impact on our airquality, on our water quality, and
what are going to be thedownstream effects on our health
as a result of all this stuff,right? So that is something that
absolutely does concern mevery much.
You know, I think to myself,in a planned economy, in an economy

(18:54):
where you have a functioningDemocracy, we don't. In a place where
you can actually makedecisions that are in the interests
of the entire populace versusin, you know, the benefit of a few
billionaires or oligarchs, onemight consider strategic placement
of data centers and maybe eventhinking about desalination plants,

(19:16):
right? Like taking salt waterfrom the ocean and using that to
generate energy as well aswater to handle these things. Again,
I'm just thinking out loud asyou're talking. There's gotta be
a way of handling these thingsother than just allowing laissez
faire capitalism to make therules for us. It doesn't feel like

(19:36):
anybody is steering the shiphere. It feels like capital is. I
mean, literally only the onedominating the decider of all things
is oligarchic capital. Am Imissing something? Because I certainly
don't see any meaningfuldemocracy whatsoever saying, hey,
you know, this is killing us.We'd like to see this done differently.

(19:56):
Like, I'm already convincedthere's not, but that's almost at
this point, conjecture,because I don't have that data in
front of me, but my eyes, youknow, do not believe my lying eyes
kind of thing. Right? I mean,why are we not regulating these things
in such a way where thesenegative externalities are either
dealt with or, or we shut itdown and we say, no, that can't happen.

(20:18):
We can't do that. You're goingto kill people.
Yeah, great question. At thehighest levels of our federal government,
AI has now received the greenlight for these elite Silicon Valley
investors and executives tobasically throw all the money in
the world they want to growingwherever they want. So at the highest
level of government, thereisn't a lot of, I don't know what

(20:39):
you want to call it,democratic pushback, quote, unquote,
where that resistance ishappening to these growing AI empires
and all the consequences thatthey're going to have. It's mostly
happening at the local leveland some people are trying to do
it at the state level. They'retrying to pass laws and regulations
to try to kind of rein thesepeople in. Right? Because right now,
yeah, it's. It's absolutelycrazy. There aren't a lot of constraints.

(21:02):
There's just. Mark Zuckerbergdecides, I want a massive new mega
cluster in Louisiana withthree natural gas power plants that
are going to consume 2gigawatts of power on average. And
that happens, you know, that'slike, sure, it'll bring in jobs,
right? But there isn't anykind of overall strategic direction
or planning from the federalgovernment besides Just saying, do

(21:22):
whatever you want and grow asfast as you want. And the way the
Silicon Valley elites, the waythe capitalists are selling this,
at least to the federalgovernment, is we need to make sure
that we stay ahead of China.Because China's also throwing now
massive multi billion dollarinvestments into AI, Right? So they're
also building out their datacenters and training their models
and doing cloud computing andall that. So the argument for almost

(21:45):
anything for the rulingclasses nowadays is, well, if we
don't do it, China will and wecan't fall behind the Chinese. So
therefore, we need to removeall possible constraints and you
need to let us just dowhatever we want. That's been a very
powerful argument thatresonates in Washington, D.C. where
there's a fear of everythingChinese. Right. And so far, the Silicon
Valley capitalists have beenable to steer this ship kind of wherever

(22:08):
they want to go. But again,because there hasn't been that effective
pushback from the federalgovernment, they're going into these
local communities, findingdesperate communities that need tax
dollars and saying, listen, ifI bring my data center here, your
revenues in your budget,they're going to go up like 40%,
50%.
Yep.
And if you're a poorstruggling community in a lot of
these counties, that may soundreally good at first. Right. Or even

(22:30):
if you're not a poorstruggling county, because some of
the counties where this stuffis going up are not poor and struggling.
Right?
Uh, yeah.
But for a lot of thesecounties, that sounds really good.
Right? Wow. I'm going to getso much tax revenues. And of course,
what you're not thinking aboutare the major consequences and implications
of your actions down the roadbecause you're so desperate for that
money right now.
Yes.
In effect, a lot of theseplaces and local politicians, they're

(22:51):
being bought by SiliconValley. And sure, there is some local
pushback. I gave someexamples, places like Becker, Minnesota
and others that are fightingback. But it's hard right now. The
scales are tilted. Unlessthere is the different sort of government
in Washington, D.C. that seesthis from a different perspective.
Right. That it's not justabout, oh my God, if we don't do

(23:11):
this, China will get ahead ofus. It's okay. What are we doing
to our health? What are wedoing to our public water systems
and our air and our rivers andour lakes and all these things. Right.
So, yeah, it's quite a sad situation.
So one of the things thatjumps out at me, and I know folks
that listen to this programhave heard me talk about this with
both Randy Ray and others.Quite frankly, anyone that has paid

(23:33):
five minutes of attention tomodern monetary theory understands
federal government as thecurrency issuer and states as currency
users. And so the states arein a dog eat dog fight race to the
bottom to either cut their taxbase so that they can lure businesses
in, or do sweetheart dealswith the devil like this so that

(23:56):
they can bring tax revenue inthrough these kinds of data center
acquisitions. And people justdon't understand how vital it is
to understand that currencyissue or currency user relationship.
Because while the federalgovernment has the ability to fund
every one of these stateinitiatives or to fund green energy,

(24:18):
or to fund whatever it needsto fund the fake news belief that
the government is drowning indebt and can't do anything but fuels
more of this federalizationwhere the states are left to fight
for themselves with everincreasing austerity at the federal
level. And so that makes thesestates literally jump at the chance

