This piece was originally published on Stand Together.

Technology and energy have always been inseparable. Every breakthrough, from the first sailing ships to modern AI, has required energy to power its development and widespread use. In 2025, as computing and AI take center stage, the demand for energy is surging to unprecedented levels. But is this growth sustainable?  

In an enlightening interview, Mark Mills, an author and executive director of the National Center for Energy Analytics, shed light on the fundamental relationship between innovation and power consumption. Mills, who served in President Ronald Reagan’s White House Science Office and is a distinguished senior fellow at the Texas Public Policy Foundation, dove into the often-overlooked energy needs of the digital age, comparing the rise of computing to past technological revolutions and highlighting the challenges of keeping up with the insatiable demand for data.  

As we look ahead, what sources will fuel this explosion of digital progress? Will the energy sector be able to keep up? And what happens if it doesn’t?

Stand Together Trust: You’ve written numerous books and articles on the intersection of technology and energy. How is technology impacting the energy sector right now in 2025? And how does this compare to your previous prediction?

Mark Mills: Except for eating food for bare survival, all energy demands are invented. We invent products, machines, and services that do things that we care about that make life more comfortable, more pleasant, easier, more fun, and safer.  

All inventions and all products require energy. That relationship is not just for “tech” as we define it today. Whether it’s a car or the first cart dragged by a horse that required energy in the form of grain, all technologies that are invented to advance humanity, even those we may trivialize — like games, movies, films, and other entertainments — all use energy. 

What happens over time, however, is that we get lulled into a sense of complacency because many energy uses are hidden. Many people seem to believe that many modern technologies somehow, magically, don’t require very much, if any, energy.

Arguably, the most interesting new energy-using invention is the computer. There was no energy demand for computers before the invention of the computer. And then it would take a while before truly significant gains in efficiency would lead to significant energy demand for computers. They did become incredibly more efficient and continue to do so. Now, computing collectively, the cloud in all of its manifestations, uses roughly as much energy as global aviation or, in electricity terms, twice as much electricity as the country of Japan.

So, is that growth in demand going to end? And the question answers itself — no. 

Look at the invention of artificial intelligence, which is actually some 25 years old. Now, it has gotten cheap enough and efficient enough that we have finally seen the eruption of demand and now an awareness of this new tool on the public stage in the form of inventions like ChatGPT. And as it’s become far more efficient, and cheaper and easier to use, we’ll find all kinds of uses for that new tool. AI has already fully invaded the entertainment industry, not least in forging fake embarrassing videos of presidents and prime ministers, but more importantly, now AI is beginning to be put to work doing useful things like drug discovery.   

Its demands are limitless.  

Every other invention has a limit to its demand. There are only so many hours you want to spend in a car or on an airplane, there is only so big of a house most people want to live in, and there is only so much food you can eat in a day. The natural limits to all products and services can even be predicted by the combination of physical limits (you can only eat so much) and the overall rise of wealth.  

For example, if everybody in the world was as wealthy as an American, then we’d know rather accurately the maximum number of cars needed in the world. In America, the number of cars equals the number of people in the United States, roughly speaking, including nondriving people, children, infants, and people who are in hospitals. So, we have about as many vehicles as people, but that points to a limit. There may be a lot of three-car households, but there aren’t many 100-car households. 

That’s not true for information. There’s infinite demand for data. So, trying to guess where the digital infrastructure is going to go in energy terms is really hard.

So, what sources can supply the energy needed to power this new technology? And what are their pluses and minuses? 

The sources for all energy technologies are essentially the same. Since the underlying energy sources of nature are functionally unlimited, the challenge is in our ability to use technology to trick nature into supplying us that energy inexpensively and reliably. This has been true for all human history.

Hydrocarbons (also known as fossil fuels), oil, gas, or coal provide 82% of the world’s energy today. They’ve been providing over 80% of the world’s energy for a century in different combinations.

There’s no likelihood of change in the rough share for the next couple of decades because there hasn’t been any in the last two. That’s because the system is so big, the inertia is so high, and because we need so much energy for the world.

So, the answer to the question of how you power tech is: If the tech is computing, it’s mostly with electricity. However, that doesn’t include the energy needed to build the extremely energy-intensive information hardware in the first place — the silicon chips, fibers, and steel and concrete used to house everything. Making a ton of silicon chips takes 1,000 times more energy than making a ton of steel. And we produce semiconductor chips by the kilotons a year.

