Certainly, you've seen the headlines and you probably have your own fears, but AI has raised the fears about energy and water and security in the broader economy.

In the end, I'm a techno optimist.

This is a problem we can identify and solve it.

If AI is here to stay and we're going to build the infrastructure, let's do it the right way.

Funding was provided by the State Energy Conservation Office.

Please give it up for Michael Webber.

Thank you.

Good evening.

Good evening, everyone.

Thanks for coming out tonight.

I'm Michael Webber.

I'm a professor here at the University of Texas, and I study energy, water and innovation, and we're going to talk about all of that tonight in the context of AI and the future of AI, what that means for us.

And there are a couple things I'll start with.

Certainly, you've seen the headlines and you probably have your own fears, but AI has raised the fears about energy and water and security in the broader economy.

It's in the news.

It's probably on your minds.

Is AI's energy use a big problem for climate change?

Data centers for AI use huge amounts of electricity, water, driving up costs and climate concerns.

Data centers are booming, but there are big energy and environmental risks.

AI is eating data center power demand, and it's only getting worse.

Will, AI data centers make or break the energy transition.

Is this going to move us backwards in some of the progress we were making for a better energy future?

So those in the headlines but also for Texans, there's this context of the 2021 Winter Storm Uri, which is still a traumatic memory for a lot of us.

How many of you were affected by Winter Storm Uri, just by show of hands?

I think a lot of us were.

You might have lost power or water, or someone you know was affected.

or someone you know was affected.

So that's still in our minds and, at a high level, the February 2021 freeze was a cascading failure of a variety of systems.

Here's a clip.

And that led to the shutdown of even more power plants.

So we have a water problem, freezing water, become a gas problem, become a power problem, become a bigger gas problem, become a bigger power problem, become a water problem and a humanitarian crisis.

That was me, if you didn't realize.

That was PBS Nova, doing an interview at the time about the cascading failures that happened.

Energy problems, becoming power problems, becoming water problems, becoming a humanitarian crisis.

That's in our memory.

And that's one of the concerns if we have another event like that, but with data centers as part of it, will it get worse or not?

And if we go back to the numbers of Winter Storm Uri, it was deadly, right?

200 plus official deaths.

700 plus unofficial deaths.

if you include all the unexpected deaths that happened.

10 million plus Texans without power for several days.

10 million Texans without water, it was the largest boil water notice in American history.

The largest boil water notice before that had been Milwaukee with a million people without water.

But now in Texas, we exceeded that.

We're number 1, 10 million people without water.

So we set the record.

And just a catastrophe.

And so when that's in your memory, it changes things.

And what was especially scary about Winter Storm Uri is how close we came to just complete grid collapse.

This is a chart some of you might have seen if you're an energy nerd like me, where the frequency was dipping in the grid and we came within minutes or seconds of total grid collapse.

If you have total grid collapse, it takes weeks or months, we don't know, actually, to bring the grid back on.

That's called a a black start.

So it was scary.

We remember that.

And now we have AI and data centers and they're power hungry.

So that exacerbates maybe some of the fears and concerns we would have had.

Indeed it is power hungry.

It's not just sort of sensationalistic headlines, it's the situation.

Here's a article from 2024, “AI is poised to drive 160% increase in data center power demand.” Here's another one from Vox, “AI already uses as much energy as a small country.” Now it uses as much energy as a larger country, by the way.

It's only the beginning in 2024 is what it said.

“AI's insatiable appetite for energy.” It's just growing more and more of these headlines.

As AI booms, data centers may create electricity scarcity among users scarcity, we might not have enough to go around.

We've had plenty of electricity for a long time, and now we might not have enough to go around.

Look at all this, “In a focus: data centers, an energy hungry challenge.” It just goes on and on, what are we going to do about that?

So a lot of concern for a lot of reasons.

And I comment on this in July 2024 as well.

The demand for power isn't some faraway strain of the grid, it's a here-and-now requirement that puts data centers in competition with other uses for new utility connections.

Those other uses are like you and me.

our water treatment, our population growth, our factories, or different things we might use.

So there's a competition here and we have to sort this out.

That's the fear or the concern.

And those same data centers are also thirsty.

“AI data center water consumption is creating an unprecedented crisis in the United States.” Huge numbers, unprecedented crisis.

There might be precedent, but still, it's bad.

“The hidden thirst of technology how AI is consuming millions of gallons of water.” “AI has a hidden water cost, here's how to calculate yours” As if it's your fault.

Which I think is kind of interesting with that headline, right?

Maybe you shouldn't be using it because you'll have water consumption associated with it.

“AI's cooling problem, how data centers are transforming water use.” On and on and on.

And it also might cause the collapse of civilization, by the way, or free us forever.

I don't know if you've seen these headlines.

Here's my favorite chart, Financial Times “The singularity is here.” This is a chart of GDP or economic activity over time, and it has grown for a long time.

We've been getting richer.

Economies have gotten bigger.

There's more of us.

We're active.

And with AI, we might have the end of scarcity.

We're all billionaires or just we're all infinitely rich, which is incredible.

Or it's human extinction.

We're all going to die, right?

This is a very different kind of outcome for the same technology or 0.2% economic boost.

Which one is it?

Are we can all get rich.

Are we all going to die.

Or is it just going to kind of keep our economy going the ways we want for a while?

Well, Elon Musk tweeted this, which I think is sort of interesting, “Universal high income,” Some people talking about universal basic income or UBI, he's like we don't need universal basic income.

We need universal high income.

Don't worry, we'll all be so rich.

It'll be fine.

“Universal high income via checks issued by the federal government is the best way to deal with unemployment caused by AI, because AI, combined with robotics, will produce goods and services far in excess of the increase in the money supply.

There'll be no inflation.” The guy who wanted to dismantle the federal government is now saying the federal government's going to give us a bunch of money, which is pretty exciting.

But the core of this is some techno optimism that AI will unleash an incredible amount of economic opportunity and quality of life for all of us, because AI's going to do the work.

We're all just going to have hobbies.

Why mow our lawn or cook or grow food or do anything when AI will do it for us?

But his outlook does not represent consensus view.

It also might cause the end of humanity, an extinction, right?

So this is the Center for AI Safety has a statement out that says, “Mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks such as pandemics and nuclear war And climate change, but they didn't say that.

Can you see who signed this by the way?