(24:39):
to bring in a data center likethis, regardless of how many people
it kills. And they willliterally tell you, hey, this is
a great opportunity for ourcommunity. We're going to bring in
this thing. It's a greatopportunity. We're going to bring
medical waste into ourcommunity because it's a great opportunity
for tax revenue. We're goingto, we're going to bring nuclear
waste into your backyardbecause it's a great opportunity

(25:00):
for revenue. They don't tellyou the negative stuff. And it's
up to a very informed publicto fight back. Kind of like the banking
politics of the early 1900s.We've got to have an AI politics.
And I got to be honest withyou, a lot of us, including myself,
find ourselves having to usebits of AI here and there because

(25:22):
not being a Luddite, there isan element of you kind of have to,
they've incorporated this intoyour work. It's impossible to do
certain things. Now if you doa Google search, you get an AI response
as your first thing. I mean,yes, I'm sure you can use Duckgo
or whatever things out there,but the point is that the average

(25:42):
person still says Google that.So they go to Google and they've
already used AI whetheranybody wanted them to or not, whether
they're the most green,wonderful human being on the planet.
They've still used AI. It hasbecome ever present, it has permeated
all aspects of life at thispoint. And obviously I don't need

(26:02):
to get into the deep fakes outthere that it's doing, but talk a
little bit about the local andstate Governments. This was really
important to me. I appreciateyou bringing it up. I don't think
enough people understand thedivision point between what drives
the federal government, whichis not revenue constrained, and state
governments that areabsolutely in a fight for their existence.

(26:26):
And so you've got thishorrible relationship where you know
you're killing people, youknow you're doing bad things, and
yet at the same time, withoutthose revenues, the state funds dry
up, schools stop having money,you try to raise taxes on people
and the rich people move to alow tax area. I mean, it's just a
lose, lose race to the bottom.Can you talk a little bit about the

(26:47):
impacts that these AIinvestments are having in terms of
state bond budgets or state funding?
Yeah, absolutely. And I agreewith everything you just said. It
points to obviouslyfundamental issues with the political
economy of the United States,especially when people view the federal
government as somehow fiscallyconstrained. And they don't realize,
of course, that the governmentis the currency issuer and has enormous

(27:10):
powers to monetize debt and toexpand the money supply and to essentially
create the money that it needsfor any initiative. Right. The only
real fundamental constraintsare biophysical resource constraints
at the real world level. Butunfortunately, yes, our ruling class
thinks that money is aconstraint. There's a fixed supply
of it, that it's kind of likegold or oil or uranium. And obviously

(27:31):
that's not what money is.Right. Money is fundamentally a social
relation. It can be negotiatedover, bargained, created, destroyed.
So money is very differentfrom the physical things that we're
used to in our regular world.And it leads to this situation that
you described precisely, whichis a lot of these state and local
governments, they're very cashstrapped, they're very desperate.
Where are they going to go?Well, they have Silicon Valley doing

(27:53):
hundreds of billions ofdollars in capex and trillions and
dollars over the next fewyears that's going to go up to and
they're going to get a lot ofka Ching. That's what they see, right?
They see the money sign inthese next few years and a lot of
them are persuaded that hey,we gotta do this, otherwise where's
the money going to come fromfor the police officers and for the
schools? Yeah, you're right.They absolutely feel so desperate.

(28:13):
And because they're in thissituation, it's kind of easier to
divide and conquer. For thecapitalists in Silicon Valley, it's
much harder to take on thefederal government or a federal government
that's skeptical of yourinvestment decisions and the strategic
direction you want to take theeconomy than it is to divide and
conquer at the local level.Right. Where you can pay off politicians
or promise people nice taxrevenues for schools and things like

(28:35):
that. So they're absolutelyexploiting this wider weakness in
American federalism. I mean,that's ultimately, I think what this
is pointing to is that thestate and local governments can very
easily be exploited andmanipulated by capital, even when
they're established. Wherecapitalists are established somewhere,
they can threaten a capitalstrike. So they can threaten to take
away their business and thejobs. Right. Unless you do this thing

(28:56):
for me, unless you give meeminent domain rights so I can bulldoze
these businesses and homes andexpand my data center or expand my
factory or whoever wants toexpand, whatever. Capitalists often
threaten that. Right. And it'sknown as a capital strike. And yeah,
a lot of these local and stategovernments, I think, in general
are powerless against itbecause if you have capital mobility
and capital can go anywhere,you can only stop that at a higher

(29:19):
level of power. Like thefederal government, right?
That's right.
If the federal government wasactually interested in whatever's
going on with AI, which itreally isn't, beyond just saying
we need to invest a lot in AI,I don't think there's anybody in
the federal government beyondthat who knows what is actually going
on or what's happening on theground, what these systems are doing,
what they're used for, whattheir consequences are. So you're
right. It's a very, very, verysad situation.

(29:42):
So when you talked about powergrids, I think this is really important.
I mean, we can go back to turnof the century, even Y and so forth.
We've been talking about howvolatile the power grids are in the
United States. Quite frankly,the entire infrastructure of the
United States, it was builtduring a time where people weren't
asking about debt anddeficits, not in the way they do

(30:03):
today, which has beenweaponized and leveraging the absolute
ignorance of the public abouthow money works. And so they've been
able to get away with anoutrageous amount of things to say,
you know, really stupidthings, and nobody has any pushback
is they just don't understandthe subject. Right. I mean, politicians
don't understand the subject.The powers that be understand full

(30:26):
well what a gullible,uneducated group of people can be
led to do. And as a result,the power grid has basically not
been tended to, and we don'thave any kind of central understanding
of power and economy, or Ishould say, you know, the electrical
grid, for example. What a Hugenational defense issue. This would

(30:48):
be, if you really care aboutnational defense, how easy it would
be to really damage the USWith a strike on these antiquated
power grids. And I think it'simportant to discuss how power is
distributed electric power,that is, folks, the stuff we power
our refrigerators and ourhomes and our apartments and so forth.