So, you fabricate all that information infrastructure by burning natural gas, and you use coal to make the steel that’s involved in these machines. In addition to coal and natural gas, we use oil to transport all the stuff. So, all the hydrocarbons are used to make computing equipment and the associated infrastructures, and then they need a surprising amount of electricity to operate. Every $10 billion of data centers — and the tech industry is talking about spending hundreds of billions — will use about $10 – $20 billion of electricity over a decade of the useful operating life of the chips. For calibration, $10 billion worth of electric cars would use one-tenth as much electricity.

The real debate today is over how much electricity the cloud will eventually need for all the data centers and the networks that connect them. How will we produce the electricity, and how will we supply it reliably?

It’s easy to answer. It’ll follow the Pareto Law’s 80/20 rule: 80% of the net new electricity will be natural gas; 20% will be a combination of the rest, depending on where it’s built. Roughly speaking, at least in the United States, that’s where it’s going to come from — because that’s where it can come from, and because you can build half a gigawatt, 500 megawatts, of natural-gas-fired electricity supply for one data center in the same 18 months it takes to build the data center itself.

We can’t build a nuclear plant in 18 months or a coal plant or a reliable gigawatt of solar panels. It’s not possible. But you can, in the best case, in 18 months, build a natural gas plant that can provide that power. To be clear, we’re talking about what is possible in the world of engineering. Some policymakers may want to believe more money, subsidies, can change all that, but it can’t. Inversely, though, regulators can make it take longer to build those things that we need.

What if we don’t build large amounts of reliable new energy required? What’s the impact that it’ll have, and what would it take to avoid this?

The probability that we won’t build enough power is near zero — not impossible, but near zero because the value of the things being built is too high.

The government does not subsidize data centers because they think it’s a social good to build them. This is not the Interstate Highway System being paid for by a highway tax. This is the market demanding the cloud services that the data centers provide, from e-commerce and entertainment, from mapping services to finance, and from health care to scientific discovery — all the things that people do in the business of everyday life. The demand is so high that the supply of data centers and communications is still underserving the opportunity.

Since that’s driving it — an economic value that’s so high — then the incentive to build the power plants is extraordinarily high. And the builders, the big tech companies, have incredibly deep pockets. So, they’ll get built.

What will get built will be driven by what you can build, not what you’d like to build. 
Again, the velocity of the demand is driving the need to build things that are buildable quickly.

And this brings you back into the real world of engineering and fuels — the kinds of power systems that you can get access to in the real world, not one of PowerPoints. And it comes back to the obvious: People don’t think that your access to shopping or telehealth should be dictated by when it’s sunny or windy. Obviously, you want power that keeps the network running all the time.

So, the economic question, as well as the engineering question, is: How do you do that?

Broadly speaking, it’s the same debate we’ve been having for some time about energy and electricity. What AI, computing, and the cloud have done, however, is sharpen that debate. It’s making clear just how silly it is to count on episodic power to provide society’s enormous demand for highly reliable services. And it’s caused a hyperfocus on the engineering problems and staggering costs of that reliability challenge using episodic power — i.e., solar and wind.

What it’s caused people to realize is that even if wind electricity is cheap when it’s windy, or solar electricity when it’s sunny, that’s not the interesting question.  
Is it cheap, then? Yes, it is. But how much does it cost to deliver that electricity when it’s not sunny?

We know the answer to that question since the costs of the associated engineering are clear. Also, Germany and England have given us the answer — both nations have radically expanded the use of wind and solar and found that it’s really expensive to deliver reliability from episodic power; grid electric rates have roughly tripled.

Electricity reliably from episodic sources is not impossible. It’s just expensive. However, the need for a lot more data centers has hyper focused us on getting sensible answers to the question of supplying enough reliable power quickly, at any cost, and of course, cost will always matter in the end.

It’s important to remember that all technologies require energy to produce and operate. And the essential purpose of nearly all technologies is to improve productivity.

Productivity is the driver of wealth. What happens when you create more wealth? Well, people do more things. So, technology has a two-edged sword, which is not trivial. This is true for both computers and cars. If the product or service becomes cheaper, whether it’s information systems or transportation systems, that is also the same as increasing wealth.

Increasing wealth means those who were poorer before now do more things like take a vacation, own a car, or get access to health care, all of which boosts energy demand.   

There is a symbiotic and virtuous feedback loop between technology and productivity, which itself requires cheap and reliable energy. This has been the engine of civilization since the discovery of internal combustion.

National Center for Energy Analytics is supported by Stand Together Trust, which provides funding and strategic capabilities to innovators, scholars, and social entrepreneurs to develop new and better ways to tackle America’s biggest problems.  

The views and opinions expressed are those of the author’s and do not necessarily reflect the official policy or position of C3.