Sam Altman, CEO of OpenAI, Anthropic, different professors, Bill Gates, here is Igor, who has xAI, which is affiliated with Elon Musk.

So it's either going to make us so rich, we all have all the money we need, or it's the end of us and we should pay attention to it.

Maybe it's both, and it's more of a choice that we have to make.

We have some proactivity or intentionality we can have around this.

So that's AI.

It's in the news, right?

It's going to undermine our energy, undermine our water, and kill the human species or make us rich.

So let's talk about that.

It's also a national security issue.

And the AI arms race today reminds me of the nuclear arms race of World War II.

In World War II, there was a race to develop nuclear weapons, and we didn't even take the time to decide whether nuclear weapons were good or bad.

We just knew that we wanted to have it before the bad guys had it.

That's what we're doing with AI today.

We don't even have the time to deliberate, or we haven't taken the time to deliberate, whether AI is good or bad.

But we know that we want it.

If the adversary is going to have it, we don't want them to have it without us having it.

And that's something I wrote about also a couple years ago.

The offshoring potential, the ability to train your AI models offshore, the offshoring potential from the US perspective creates a national security risk.

If we don't win the race to lead on AI, then certainly someone else will, and that might be a foreign adversary.

That's the risk for us that we need to manage.

And so I say, what should we do and how should we think about these risks?

Well, let's talk about AI, what it might mean good or bad for us?

I want to give a little more background that just in general, new technologies are unnerving.

We don't like it when new stuff comes along especially if you're of a certain age and older.

We tend to be more suspicious and skeptical of new technologies, whereas younger ages like to adopt them.

Here's something I think about on campus.

I've been on campus since I was two, so campus is my life.

And when the internet came to life in the 90s and I was a grad school, it was a real concern.

What about internet searches and Wikipedia and things like that?

Is that cheating?

We used to think in the 90s you had to go to the library to look things up on index cards.

Some of you who are of a certain vintage will remember that as I do.

And then internet search came to life and we considered that cheating.

But now I can't imagine doing research without starting with an internet search.

Right?

So our view of the technology has changed, and that might happen for other new technologies that come along the way, including AI.

But this tradition goes back thousands of years.

Socrates is worried about a new technology called writing.

He's like, “Oh, that's going to be bad for us” “If men learn this new technology, writing, it will implant forgetfulness in their souls.

They will cease to exercise memory because they rely on that which is written, calling things to remembrance no longer from within themselves, but by means of external marks” They were worried that this new technology, writing, would weaken your brain because you wouldn't have to remember stuff.

Just think like the old Greek tragedies or things like the Iliad or wherever that used to be remembered and spoken.

They weren't read.

So this idea that something new will make us dumb is not a new idea.

It's been around for a while.

This new technology can be unnerving.

But new technologies can also be empowering.

It could be liberating.

“AI is the most enabling technology since electricity.” That's what Chris Crosby said at U.T.

Energy Week.

Now he's a data center developer, maybe he would believe that, but it's just interesting that some people see it as the tool we need that will liberate us, or the thing that will make us dumb and weak and insecure.

So that's the tension we had.

I would also like to remind you that AI is not a new concept.

It was actually coined artificial intelligence as a word in 1956 by John McCarthy, a professor at Dartmouth College, who pulled together, McCarthy and his colleagues, a proposal for the Dartmouth Summer Research Project on Artificial Intelligence, funded by the Rockefeller Foundation, they pulled together a group of great thinkers on artificial intelligence, and said let's talk about this.

What does this mean for society?

Machines that can learn.

And they convened a meeting and Claude Shannon was there.

And Claude Shannon is the father of information theory.

He's the guy who essentially figured out that you can use Boolean logic or Boolean algebra, sort of philosophical concepts, and attach it to ones and zeros with transistors, ultimately, that could enable the digital age.

You could use circuits to be on on or off and if things are on or off, you can do math.

And if you could do math, you can start to learn, do other things.

He coined that whole concept basically as the father of information theory, eventually at Bell Labs and other places, and he is the man for whom the AI platform Claude is named.

If you ever wonder where Claude came from, that's Claude Shannon, great, great mind and a great thinker.

His master's thesis at Michigan is the most important master's thesis in history, because it basically enabled the modern age of information theory.

So that was 1956, has been with us for a while.

And think of the computing power in the 50s.

You have more computing power in your watch and your your phones than we had back then with these large mainframes.

Any of you ever program on one of these, maybe if you were a student here in the 70s, you might have something similar.

I see some of you might have at least touched one of the machines, and certainly some of you have seen punch cards, the old ways of doing ones and zeroes and doing your programming.

So this is actually a great age test.

If people know what this is, that means you're above a certain age.

I don't want to say what age you're above.

The foundational math for artificial intelligence has been around for decades.

This is my textbook from grad school, 1999.

I took a class on neural networks.

And this is not the same math as artificial intelligence, but it's like same family of math, about math you could use to predict things and you could use math to learn.

And I took this class from Bernie Woodrow, famous professor at Stanford, his first student was Ted Hoff.

Does that name mean anything to anybody?

Ted Hoff was employee number 12 at Intel, and he was one of the co inventors of the microprocessor.

So Bernie's first student was Ted.

And I was like his last student.

So the world had fallen a long way.

So the world had fallen a long way.

I took this class.

I have no idea what neural networks are.

I got an A because I repeated back to professor what he wanted to hear, and by the way, that was actually the point of neural networks in math.

This pattern recognition, repeating back to people what they want to hear.

That is the basis of AI.

So even people like me who aren't AI people, I took classes on this decades ago, and I've even been talking about AI and energy for a while.

This is a picture me in 2019, in the United Arab Emirates.

In Abu Dhabi, I'm interviewing His Excellency Omar bin Sultan Al Olama, who is the Minister of AI for UAE.

So as of 2017, this is two years later in 2019, UAE had someone in charge of AI.

We still don't have someone in charge of AI in the United State but there are other countries that are further ahead on this.

And he was the guy in charge of AI and I got to interview him on a stage in front of a hundred--- I promise you, there were hundreds of people there It was empty behind us, but it was like, it's like a full audience, really, they were hanging on every word.

But we talked about what does AI mean for energy and we're talking about the energy use cases.

What about the energy consumption?

So even the idea that AI might be important for energy as a way to get solutions or a way that might exacerbate the grid, even that's not a new idea we've been talking about it for a while, and we spent a lot of time talking about what jobs are at risk.