(31:08):
Can you talk a little bitabout the impact? Guess. Describe
the power grid as it is todayand describe the impacts this is
having on. I know you talked alittle bit about the brownouts, but
let's go deeper into that.
Yeah, absolutely. And so theUS Power grid is humongous but relatively
disconnected. There's a bunchof different regions that have their
own, like, transmission lines.And Texas, as you might know, has

(31:30):
its own, like, independentpower distribution system. It's not
really connected to otherstates because it doesn't really
want to be, you know, itdoesn't want to, quote, unquote,
suffer from federalregulations as it sees the situation.
So the way it basically worksis we have power plants, whether
those are fossil fuel powerplants, like coal plants or natural
gas plants, or more renewablesources like nuclear power plants,

(31:53):
or obviously wind farms andsolar farms. They generate the electricity.
So they generate theelectricity. Then we have transmission
lines. They carry it acrossthe United States. And then there's
all these substations withtransformers and transformers. Take
the high voltage thatoriginally comes out of the power
plants, and then theyessentially reduce it. So it's safe
for you to use at home. So thevoltage that you're actually getting

(32:14):
at home, it's way, way lowerthan what's actually produced at
a power plant, becauseotherwise it would be unusable. Right.
It would just destroy yourhouse. So it has to be brought down
and controlled. And that'swhat all these, like, substations
and transformers do. Youmentioned how old the grid is and
all the problems that it has.The average transformer in the United
States is 38 years old, right?It's 30 years. A lot of these things
haven't been replaced ortended to in ages. And it speaks

(32:38):
to broader issues, of course,with American infrastructure, not
just with our power grids.But, yes, this is one area where
it's absolutely relevant is wehaven't had the necessary investments
as we've needed in our powergrids. What you had over the past
couple of decades is totalelectricity production in the United
States flatlined or stabilizedbecause just the extra demand wasn't
there. So there just wasn't alot of new demand. And power companies

(33:01):
thought, well, we don't needto produce a lot more. And so they
didn't. And now of coursewe've had this huge mad scramble
towards AI and data centers.And now suddenly there is a lot more
demand, but the productioncapacity isn't really there. And
so there are a lot ofutilities that are scrambling to
catch up. But as you know, itcan take years and years to build

(33:21):
a power plant, right? It's notsomething that you just turn on tomorrow.
And because you can't justturn to it tomorrow, what a lot of
these hyperscalers, you know,Google, Amazon, Microsoft. What a
lot of these big data centercompanies are doing is they're turning
to their own sort of smallindependent power generation units
that operate on site. So Imentioned some of them, like diesel
generators for backup power,natural gas generators. Some people

(33:44):
are interested in smallmodular nuclear reactors, although
right now those haven't reallybeen commercialized at scale yet.
You are listening to Macro NCheese, a podcast by Real Progressives.
We are a 501c3 non profitorganization. All donations are tax
deductible. Please considerbecoming a monthly donor on Patreon

(34:06):
Substack or our websiterealprogressives.org now back to
the podcast.
But the point is that theywant these independent power units
because one, a lot of thecapacity isn't there on the grid.
And when it is there, it leadsto a lot of issues with instability.
And it's gotten so bad thatsome of these companies have gotten

(34:29):
so desperate that they'reprobably skirting laws and regulations,
right? So when Elon Musk wentup and started Colossus last year
In Memphis, those 35 naturalgas generators collectively produced
enough power, almost 100megawatts. They produced enough power
that they were large enough tobe covered under the Clean Air act.
Which means they need permits.So they need official permits and

(34:50):
a bunch of other checks. AndElon Musk's company xai basically
ignored those permits, didn'tget them. They argued that, well,
these systems are kind of likeindependent, autonomous, they're
portable. They're not really apower plant like a large power plant
that should be covered underthe Clean Air Act. And so they just
got these systems installed onsite without any permits, any environmental

(35:11):
reviews, again, likelyskirting a lot of laws and regulations.
So not the only place whereit's happening. But the point of
that story is that they'regetting so desperate to produce energy
and electricity because theycan't get a lot of it from the regular
utilities that they'rebypassing a lot of Laws and regulations
in order to make it happenbecause they're so desperate to scale

(35:31):
up these data centers. Youknow, the bigger the data centers,
Obviously the more GPUs andcomputer chips you can put in there,
the bigger the models you cantrain, the more inferencing you can
do, right? The more things youcan answer from users and stuff like
that. So everything is bigger.And it's gotten to this point where
just laws and regulations arebeing ignored. Obviously a lot of
people's lives are going to beharmed along the way. And it's just

(35:52):
absolutely crazy when you.
Look at the regulatory bodiesin this country alone. And I'm not
going to speak to the rest ofthe world because I don't understand
the rest of the world'sregulatory bodies nearly as well
as I kind of do ours. And oneof the things that jumps out at you
is there's a lot of wellintentioned people that think, hey,
you know, the government couldjust regulate this and so forth.