And this was really fascinating because there are different kinds of jobs at risk from AI than from other waves of history we've had.

And I told him that I thought AI would be unlike other waves of mechanization, because usually when you get new tractors, they offset or displace lower wage earners on the farm or new machines in the factory, offset or displace lower wage earners and then AI might come in and displace $1,000/hour lawyers or higher pay people instead of the burger flippers.

So it's going to have a different effect on the labor supply chain.

And I said the one job I thought would be safe would be a joke tellers.

That the comedians would be the one job that's left, because you have to be smart and improvise.

And it turns out, tell me a joke is actually a great use case for AI.

Tell me a dad joke that rhymes, or something like that.

Anyway, so I was totally wrong on that, but it was interesting to have this conversation in 2019, what does AI mean for energy?

He was recognized as one of the 100 most important people in the world So he's been on this issue and thinking about it.

So he's been on this issue and thinking about it.

And for me, this fits into this broader wave.

These are headlines from 2016.

In 2016, there's already a sense like, uh oh, AI is going to be different.

It's going to displace different people than we're used to.

It's going to be disruptive.

“March of the machines,” a special report on AI in The Economist June 25th, 2016.

So this isn't even like a here or now issue, like it's on our minds a lot the last couple of years.

But people have been seeing this for a while and going, uh oh, maybe this is bad, maybe this is good, but it's different for sure.

And if you back up to the 80s, after the rise of the personal computer, our movies, our popular culture, even captured this we had a bunch of movies pop up about the technologies that might kill us with Blade Runner and The Terminator and WarGames.

And I want to play you an extended clip from WarGames.

So forgive me for how long this clip is, but this movie holds up well and it captures some of the key points.

So let's play this clip.

Because... What's so special about playing with some machine?

Oh, no.

No, it's not just a machine.

Here, look at this.

This is a tape that I got from the library.

It's about this guy named Falcon.

He was into games as well as computers.

He designed them so that they could play checkers or poker, chess.

What's so great about that?

Everybody's doing that now.

Oh, no.

No, no.

What he did was great.

He designed his computer so that it could learn from its own mistakes.

So they'd be better the next time they played.

The system actually learned how to learn.

It could teach itself.

Intiate internal power.

Generators on and functioning.

External power disconnected.

It's a bluff, John.

Call it off.

No, it's not a bluff.

It's real.

Hello, General Barringer.

Stephen Falcon.

Mr.

Falcon, you picked a hell of a day for a visit.

General, what you see on these screens up here, is a fantasy, a computer enhanced hallucination.

Those blips are not real missiles.

They're phantoms.

Jack, there's nothing to indicate a simulation at all.

Everything is working perfectly.

But does it make any sense?

Does what make any sense?

That.

Look, I don't have time for a conversation right now.

General, are you prepared to destroy the enemy?

You betcha.

Do you think they know that?

I believe we've made that clear enough.

Then don't.

Tell the president to ride out the attack.

Sir, they need a decision.

General, do you really believe that the enemy would attack without provocation using so many missiles, bombers, and subs, so that we would have no choice but to totally annihilate them?

One minute and 30 seconds to impact.

General, you are listening to a machine.

Do the world a favor and don't act like one.

What's it doing?

It's learning.

I don't want to ruin it but, they lived.

So there wasn't a total nuclear war.

But there are several key points of that that hold up really well over time that I think are relevant today.

So you heard that the machines can learn, the AI systems can hallucinate.

Hallucinations still happen with AI platforms today.

Those AI computers might have their own power generation.

He turned on his own power.

Independent power.

The risks of machines could be existential.

And it's a national security issue.

Like it's so evergreen.

That movie in many ways holds up well as we grapple with these issues related to AI.

So our popular culture can get it every once in a while.

Another part of this is that the information energy nexus has been around for a while too.

That's not new either.

In fact, this information intensity of the energy system that's been growing for decades with smart meters and smart grids and smart sensors.

There's so much more there's data on everything.

Data in your hand.

Data is being collected for so many things, including the energy sectors.

So that's part of this trend.

So the idea that information needs energy and energy wants information is not new.

It's been growing for decades.

It's just kind of at this point where it's growing exponentially and they're more coupled than ever.

But there's also some counter trends.

Energy is becoming more distributed with time.

We're going from far away, large power plants that are centralized to rooftop, smaller systems like our solar panels.

It used to be all the information was in our laptop, but now might be the cloud somewhere else.

So one's going from local to far away.

One is going from far away to local.

So they have a counter trend in how they operate.

And then all this new wave of information energy leads to a lot of CapEx or capital expenses, a lot of investments that are happening, among the biggest in U.S.

history.

So let's compare it with different waves of investment that have happened over time.

The canals were huge in the 1800s.

Abraham Lincoln ran on a platform of building canals and getting more canals.

Canals are so important to national security and economy and everything else but eclipsed a little bit later by the railroads, which were huge, the railroads were so big.

The interstate highways, therural broadband of the Bush administration, now AI infrastructure, the amount of money here we're talking about $1.2 trillion that is being invested for this wave.

And that's impressive and fun, exciting, but also kind of scary about what kind of things we're building, how long they'll last, and how long will we have to deal with them.

Let's put the same information on a different kind of chart.

Let's put it in, money on the y axis in, money on the y axis and time in years on the x axis.

And what's especially interesting about the data center build out is how steep that is, how much we're spending so quickly.

The railroads were huge, Interstate highways were huge.

But the railroads unfolded over decades.

The space station decades, Apollo program say a decade F-35 program over a couple of decades, Interstate highway over decades.

Most of it unfolded over decades.

And we're at these levels of investment in a couple of years.

So the pace feels kind of stunning.

Let me take that same information and normalize as a percent of GDP it will look different.

Railroads are off the chart.

Railroads were a big part of GDP And the data center system here is early but steep.

Is it going to overtake railroads as a fraction of GDP?

Is it going to level off like Apollo program?

We don't know.

It's among the biggest things that ever happened in United States history, at least when it comes to the infrastructure investments.

So that's interesting.

And you might have a job related to this.

You might help build things or design things.

You might use the things.

It's going to touch all of us one way or another.

For us here in Texas, it's relevant because Texas can lead the way when it comes to AI.

We could lead in AI for all sorts of reasons.

We've got a standalone grid, we're a very large scale energy producer, and we have the infrastructure to back it up, Rapid growth in new forms of energy a deep bench of talent, a lot of money already flowing here, and we've got really good fiber infrastructure, optical fiber.

So we have these ingredients that make Texas really poised for success.

Let's talk about how we're a global scale energy region, Texas.

If we were a country, which I know we like to talk about in Texas a lot let's say we're a country.

We'd be number three in the world for natural gas.

Number one's the United States.

Number two is Russia.

Number three is Texas.

Like we're bigger than Iran or Canada or Qatar.

You name the place.

Texas is a dominant natural gas producer Number four in the world for oil production and refining.

Number six in the world for installed wind capacity.

If you look at actually wind generation in gigawatt hours, I think we're number four.

Number nine in the world for utility scale solar.

I wouldn't be surprised if we're number seven by the end of 2026.

Number three in the world for battery energy storage.

We were basically not even on the chart a few years ago.

We'll probably be number three or possibly number two within a year or two.

We're number six in the world for CO2 emissions.

We're a filthy state.

We're up there with Germany and Canada and Australia.

If we were a country in terms of our emissions, but we also are number one in the world for carbon capture and sequestration.

So we're really good at taking carbon out of smokestacks, or out of the sky, and putting it in the ground.

And we're really big at hydrogen production.

So we're a global scale energy region, producer, consumer, user, mover, everything else, and that makes us poised to handle a rapid buildout for something that needs a lot of energy.

We also have our own grid.

There are three grids in America East, West, and Texas.

And for better or worse, that's what we got.

And that means we can't buy power from our neighbors when we have a freeze like Winter Storm Uri.

It also means we can't sell power to them if we have access and we'd like to make some money doing that.

But it means we have a little more control over the policies and can build more quickly In Texas.

We can build out the power that might be needed, and that's ERCOT, the Texas grid, the Electric Reliability Council of Texas, and that grid has been building.

And it's been getting cleaner.

This chart shows by color the amount of our electricity we get from different forms of energy, different fuels.

Natural gas in blue, coal in orange, wind in dark blue, nuclear in green, solar in that yellow color, and then purple is other, mostly hydroelectric or some other things.

And what I think is facinating here is, how much coal has shrunk over the last 20 years, as wind and solar have come online, and wind looks like a tornado.

That's kind of on purpose by Doctor Josh Rhodes, who made this chart.

A lot of wind came along and now solar, they're displacing coal, so our grid is getting cleaner as we add more things into the grid, which is great.

And we have natural gas to back it up if wind and solar aren't available.

Now we're adding batteries and other things.

So this is a pretty exciting grid.

In fact, we host delegations from around the world who come here to say, how did you do that?

We're like, oh, we've got a special market design.

We got easy regulations.

We got clear rules.

Come build here so we can build the power we think that we need for data centers, and as a result, data centers, they come here.

We have a lot of data centers.

Number one in terms of states is Virginia.

Number one in terms of states is Virginia.

So it's hard to see here on the chart, but Virginia has 643 data centers as of this chart in October 2025.

Texas at 395.

This is growing.

We'll probably overtake Virginia within a year or two.

We'll see how it goes, because we're building so many more data centers here.

And then there's clusters in Dallas, in Atlanta, Northern Virginia, the federal government is a big user of data, Now let's zoom out and look at a national map.

This is a map of infrastructure in the United States.

The colored lines are transmission lines that show the power that's moving.

The white lines are fiber optic capacity.

And then the circles are where the data centers are either already operating, under construction, or planned.

And there are these clusters in Dallas and Amarillo, maybe in Abilene, Atlanta, Northern Virginia.

And that's because that's where there's a right kind or right mix of infrastructure bringing it all together.

So this map tells us about that.

If we zoom in on Texas, we've got a lot of fiber in Texas historically because of Dallas, because of AT&T and Nortel and the different telecom companies based in Dallas, there's a lot of fiber optic capacity there.

So Dallas has been a magnet for data centers and for AI today.

Let's take the same map but overlay in blue natural gas pipelines.

So a lot of natural gas pipelines, a lot of natural gas pipelines in Texas, part of the same place where you have the fiber optic lines.

So we have a lot of places where we can build here.

It's the same kind of story.

We have fiber optics, natural gas, power, land, a smart workforce that all comes together.

Let's build data centers in Texas.

So that's why we have to confront that issue today.

There's a group at the Bureau of Economic Geology, a research entity here at UT that's put together a consortium, a group of researchers called compass, where they're trying to optimize where you might put data centers.

And they have the same kind of concept.

Like, where are the people?

Where's the energy?

Where's the infrastructure?

Where do you have the appropriate land?

Where can you avoid the weather hazards?

If we're too close to the coast, you might have to worry about hurricanes.

And they put this together with a series of layers of maps to help identify the spots.

So researchers are trying to work with developers to identify places that are least impactful or most beneficial to everybody.

West Texas is kind of appealing because there's a lot of wind, a lot of solar, a lot of land, and there's some fiber optics out in West Texas.

So that's attracting a lot of attention.

But that's not really good news for everybody.

Let's look at some of the headlines.

The Stargate data center.

It's poised to complete construction this year in Abilene.

The Stargate complex is expected to be at maximum build out five gigawatts in peak power demand.

Five gigawatts.

And the gigawatt might not make sense to you, So one gigawatt for a nuclear power plant, five gigawatts, five nuclear power plants, a power at just one complex Stargate.

The peak power demand in the city of Austin, we're one of the largest cities in America, our peak power demand is three gigawatts.

So one data center complex, will put one and two thirds Austins at the end of the road in Abilene.

That's a lot of power in one spot, and it's much bigger than all the demand for everything else in Abilene right now.

It's not that many people there.

There's not that much industry.

And there's some excitement for the jobs and economic development.

But a lot of concern about the environmental impact.

So that's Stargate.

Meta is boosting investment at their West Texas data center near El Paso, and we have Amarillo residents voicing concerns over the new data center.

They're worried about the water that might be required for that data center.

So that's been in the news and then some others on the way.

The Texas Tech University system teaming up with Fermi America.

This is a group that former Governor Rick Perry is working with to develop data centers using natural gas, although maybe someday also nuclear up in the Panhandle.

So that's another place where we're getting data centers where there didn't used to be data centers.

And now it looks like we're about to pass Virginia for the most data centers, and people are worried about the water and electricity.

It's the same concern I talk about every day at home.