(36:14):
And the government has theseregulations on the books. But what
they do is they severelyunderfund the actual enforcement
of those regulations and thenthey allow business to decry the
cost of maintaining thoseregulations. And then in the absence
of having anyone that canactually manage or control or say,

(36:35):
hey, you know what, you'regoing to follow these regulations.
And if these regulations makeyour business untenable, maybe your
business isn't worth keepingaround, maybe your business isn't
good for us. But theyfundamentally don't understand that
when you underfund whateverthat thing is, right, if I underfund
the people delivering food,but I make a food program, you're

(36:57):
going to see very bad resultsbecause I didn't fund the necessary
elements of it. I didn't planit out, I didn't ensure the real
resources are there. So theydo these things with the regulatory
bodies to say, see, we didsomething, but they intentionally
underfund the enforcement.That makes it like a toothless non

(37:17):
thing. A thing that we can allclap to ourselves and feel very smug
and self congratulatory forthe hey look, we the Democrats went
ahead and passed thisenvironmental regulation. Yeah, but
tell me about the enforcement,tell me about the enforcement. Tell
me about the apparatus thatensures that any of this stuff has
teeth to it. Oh yeah, I meanyou could look at what happened with

(37:41):
all the regulating of Wallstreet that is so fraudulent and
just pathetic and thedismantling of all the different
boundaries, if you will,between investment and banking, it's
just completely out thewindow. And you look at it for this,
for energy, and every singlething that they could regulate. It's
not that the regulations arebad, maybe they aren't. It's that

(38:03):
there's no actual meaningfulenforcement. There's no actual meaningful
way of exacting a price orholding the line on those things
because there is no support atthe top. Like you said, the federal
government literally does notsupport these moves. We talked to
Bill Black, who notoriouslywas key to bringing down the Keating

(38:24):
Five years ago, and every timehe knocked on the door trying to
enforce the regulations thathe was in charge of enforcing, he
was not supported. It's alwaysbeen this way. And it's a power dynamic.
It's like the rules are therefor the little people that don't
have any power. They're notthere for the big people who do.

(38:46):
And when you look at thingslike what we're talking about here,
these data centers in generaland AI overall, they're not there
to serve we the people in anyway, shape or form. And so the regulatory
bodies are naturally going tobe starved, underfunded and toothless.
What is your experience here?Obviously, the federal government

(39:06):
is all in, especially nowwithin this Trump administration,
but I don't think it was anydifferent under Biden. I really don't
think there was any differencewhatsoever. I think that the Red
scare, part two, now China,Sino scare, and the militarization
of this country in terms ofanybody out there that could possibly
leapfrog us, instead ofworking collaboratively or cooperatively

(39:27):
or thinking in terms of what'sbest for everyone, you know, let's
be fair. They are literallyensuring that there is no meaningful
way to stop this freight train.
Yeah, you're absolutely right.And it's funny you said there was
nothing different under theBiden administration, and I think
that's true at a practicallevel in terms of regulatory enforcement.

(39:48):
But what was different was thetune that the Silicon Valley capitalists
were singing. So back then,you might remember, Sam Altman was
quite excited about the idea,or pretended to be excited about
the idea about AI regulation.So this was a big talk of the town,
you know, when ChatGPT firstburst onto the scene and shortly
after, we need regulations tocontrol this stuff. And part of that

(40:10):
was the new emerging gen AIindustry wanting to seem respectable
and accepted. And part of theway you do that is by having federal
regulations. But once Trumpcame into power and Biden left, they've
kind of changed their tune.Now they don't really want regulation
now it's sort of like, leaveus alone. So we can best decide what
to do because we're the onlyones that know what's actually going

(40:31):
on. So Silicon Valley has kindof changed its tune on whether they
even want a regulatoryframework on AI and on what kind
of framework they even want.And so now when you have states like
California pushing proposalsto provide some more regulations
on these companies, there's alot of resistance now and hesitation
and skepticism from SiliconValley. So I think that's going to

(40:53):
be an emerging battlegroundgoing forward. To what extent are
you going to have regulationat the state and local level as opposed
to the more federal level?
We have talked at some greatlength about the climate crisis that
we're experiencing and soforth. And there's nothing, by the
way, going on to change that.Like all systems are go in terms

(41:14):
of climate destruction. Whatexactly do you think will happen
at the state level? I mean,they don't have the resources really,
or maybe they do, and I justdon't understand. But what can a
state do to really stop this?
You know, unfortunately, notmuch. Right. Because even if some
states stop it as localcommunities have stopped it, other

(41:35):
states are going to go inthere with open arms and say, you
know, hey, this bringsconstruction jobs, this brings, you
know, tax revenue. So come onin. And that's what's happening right
now. But I'm glad youmentioned climate change and global
warming, because I did want totalk a little bit about the global
aspects of this as well. And Ithink some of the things that people
are fundamentallymisunderstanding when they talk about

(41:56):
the rise of AI and when theytalk about the rise of data centers.
So right now, what a lot ofpeople will tell you, like the International
Energy Agency or people fromour world in data, like Heather Richie
and all that, they'll tell youdata centers and crypto mining together,
they're only responsible forlike 2% of global electricity consumption.
And even in the next fiveyears, the projection from some of

(42:18):
these organizations andcorporations are calling for it to
double. Right? So you might goto like 3 or 4% of global energy
consumption by the year 2030.And a lot of people are kind of dismissing
those numbers as sort of,well, sure, they're growing fast,
but in the grand scheme ofthings, they're kind of irrelevant.
And I think a couple of majorthings people miss in this conversation.

(42:40):
One is that electricityconsumption is not the same thing
as energy demand. Right?Because electricity is not the only
kind of energy that's impactedin what we're talking about. So there's
also construction andtransportation and burning Fossil
fuels that are associated withthe rise of data centers. And there's
a lot of complex upstreamfactors that are involved in making

(43:01):
data centers happen in thefirst place. So let's just go through
some of them. So one,obviously you have to build a data
center. Who's doing that?Well, construction companies are.
And what are they using? Well,they're using bulldozers and excavators
and cranes and things likethat. Tandem rollers. Most of those
things are powered by dieselengines. And so they're emitting
greenhouse gases and otherharmful pollutants as they're building

(43:23):
these things. Again, ties bothto the global warming angle and the
public health angle. Right. Sojust the mere construction of these
massive data centers producesa lot of pollution. That's one issue.
You produce a lot of thesepollution from these data centers.
But another problem is thatthen you have to put specialized
equipment in these datacenters. So you got to put the servers,
the server racks and thenetwork switches and the computer

(43:45):
chips and take Nvidia's GPUs.Right. Where are they produced, Steve?
They don't just magically showup at the data center. You know,
most of them are produced byTSMC in Taiwan. Right. So they're
produced all the way in Taiwanby TSMC as these massive production
facilities, foundries overthere. And obviously that's using
a lot of electricity inTaiwan, but then they have to be

(44:06):
shipped and transported andother things. Now, all of that's
using energy. And think aboutwhat it takes TSMC to make a lot
of these advanced computerchips. It takes a lot of highly specialized
devices and equipment likeextreme ultraviolet lithography machines,
which are used for thoseetching patterns on the silicon wafers.
So you need a lot of thesespecialized machines in order to

(44:27):
build, you know, very advancedchips that have a lot of transistors.
They're very, very, very tinytransistors. And where does TSMC
get a lot of those specializeddevices? Well, he doesn't produce
them himself. He gets themfrom other companies around the world.
So he gets a lot of them fromthe ASML in the Netherlands. That's
what produces the world's mostadvanced EUV lithography machines.
And this is fascinating. Thisis from ASML's own website. It takes

(44:50):
ASML to transport one EUVlithography machine to TSMC in Taiwan
takes ASML 40 freight tankers,three cargo planes, and 20 trucks
on average. Wow. And that'sbecause these lithography machines
are, you know, these hugedevices and they have to be disassembled
into various components.Right. And so all of these things,
the cargo planes, the merchantshipping, the trucks, they're basically

(45:13):
shipping different componentsof this thing to tsmc, and then it's
reconstructed on site. Right?And then TSMC uses it to make GPUs
and other advanced chips. Soobviously, you can imagine there's
a lot of, again, greenhousegases and pollutants associated with
all of that transportation.And ASML itself has over 5,000 different
suppliers around the world. Soit needs all of these like sensors

(45:35):
and lasers and otherspecialized equipment to produce
its own machines. Right. I wasjust giving you one example, but
you can see where this isgoing, and where it's going is that
our global energy networks arehighly complex and extended and interdependent.
That's especially the casewhen you're talking about the semiconductor
industry. So the semiconductorindustry has the most globally extended

(45:55):
supply chains of any industry.And it's got its tentacles kind of
everywhere. Right. So if onecompany is doing something over here,
it needs to source a milliondifferent parts and components from
companies all over the world.And that's true to a large extent
with a lot of other industriesas well, like the auto industry and
so on. But again, it isespecially true with the semiconductor
industry. And that's relevanthere because that's where a lot of

(46:16):
the specialized equipment anddevices that this industry is producing
is going into these datacenters. So I think this is part
of the story that's oftenmissed when people talk about, well,
what are going to be the extraenergy demands associated with the
rise of AI. It's really acomplicated question to answer. I
mean, you would really needsomebody doing just a whole PhD thesis

(46:39):
just on this subject to trulyanswer it, because it's not as simple
as saying, well, they consume2% of the world's electricity. That's
not such a big deal in thegrand scheme of things, really. Once
you include, of course, thefull energy spectrum, all the different
ways that we use energy.Right. And all the different emissions
that we have and all thingslike that. The rise of AI and data
centers, I think likely ishaving a much bigger impact than

(47:02):
that. And of course, then I'malso leaving out the downstream effects.
Right. So all of us now thatare using AI in our lives, well,
when we're at home or at work,who's powering that equipment? Right.
So it's our utilities, right?We get electricity from the grid,
so they have to power ourdevices. And we're using them more
now because we've got to ask,you know, chatgpt you know, a million

(47:22):
questions or like you said,people need to use it for work because
a lot of these agentic AIsystems are coming online at work.
So it's leading to a lot ofthe downstream effects as well. And
it's crazy how fast this stuffis spreading around the world. Steve,
I gotta tell you, I'm justlike flabbergasted by some of these
numbers. Saudi Arabia, inorder to ingratiate themselves with
the new administration inWashington, is now building out dozens

(47:45):
of massive data centers in thedesert, right? So you can imagine
what a great idea that is in awater parts region. They've inked
deals with Nvidia where Nvidiais going to ship them their latest,
you know, Blackwell serieschips. So Saudi Arabia is getting
all the latest stuff eventhough it's not a tech powerhouse,
even though it doesn't haveany need to this scale for cloud
computing. But it's bringingon these just massive investments

(48:07):
as part of their Vision 2030initiatives. And it's just crazy.
And it's happening all aroundthe world. Brazil is also now constructing
dozens of hyperscale datacenters and Japan's constructing
a lot of them. China obviouslyis having a massive build out and
it's just crazy, right? Andagain, when you look at how entangled
all of these energy networksare. So this is one of the fundamental
points I brought up in thebook, is that our energy networks

(48:29):
are highly entangled. So whenwe change conversional devices or
things over here can have alot of complex upstream and downstream
effects in other parts of thenetwork because all of these things
are interconnected together.When you do your ChatGPT query, Nvidia's
chips have to calculate andgive you the answer. But Nvidia's
chips are produced by TSMC inTaiwan. But in order for TSMC to

(48:49):
produce them, it has to getstuff from ASML and a bunch of other
people. And in order for ASMLto do that, they have to get equipment
from other people. It's all sointerconnected that I think people
don't appreciate the scale ofthe changes that are happening when
they cite these throwawaystatistics about, well, it's only
2% of electricity consumption.That's a common way that it's dismissed.