What are we gonna do with all the water and the energy?

And this just a good reminder that these data centers, though exciting, have a lot of environmental impacts.

Let's go through them.

There's the heat.

They are hot, they generate heat, and the heat could be felt nearby.

They have a lot of truck traffic the trucks can tear up the road or cause accidents or congestion or noise or dust, you name it.

There's light pollution.

For security, they have a lot of lights that can ruin the dark skies.

There's the noise.

You can hear them from a distance.

They're certainly loud within.

I'll come back to that.

There are the fumes from all the generators on site.

The land impacts, of course, just like all the landscaping, all the land work the energy and the water.

So let's talk through some of those.

Let's talk about the urban heat island effect.

These data centers are so hot it will raise the temperature at the site by 2 degrees Celsius That's about 3.5 degrees Fahrenheit, for those who don't speak Celsius.

that's several degrees on site, but it can be as much as 9 degrees Celsius on site, which is like 16 degrees Farenheit.

And by ten kilometers away the temperature's back to normal.

But even within mile or two, you still feel the heat from the urban heat island effect of that data center.

And so in a hot state like Texas, that's not going to be fun.

So that's one environmental impact.

There's also the noise from the uninterruptible power supplies, the generators, all the fans inside the system.

If you've ever toured a data center, they probably gave you ear protection because it's so loud inside of it.

And if you don't do your noise dampening the right way with the walls, you can hear it outside.

So this is a concern for people.

So you have to pay attention to all of the environmental impacts including noise.

And then of course, there's the energy that drives a lot of the conversation.

And it needs energy because in the end, AI is pattern recognition.

It's doing mathematical calculations to recognize a pattern and predict the future.

So if you have to do a lot of math, you'll use microprocessors and have a lot of microprocessors in the data center, doing a lot of math to make your predictions.

So those high speed computations require electricity.

And the AI world has two phases of its life the training and then the inference.

The training is taking all this data out there that you will then train your model with, and that model's here with the patterns and connections.

And then you use that model, you deploy that model, with some inputs or prompts or queries to make a chart or make a logo or whatever it is you're using AI for.

So you have to train the system and then deploy the system.

We call that deployment inference.

You're inferring the future.

It's like when you type in a search bar and it starts to predict your next word, or you're trying to text on your phone.

It's inferring or guessing what your next word might be.

That's a form of AI, that kind of pattern recognition.

So we've dealt with it already just in our day to day life.

So those two phases have different kinds of power requirements.

The training process, which is really in the news these days, is very power intensive, very space intensive.

It's incredible how much you need for all the chips and all the power, but it's flexible in location.

This is the national security concern, which is the training can happen out of the country.

And this is what makes the Pentagon nervous.

Like, well, we want the training to happen here.

We don't want an adversary to do the training there.

How do we build it faster, so the training is done here?

But man, you need a lot of power quickly.

The inference is not as power intensive, doesn't need much space, but it does need to be close.

In fact, it might be at the edge.

It might be in your watch or on your phone, or in your device, or in your car, or in your Waymo or whatever it is.

So the inference can be at the edge, but the training is at a large centralized system and the workloads are switching.

We used to be almost all the workload was training the large language models, the LLMs, and a little bit was inference, but it's changing as we train up the models, we'll go more towards inference, but the power consumption will still be high.

We might take those training data centers and turn them into inference data centers.

So that's part of a life cycle of an AI system.

And those data centers are just converting electricity into information and heat.

The information think of it as photons, you're taking electrons in and photons out with your fiber optics, but also generating heat and this shows you the energy we need the electricity in.

But then we have to manage the heat, which needs more electricity for your cooling.

And then you get photons out, electrons in heat and photons out.

That's the AI data center.

And that's the energy implication and the energy densities or power densities are just remarkable.

So let me give you some data on these racks over time.

So a computer rack is about my size.

It's about my height, about my width, much deeper than I am.

And you have a stack of computers.

And historically, for about the last decade, a computer rack was like 15kW.

A home in Texas is like 3 to 5kW.

So this is like 3 to 5 homes of power in one rack.

So you think like homes in a cul-de-sac or something like that.

But they've been growing.

By last year, by 2025, the server racks were ten times more powerful, say 150kW or so, over 100kW.

In 2026, there'll be 600kW per rack.

By 2027, one megawatt in a footprint the size of me, one megawatt is about the size of like an H-E-B superstore.

An H-E-B superstore in the size of me and had me lined up hundreds of times inside a data center.

That's how you get these multi 100 megawatt data centers.

The power density is incredible.

This blows away any appliance you have in your house by orders of magnitude.

So it's incredible power density.

And that also means a lot of heat has to be managed.

And you get these tensions which is as the chips get smaller, they're more fragile.

They're less robust to heat.

So cooling becomes more critical so you don't break the chip.

There's a lot of research by a lot of semiconductor manufacturing companies on how to manage the heat, how to do thermal management to shed the heat quickly, or how to make the chips more efficient so they don't generate as much heat.

You have to get the heat away from the chips so you don't break the chips.

The older chips are less power intense, but also a little more robust, which is kind of interesting.

And then as the racks go up to megawatt, they're also generating more heat.

So they're more sensitive to heat and they're generating more heat.

You have to work more desperately to get the heat away because there's so much of it, but also because the stakes are so high.

So these trends kind of compound.

And that puts it all together in these data centers, which are growing.

And just as a quick reminder, it's not just for AI.

We have data centers for enterprise, like for banking, we have it for cloud services.

We have it for edge computing.

We have it for high performance computing.

The Texas Advanced Computing Center here at UT has a five megawatt supercomputer.

So we've got a very large supercomputer here at UT for all our calculations, for all of our science.

So there are different types of computers and data centers and AI training is just the latest version of it that's in the news.

So we've been dealing with data centers for a while and thinking about their energy needs.

If we stack it all up, data centers are a lot of electricity, but they're kind of not.

Like they're not really overtaking air conditioning yet as a cause of electricity consumption for all the data centers.

That's this orange bar at the top, but that includes AI.

If you look at just AI data centers in 2024, they're about at, like televisions, and they're going to overtake refrigerators in 2026 and then at some point will grow.

I guess the difference is air conditioning is not growing as quickly as data centers, but the data centers yet don't overwhelm the electricity system.

And if we look to the future for electric cars, which are not that big of a deal today, electric cars just barely overtake hot tubs, which I think is kind of funny.

But electric cars are growing quickly, so that might really accelerate to the right as well.

So there's a lot of electricity consumption for a lot of things.

And data centers are one part of the puzzle.

And AI ata centers are one part of the data center puzzle.

And then if we put it in some sort of national numbers, So it's like 4% or something like that, of our total energy consumption is in the data centers, and that's not that big a deal yet, but locally it might be almost all the new power.

So in Abilene it is going to be a big deal because it's going to overwhelm how much electricity they use in Abilene.

But across Texas, it's not that big of a deal across the nation, but locally it's a big deal.

And that's the tension, which is it might be all the power locally, but a small fraction of the power on average across the region or across the nation.

So if we look at electricity consumption across the years, refrigeration the red line lighting has been dropping as we've added LEDs.

Lighting has been getting more efficient with time, which is great news.

So lighting actually drops.

Space cooling grows as we have hotter climate and more people, more houses.

Ventilation grows.

Data centers are going to overtake.

They're overtaking refrigeration now going to overtake the others within a decade or a few years.

So it's the rapid growth that's a concern, not the level of consumption today.

And it turns out it's a global phenomenon.

This is the total energy consump in terawatt hours for electricity for data centers around the world.

Just for a reference, Texas for all purposes, consumes about 500 terawatt hours of electricity.

So if you take all the data centers in the world, They're going to reach Texas around now-ish, maybe overtake and grow, but all those might be in Texas, so Texas might grow alongside But we're talking about a Texas worth of electricity.

Okay.

Is that a big deal or not?

Texas is a big state, consumes a lot of electricity.

That's a lot.

But we're just one region out of the world.

So the numbers grow.

They grow here in Texas.

They grow around the world.

If we zoom in on the Texas numbers, though, the peak demand here in gigawatts is pretty fascinating.

Here's our peak demand in Texas over time.

We're here in the last few years, around 85.5GW in 2023.

Do you remember the heat dome and how hot that was?

That was our peak demand of Texas 85.5GW.

The peak demand was lower in 2024 and 2025 because of milder temperatures.

The peak demand will grow because of weather and because of people and moving here and all the industrial activity.

But then if you add in the data centers here in gold, look at that growth for non-crypto data centers.

The red is cryptocurrency mining.

Oil and gas is there.

But it's just a few gigawatts.

Doesn't show up that much, our other industrial uses.

This does not include population growth or economic growth.

There'll be some growth just from population growth and economic growth and other industrial activity.

But the data center numbers are just wild.

Those won't all get built.

We don't know what will get built.

Maybe 20%, maybe 30%, it won't be all of them.

But even 20% of a really big number is still a big number.

So that's the curve we're at.

And we're right here at that inflection point.

We'll see how it goes.

The AI training, that has all the chips, everything else we're talking about, has some bottlenecks that will slow its growth.

It has a power scarcity.

That's the biggest bottleneck to That's why it's in the news because we got AI data centers paying a lot of money to overcome the power bottleneck to get power very quickly.

There's chip scarcity, but the chip scarcity is masked by the power scarcity.

Then there's data scarcity.

We're not out of data yet, but we're using the entire internet to train these models.

We're going to run out of internet soon.

Isn't that wild?

It feels like the internet's unlimited, but we're going to run out of internet we're going to run out of data to train these models with so people have to get new data or have synthetic data, make up some data.

So think about your proprietary data, and how you might use that for training.

And eventually you have latency problems.

Where actually communication from one chip to another from one end of the sprawling complex to the other, that latency or lag time in communications will be a problem as well.

There are some bottlenecks, so I don't expect AI to grow quite as fast as some of those numbers, but still, it's going to grow quickly.

Those AI data centers have a few options to meet their energy needs.

They can connect to the grid, which is what they would like to do, but we're worried they will drive up our prices.

They could build their own power behind the meter, but we're worried that that will drive up emissions.

They can revive old power plants.

Here's one, a Pennsylvania deal to get an old hydroelectric power plant revived for a Google data center or you might seen the news around Microsoft reviving Three Mile Island to get some nuclear power for data centers.

You take old power plants and turn them back on, or you can use efficiency.

And there are several different pathways for efficiency.

You can think of algorithmic efficiency, be data wise instead of data heavy.

And that's what the the DeepSeek algorithms from China were really interesting.

They were really efficient.

That caught a lot of people's attention.

Like, holy cow, that was a very efficient algorithm.

You can think of your chip efficiency in terms of Flops per watt.

A flop is a floating point operation or a calculation, just like how many mathematical calculations you can execute per watt of power.

There's a cooling efficiency measured by power usage effectiveness and then communications efficiency.

Instead of using copper, maybe we can use fiber to communicate chip to chip.

Photons move faster than electrons.

So there are a variety of things we can do to improve the efficiency of a system that will lower the power.

And we're probably going to have to, because you probably can't get enough power to drive the whole system.

We got to talk about the water as well.

AI needs water.

Sometimes the data centers are built in water stressed regions.

So there's a map of Texas I stole from a postdoc, Doctor Drew Kassel is here.

We've created this map based on where large loads will be added around Texas, different counties in darker colors where you have more of them.

And we also have a lot of water stress in Texas.

You've probably seen the news about Corpus Christi, where Corpus Christi is like running out of water.

It might run out of water in just a few weeks.

This is crazy.

If you zoom in on that map, we have a problem, right?

That there are data centers that might be built in a place that doesn't have water already or doesn't have the spare capacity.

So this is the stress we have to think about with Texas.

If you look at the water use of a data center, 40% of the water use of a data center is at the data center.

60% is to cool the power plants that supply the data centers.

So there's different water use, cooling the power plant to make the electricity and cooling the chips and cooling the data center directly.

So that's a lot of water.

And if you add it all up, it becomes non-trivial.

Doctor Margaret Cook, a former student of mine, we're very proud of her, at HARC, The Houston Advanced Research Center, did this really great report in January 2026, talking about thirsty data and the Lone Star State, how much water we're using.

And she said if you add up all the water use directly on site, as well as the water used for cooling the power plants to provide electricity to the site, it's about 25 billion gallons, 25 billion gallons of water in 2025.

That's a lot of water.

But it's only 0.4% of total water use in Texas.

Okay, it's a big number, but we use a lot of water for a lot of things.

If you think data centers use a lot of water, wait till I tell you about farms.