(49:10):
And I think, just as I justexplained, I think that way of reasoning
is so flawed it's dangerous toglobal civilization if we keep thinking
along those lines.
I gotta say, just for kicksand grins, you know, as you're talking,
I was looking up someinformation about ASML and there's
great Yahoo Finance article.An analyst explains why Nvidia China

(49:31):
News could be huge for ASMLHolding. And I'll just read this
one quote, because they'redown 18%, okay? And obviously earlier
in the year the US had said,you can't sell these Nvidia chips
to China. But now all of asudden US government has backed off
of that. And I'll just readthis just for the purpose of filling
in some gaps here. This guyJoe Tagay from Equity Armor Investment

(49:56):
said in a recent program onSchwab Network that the US government's
decision to allow Nvidia tosell chips to China would be huge
for ASML holding, nv, nasdaq,asml. He explained why the company
is important in the AIindustry. ASML has been a company
I've been following for thepast few years. I'm obviously a big
fan. It's been behind the chiprevolution. It makes what people

(50:20):
have called the mostcomplicated machine humans have ever
built. And as you can imagine,it is very expensive to produce.
They are essentially themachine that makes the machine. So
we can't get these Nvidiachips without these ASML machines.
And I think this China Newscould be really huge for the company.
Remember, they took a bigwrite off, big leg down when the

(50:41):
China News came out thatNvidia will not be selling to China
early in the year. So it'sgoing to be interesting to see anyway,
point is that I wonder whatthe impetus was. I mean, I guess
capital won and said, you'regoing to let us sell to these guys
no matter what. But obviouslythat's a big deal. That's a huge
deal.
Yeah. And the impetus was thatJensen Huang talked to Trump and

(51:03):
convinced him to let him sellsome low end chips to China under
the argument that, well,they're so less advanced than the
latest stuff that we'rebuilding, that it's not really a
national security threat if wegive this stuff to China. Right?
It was like the H20 chips.Those are much older variants. This
is not like the Blackwellarchitecture, which is Nvidia's latest
AI architecture. Right. Packs200 billion transistors in a single

(51:27):
chip. It's the world's fastestcomputer chip. So that's not what
China is getting. That stuffis still under export controls. China
is getting some of Nvidia'searlier AI chips which are still
powerful but still kind of, Ithink, not as concerning to the US
security establishment. I willsay though, China itself now is building
their own AI chips, Right? SoHuawei is Doing this, they've already

(51:48):
gotten some chips out. SoChina is having a massive AI buildout
too. And because of all theexport controls that the United States
targeted China with over allthese years, China has put enormous
investments into developingits own domestic semiconductor industry.
And that has paid offenormously. So the Chinese are now
producing huge quantities ofhigh quality chips and across sort

(52:11):
of the market spectrum. Right.Whether it's car chips or, you know,
AI chips or chips for computerchips, smartphone chips. So the semiconductor
industry there in China isbecoming highly independent and highly
diversified. And that was inlarge part a response to the sanctions,
to the export controlslaunched by the United States as
a way of choking off China'stechnological development. So that

(52:33):
was Washington's sort oforiginal goal. But now I think there's
a lot of politics happeninghere too with the trade deals and
the negotiations that arehappening with China. This was probably
kind of like a bone to theChinese as well, like, hey, yeah,
you can have some of thesechip. But yeah, I think, yeah, the
original impetus was JensenHuang talking to Trump.
So as we come towards theclose here, what are some of the

(52:55):
important things that maybe wehaven't discussed yet that we should
maybe bring out onto the table?
Yes, I think the importantthings are that we've talked a lot
about what the problems are,so we've talked a lot about what
the local consequences are,sort of the global consequences and
the global dimensions of this,and we haven't talked much about

(53:16):
potential solutions. Right.What could we do ideally in the face
of all of this? And that's notto say that it's going to get done
because the politicalenvironment isn't there. So I don't
think any of what I'm about tosay is going to get done in the short
term, but thinking more longterm, like where should we set the
boundaries of AI and energy?Should we just continue operating
as if there are no boundariesand just keep building out data centers

(53:38):
for the rest of the century?Obviously, Steve, we've discussed
my book before. My answer tothat would be no. Right, so then
what kind of rationalconstraints and limits should we
be imposing on this build outon these systems? And I think there's
a lot of different things thatwe could talk about here in terms
of constraints. So one kind ofconstraint that a lot of people like
NOAM Brown at OpenAI havepointed out is that AI systems, you

(54:01):
can generate huge improvementsin performance just by letting them
think a little bit longer. Andso you don't need to train these
humongous systems. The waySilicon Valley is doing, right, like
GPT4, OpenAI's latest, whichis consumed as much electricity to
train, I think as likethousands of American homes in a

(54:22):
year, thousands of typicalAmerican households in a year just
to train ChatGPT4. Right. Andthe idea is that, well, we need to
spend so much computingresources because that's how we'll
get a boost in performance.Right? These are known as the AI
scaling laws. And the thingis, there are ways to improve the
performance of these modelswhere you don't have to train them,
you don't have to train suchlarge models. Right. And an easy