That's pretty interesting.

But it's 0.4%.

But it's going to grow, and that demand might increase by several factors to something like over 150 billion gallons by 2030.

And that will be a couple percent of total Texas annual water use.

The challenge that she identified is the Texas State Water Plan does not include this projected demand growth from data centers.

Our plans are not prepared for data centers when it comes to water.

And so a first step might be okay, well, so you think about it and build that into the plan.

And then those numbers, while scary, are still smaller compared to corn irrigation or the amount of water that we leak from pipes across America, which are like a trillion gallons or several trillion gallons.

So it's a lot of water, but it's not.

But it's the same kind of thing as earlier.

It's not a lot of water on average, but it might be a lot of water at that particular location, and that location might be very water stressed.

So that means we need to think about the power sector as well.

And there are some types of power plants that are very thirsty, like nuclear or coal or natural gas systems for carbon capture.

There are some that are just thirsty, like natural gas without carbon capture.

And then there are the water lean power plants wind and solar.

So you want to think about what kind of power sector or fuel mix we're going to use so that we don't have as much water consumption there, and that will help ease up the system a little bit.

And then we can think about the water use on site.

We can think about efficiency, water efficiency, better water pumps, better water pipes, make sure we're not losing any there for leaks.

Also, switching to air cooling.

So you might use water cooling at the chip, but you might use air cooling to cool the water that touches the chip.

And so we could switch from water cooling to air cooling at the data center at the outside facility, and then also offset the water use elsewhere.

Let me give you a couple examples.

There is this man, Scooter Mangold of Yancey Water Supply Medina, who had a big data center coming to town, he talked to them and said, you know what?

You want a lot of water.

I don't have the water.

How about you switch from evaporative cooling to air cooling?

And that lowered their demand from 1 million to 150,000 gallons of water a day.

Okay, 150,000 gallons of water a day is still non-trivial, but it's 85% smaller than the million gallons a day.

So that's good news.

Their total needs will be 55 million gallons a year.

So you could do this.

These are headlines that are happening every day, these negotiations and these conversations.

So we can change the cooling system to use less water.

The trade off is air cooling is not as efficient.

It will require more energy.

So this is one of the subtleties we have to be mindful of.

And if we put that 55 million gallons in context, there are other facilities that use a lot of water.

This is news from a couple days ago.

This is the Gigafactory.

They identified that, oh, they needed 200 million more gallons of water per year than they had the prior year.

Their total use is over 550 million gallons in 2025.

This one factory here in Austin uses more than ten times as much as that new data center in Medina County.

So data centers are water intensive, but they're not the only water intensive thing around, and we've got to put it all in context for which things we think are most valuable for society.

Meta is building a data center in El Paso where they have actually promised to be water positive.

So this is interesting, how can you be water positive?

They're using a lot of water, but they invest locally in irrigation efficiency, or other water projects, that frees up more water than they are actually going to use.

So they will actually generate more water for the system than they will require, and that is our goal to do globally by 2030.

Now that's a goal.

That's a pledge.

That's not quite a law, but still it's a step in the right direction that we have a major company saying we're going to have water positivity, we're going to pay for water efficiency and irrigation efficiency, you name it, elsewhere so that we have more water for the system.

This is a step in the right direction, I think.

This kind of raises the question, is AI part of the problem or part of the solution?

The answer is yes, right.

It's part of both for everything.

This is the way I talked about it in 2024.

I said, look, is AI too power hungry for our own good?

Projections suggest AI platforms may need gigawatts of electricity, but AI could offset that by unlocking new means for saving energy or producing clean power.

This is the way I wrote about it in July 2024.

The New York Times thought that was a good idea, so they wrote the same thing a month later.

But it'skind of the idea, right, is AI going to ruin the planet or save it?

Like, it can do both you know, it can save it.

And let's use it to save it.

So let's use it for good where we can.

There are a variety of ways why AI centers are bad neighbors.

The diesel generators are a huge nuisance.

They're very loud.

They have a lot of fumes.

You have to turn them on every two weeks just to make sure they work.

They have all these trucks delivering diesel, that's annoying.

The trucks tear up the roads.

And also they don't increase the population.

Data centers don't have a lot of employment.

And so if you're a rural county that would like more people to move there, this isn't going to satisfy that desire.

And so that can be a challenge for areas that would like population growth.

But they can be good neighbors because there are some places that they're crowded, like, we don't want more people, but we would like more money, like Northern Virginia.

And so you have Loudon County there with a lot of data centers and the data centers as they get built, here in the blue, the tax rates drop because the data centers pay taxes, but they barely increase traffic.

They don't put pressure on the schools or the social services.

They just get more money.

And that's actually pretty interesting if you are in that community.

John Arnold, some of you might recognize that name, a big Texas figure when it comes to energy, he put this out not that long ago.

“The data center buildout has allowed Loudon County to lower property tax rates by 38% since 2010, translating to average savings of about $3,400 per year for homeowners.” That's a good neighbor, and we just need to make sure that the data centers are paying their taxes, so that we get those benefits.

What will those data centers do to rates?

Well, they will increase them or decrease them.

This is like the constant conundrum with AI.

This is a chart from the Lawrence Berkeley National Laboratory, came out in April 2026 that looked at these rates.

They're saying, well, basically the prices will fall if the demand increases, but at a faster rate than the investments that are required to improve the system.

If you put in new demand but don't spend that much to expand the system, your rates will go down.

If you had to put a lot of money into expanding the system, then your rates might go up.

And we kind of have a choice here as citizenry and rate makers and taxpayers and policymakers, everyone else, to think about this, to make sure that the rates will go down, not up, wherever possible.

So there's actually some scenarios where the rates will go down.

I've got this quote, so I want to bring it up for you “I actually like the idea of these data centers connecting to the grid, because it's good for all of us to have a really rich customer who has the money to invest in making the grid better.” I like having rich customers pay to upgrade the grid so I don't have to.

Like, to me, that sounds pretty appealing.

As long as the rich customers will invest to make the grid better.

So we have to make sure that's part of the deal.

But generally speaking, new customers for electricity are often poor, like we think of like rural electrification 90 years ago, other activities, it's about bringing electricity to poor people, to lift them out of poverty.

This is a very different situation.

It's a group that's already very rich that would like more power to get richer, but they have the money to pay for the system to be better, for all of us.

So let's make sure that's part of the deal.