(54:43):
way is just to let them thinka little bit longer. They get a huge
boost you in of frontperformance just from that. If you're
willing to wait, you know, twoseconds for the answer instead of
like one second or something.Right. I'm just tossing out a tongue
in cheek example, but you getthe point, right? And that'll save
enormous amounts on thetraining side. Now it might consume
a little bit more on theinferencing side, but for that you
could do other things. Likeyou can impose rational curbs on

(55:06):
supply and demand. So youcould tell people, hey, from 10am
to 9am or something, I don'tknow. You can't use ChatGPT, obviously.
OpenAI is never going to dothis willingly, of course. What I'm
talking about is policies,right? Social policies, public policy.
What is the role of that?That's one thing you could do is
you could just limit people'saccess to these AI systems and that

(55:27):
would scale down energy useenormously. I don't need an AI overview
for Google every time I searchfor something, that's what Google
gives me now. And Google'sdoing that because it doesn't want
to be left behind. It doesn'twant Everybody going to ChatGPT or
to some other system that'sgiving them an AI response. And so
now Google has essentiallyinsidified, you know, to quote Cory

(55:49):
Doctorow, it's inshidified itsexperience and it's made every single
answer pop up with the AIresponse first because it doesn't
want to lose its potentialcustomers to ChatGPT. It doesn't
want people going away fromGoogle the search engine and replacing
something like Deep SEQ orchatgpt for their search needs. Google
wants to stay. So, you know,it's put AI for everything. You could

(56:10):
do things where you could belike, hey, you don't need an AI response
for everything. Google oranybody else and the correspond for
that. If you're worried aboutsome kind of advantage. We can impose
limits on how often people canuse ChatGPT or something. Right,
If Google's worried aboutthat. But the point is you can do
things like that. You can curbsupply and demand so that we're just
not using it as much. Once youdo that, then you can begin to think

(56:32):
about a more comprehensivesolution about, well, what are the
actual data center resourcesand other resources that we need
in order to sustain sort of amore stable AI regime. Because I'm
certainly not an advocate forjust getting rid of generative AI
and agentic AI and not usingany robots whatsoever. I don't think
that's realistic. I thinkthere's a lot of fundamental things

(56:53):
that are changing in theworld, like the world's population,
especially in more advancedcountries, is aging. So it's nice
to have robots so you canoffload some of that labor and tasks
when our populations start todecline, as they're already happening
in many areas. So you may knowalready that Japan is a big proponent
of robotics. There are a lotof robots in Japan, and Japan's population

(57:13):
is coming down. Right. So ifyou're facing these long term demographic
challenges, AI and roboticscould be an important part of how
we address that so that westill get the economic resources
that we need to people. So Ithink there is a place for AI in
our world. I'm not saying AIall bad, so let's just destroy all
the data centers and get ridof them. No, that's not what I'm

(57:34):
saying. But I do think we haveto be careful. Well, we certainly
can't keep going thisdirection where we just double data
center usage every five yearsor so. I think that's courting ecological
catastrophe. It's courtingmany public health crises, a lot
of economic crises with jobsand things like that. It's the kind
of rapid, poorly plannedchange that will lead to a lot of

(57:55):
social and politicalinstability and a lot of harmful
consequences for globalcivilization as a whole. So I think
we absolutely need to thinkabout how we can have AI in our world
in such a way that it doesn'tbring down the ecological stability
of the biosphere and in such away that it doesn't induce massive
political and economic crisisin human society.

(58:16):
You know, I brought it up alittle earlier, but I am serious
about this. I know that thereare deep water filtration plants
that can be used to bringclean, potable drinking water to
communities. And God knows weuse pipelines for fuel. What would
prevent us from Doing deep,well, desalinization type, you know,

(58:38):
energy, you name it. I mean,I've seen some of these massive units
out there that both usenatural elements of the wave to create
the energy. It also at thesame time cleans or desalinates the
water so it's drinkable. Whatwould prevent something like this
versus destroying the aquifersin these local areas? Again, I'm

(59:00):
a nobody. I'm a noob. I'm justcoming up with things.
It's a great question. Andyou're right, a lot of the resources
are out there. A lot of thatis already happening. So some of
these data centers are usingwastewater from treatment plants
and things like that, and wecould build more. Certainly one area
where that's a problemspecifically with data centers is,
like I said, they'reinterested in going where the power
is cheap. And sometimes wherethe power is cheap might be some

(59:22):
random rural community out inthe middle of nowhere. And unfortunately
there they may not have awastewater treatment plant or desalination
happening nearby or. And soyou kind of need to use the local
water supply that you have,and you may not have time or resources
or money to run it througheverything else. And so that's happening
a lot, unfortunately. And it'swhat's leading to a lot of these

(59:44):
water shortages and waterquality problems that I mentioned.
And I think what your pointultimately highlights is the need
to have a more comprehensivestrategic direction so that we're
not building data centers inPhoenix, Arizona, which is where
many of them are going rightnow, which, as you know, is the western
US has been struggling withmassive droughts over the past few

(01:00:05):
decades because of globalwarming. So they've been struggling
with, you know, an on and offhistoric drought for the past four
or five decades. Right. Andit's leading to a lot of water shortages,
low levels in the ColoradoRiver, a lot of political fights
and negotiations among thestates about how they're going to
divide the water. And it'slike in these areas, we're just saying,
yeah, let's build a bunch ofdata centers here and make everything

(01:00:25):
worse. Right. So you'regetting back to more comprehensive
thinking, thinking about landuse dynamics and where you want to
place things. Does it makesense to build data centers in Phoenix,
Arizona, or in the desert ofSaudi Arabia? The answer is no. That's
going to lead to disaster in acouple of decades if you keep going
down this path. And so that'swhere you go and you say, what are

(01:00:46):
the resources available rightnow? Right. Where are the water Treatment
plants? Where are the thingsthat we need or where can we build
them? Well, let's build datacenters just over here or let's rein
in the number that we'rebuilding because we want to put these
wider constraints on AI, onthe prevalence and diffusion of AI
in our world. Right. So all ofthese questions sort of have to be
tackled collectively andcomprehensively. Steve, you can't

(01:01:06):
just look at desalinationplants and sort of say, well, if
we just build a lot more ofthose, I think we'll be good with
the water.
Right.
The problem right now is youhave to think about land use planning,
because right now, since thereis no coherent direction from the
federal government, thesethings are getting built everywhere.
Basically. There's no widerthinking about whether they should
be built there. What are theexisting resources there to support

(01:01:28):
them. And in many cases, asI've just explained, there are very
few existing resources thereto support these things. So then
they're cutting laws andskirting regulations, all the things
that we talked about, right?So it's such a complex multidimensional
challenge.
If I can recap this, We'vetalked about regulation, we've talked
about energy consumption,we've talked about water use, we've

(01:01:49):
talked about the misplacedworkers, we've talked about the role
of federal and state in termsof understanding currency issue or
currency user dynamics and therace to the bottom of the ever present
desire for bringing the jobsand the taxes and all the other stuff
that goes into a fundamentallack of understanding of the public

(01:02:09):
to how money works. I mean,they are cheering on Doge, which
is slashing and burning theinfrastructure that could in fact
regulate these things. It'sjust an unfortunate truth, but that's
kind of the essence of this.And we talked about the antiquated
power grid and we've talkedabout integrated solutions versus
capital, which is large and incharge. With all that said, we even

(01:02:31):
looked at the global elementshere, including the manufacturing
of these chips and all therest of the elements that go into
the AI ecosphere. Tell me yourfinal thoughts on what you would
like our listeners to takefrom this podcast and to maybe consider
as they go forward.

(01:02:52):
I think what I want to tellyour listeners is that you're not
powerless. There's a lot oflocal communities that are fighting
back and fighting backsuccessfully against the expansion
of these data center systems.They're saying, no, you will not
build here unless you do xyz.So what I would say, especially at

(01:03:13):
this moment, where I think alot of people might be distracted
with the situation at thefederal level, because there's always
a lot of news coming out fromthe federal government and let's
just say most of it not great.So that I think there's a lot of
people who are distracted atthat level. But make sure that you
also pay attention to what'shappening in your local communities.
Right. Because before you knowit, a data center could come nearby.

(01:03:34):
And if you're not aware of howbig it's going to be, how much power
is it going to use, how is itgoing to relate to the water supply,
These are all things thatcould profoundly affect your life.
Right. And so I would sayeducate yourself, organize. Like
I said, there have been a lotof successful efforts from different
cities and counties to resisteither the arrival of these data

(01:03:56):
centers or their furtherexpansion. Even in places like Northern
Virginia, where there are alot of them already, there's some
efforts to sort of try andconstrain how they're built in the
future. So I think it's agreat opportunity to apply some political
pressure at the local level,get organized at the local level
and understand that this AIcould be coming for a community near

(01:04:18):
you, not just in front of yourscreen, but literally physically
right outside your home ornear your home. Right. It doesn't
have to be obviously rightnext to your home. It could just
be in your general communityand it'll affect your life just the
same. So, yeah, definitely geteducated, organize, agitate at the
local political level becauseyou do have some power to push back
against this.
Very good. All right, Errol,thank you once again, man. I love

(01:04:41):
having you on and I hope wecan have you back on a million more
times in the future with that.Tell everybody where we can find
more of your work. Obviously,we pumped your book. I'd love to
keep pumping your book. It's agreat book. But where can we find
more of your work?
Absolutely. Appreciate it.Yep. Substack is generally where
I write, so I have a ton ofgreat content out on Substack. I
have no plans to monetize anypart of my substack, so people can

(01:05:03):
go in there and read all theposts I've ever written, you know,
all my notes. All of that isfree. That's generally where I'm
most active. You can alsocheck out my website, erickolassi.com
and yet, like you said, mybook the Physics of Capitalism was
published earlier this year,so you can check that out, too. Thanks.
Fantastic. All right, so withthat, my name is Steve Grumbine.
I am the host of Macro andCheese. And this podcast is a part

(01:05:25):
of a larger ecosystem calledReal Progressive, which is a 501C3
not for profit. Folks, we liveand die on your support. I know there
are other groups out therethat pay all their stuff. Thank you
Errol for not paywallingyours. We don't pay wall ours either.
And sometimes you wonder iffamiliarity breeds contempt because
if it didn't cost something,is it worth anything? I think it

(01:05:46):
is if you think it is. We needyour support. You know, small donations,
large donations, again, taxdeductible. It's kind of a win win
for all of us here, folks. Youcan catch us on patreon patreon.com
real progressives. You can goto our website realprogressives.org
where we have a dropdown to goto donate. You can also go to our

(01:06:10):
sub stack Real Progressivesand become a monthly donor there
as well. We really need yourhelp folks. This is not idle chatter.
We need your help. So if youconsider the work we do here worthwhile,
please consider supporting us.And also of course, please buy Harold's
book. It is worth your time.And with that, without further ado,

(01:06:30):
we bid you adieu on behalf ofmy guest, Errol Colossi and myself,
Steve Grumbine, behalf of thepodcast Macaron Cheese for the organization
Real Progressives. We are outof here.
Production transcripts,graphics, sound engineering, extras

(01:06:53):
and show notes for macro ncheese are.