We could also think about making them more beautiful.

Data centers are not pretty.

I don't know if you've ever seen one, but they are ugly.

And also these public storage buildings that you see them all over Austin, all over Texas, they're also ugly.

And people are like, we should make them prettier.

So here is a public storage facility, where they pretend it's townhouses.

And they're like, let's make it look like town, this is The Wall Street Journal, April 15th.

This is a recurring story, like, why are we building ugly things?

And we could do something with data centers that are pretty.

What if we made a bunch of, like, Gothic or Art Deco data centers and it becomes like a tourist destination?

We have ren faire out there or something like that.

Be a lot of fun.

So we could consider making them beautiful and that would help as well.

we can use AI to accelerate solutions for energy and water and transportation.

There are all sorts of energy use cases for AI.

We can use AI to accelerate geological exploration for heat.

That's the Zanskar startup.

They're using AI to review the old geological records, to find geothermal resources, to bring clean power to market.

Kobold's doing the same thing, to find cobalt, using maps to find places to get cobalt where you can get it without all the labor justice and environmental justice concerns of where we're doing it right now.

What about accelerated fusion reactor design, or accelerated discovery of advanced materials, or ship navigation, or mining, or optimizing our oil and gas networks, or extending the life of critical assets like transformers on the grid?

There are all these ways we can use AI to improve the energy system, and that's really exciting.

And we can give a couple examples here.

One is autonomous robot solar installers that will speed up deployment of solar panels at lower cost.

So let me show this clip to you of a robot installer.

This is autonomously taking the panels and putting it on the rack.

And so now a crew of a few people can install a lot more solar panels, much more safely or quickly, and more cheaply.

And this will accelerate the arrival of clean energy, because we will lower the cost of solar, which is already cheap.

So this is one fun application.

Let me show you another video of an autonomous weedkiller that will improve crop yields and reduce use of agrochemicals like pesticides, because you can use the autonomous machine to zap the weeds with lasers.

And I love lasers, because I'm a lasers guy.

It can differentiate the weed from the thing you want to keep and zaps it with the laser so you don't have human labor, go pull it.

And you don't have to apply chemicals that are dangerous for the soil or for human health.

So these are fun applications that are really interesting for us.

So when we have AI combined with autonomy and combined with robotics, we can do a lot of the jobs we would not have been able to do otherwise.

Autonomous cars are already saving lives.

I don't know how many of you already ridden like in a Waymo or an autonomous car, so it's kind of fun, is it?

You don't have to talk to anybody.

It's clean, all that kind of stuff.

They're also super safe.

if you look at the statistics, it's incredible.

After 170 million miles, Waymo is now going past 200 million miles, the safety improvements are dramatic.

92% reduction in serious injury or worse crashes, 83% reduction in airbag deployment, 82% reduction in injury causing crashes.

Great numbers here on 92% reduction in pedestrian crashes.

So that's huge in terms of human safety for people walking, and same kind of numbers as cyclists and also motorcycles.

They're so safe.

If we were testing a pharmaceutical and you have the placebo in the pharmaceutical treatment, and the pharmaceutical treatment was this much better than the placebo, it would be unethical not to give the pharmaceutical treatment to the placebo people.

And we're there with autonomous cars.

They are so much safer.

At some point, it's unethical to let people drive cars, right, because we're just so distracted.

We're so dangerous.

Or maybe that's just for rich people can pay extra to drive because they can afford the insurance.

Insurance will be cheaper for autonomous cars because they're safer.

They don't need bathroom breaks, they don't need cigaret breaks, and they're not distracted by their phones and they have vision, but they also have radar and lidar and acoustics.

They have all sorts of things.

So this is a great application of AI that is already yielding benefits for us, and I'm really excited about that.

We could also design our data centers to be carbon negative and water positive.

I already mentioned the water positive, but at the data center that has all that waste heat, you could use that to treat water.

You can also use that waste heat to capture carbon.

So we can make these data centers a place where good things happen for the environment.

And I would just say, if AI is here to stay and we're going to build the infrastructure, let's do it the right way.

So to kind of wrap up these points, we can build a data center that's beautiful and clean and quiet and dark and energy efficient and water positive and carbon negative.

That's where we need to go.

And I think that's where we need to be.

And then I'll close by saying, in the end, I'm a techno optimist, so I feel like we can solve it.

I can manage it, or we can solve this, right?

This is a problem.

We can identify and solve it.

I'm not an ethicist and maybe that's why I'm more worried about the ethical problems of AI than the environmental problems.

So I want to make one last set of comments for you about what this means for humanity in society and higher education.

So if we look at AI, it does a lot of things well.

But this is Michael Webber's list of 6 things AI doesn't do well.

I'm not an AI specialist, so I'm sure an AI person might disagree with this, but this is kind of my view.

And my view is AI is bad at spirituality.

It's not spiritual.

It's bad at emotional connectivity.

It doesn't have emotions.

It can fake empathy, but it's not actually, which is what I do, don't tell my wife she's here.

But anyway, so, it can fake it, but it doesn't actually have emotions.

It's not good at nature connectivity.

It's not good at discerning fact from fiction.

It's not good at creativity.

Creativity is inherently divergent from the mean and AI is inherently convergence to the mean.

So it's generative but not creative.

You can generate with a AI, can generate a logo, generate a song, generate a piece of art, but it's not creating it.

It's different.

That's a human endeavor.

And it's not great a critical thinking.

These are things, in my view, that AI is bad at.

And it might become very good at some of those things someday.

Maybe the list gets shorter, but this is my list as of today, and when I put that in context of being a professor here at the University of Texas, I think of the ways we teach to think.

There are three ways of thinking that we teach.

We teach analytical thinking, that's for the engineers, and we seek to answer the question how?

And then you have the critical thinkers where we seek to answer the question why?

And then we have the creative thinkers, the innovative thinkers, who seek answers to the question, what if?

And the analytical thinking is the easiest to replace with AI, which is pretty interesting because we're obsessed with STEM.

And I'm a STEM guy and I love STEM.

I think STEM people and STEM is important to the world, but we've ignored some of these other things, and I think we're going to need those more than ever.

I think we need fine arts, liberal arts, more than ever, and we need innovators and philosophers.

So that's the part I think AI will struggle with.

And that's you.

And I'll stop there.

Thank you very much.

Funding was provided by the State Energy Conservation Office,