In February, 2021, the state of Texas experienced one of the most brutal winters in its history. Snow storms overwhelmed power plants and left Texans in the cold and dark. Dozens died, some from exposure to the cold, with no electricity to heat their homes.
And yet one business in Texas was getting electricity – Bitdeer – a bitcoin mining data center near Austin, that was “using enough electricity to power about 6,500 homes as they raced to earn Bitcoin,” according to this New York Times report. To add insult to injury, such was the agreement that the power grid operator had to pay Bitdeer $175,000 per hour just to keep their computers off during the crisis.
Digital mining for Bitcoin can be highly lucrative. But crypto currency mining is extremely energy intensive. The total 34 bitcoin mining operations in the US consume the equivalent of the electricity consumed by three million households, according to the Times.
Energy is being used at historically worrisome rates, not just by Bitcoin miners, but also by AI data centers and EV automobiles. Energy usage is so great that those at the heart of the AI and EV industries, like Elon Musk, are sounding the alarm. In this interview with Lex Fridman, Musk talked about how the intense demand for chips that run AI algorithms has resulted in a silicon shortage today, which will lead to a shortage in step-down transformers, and eventually, electricity shortages in general.
I gave a speech for the world gathering of utility companies, electricity companies, and I said, look, you really need to prepare for tripling of electricity demand, because all transport is going to go electric with the ironic exception of rockets, and heating will also go electric. So energy usage right now is in very rough terms, one third electricity, one third transport, one third heating. And so in order for everything to go sustainable, to go electric, you need to triple electricity output.
As former NBC journalist, Jacob Ward put it recently when asked what tech leaders say is the most overlooked story right now, “they all say – the electrical grid, and the fact that we’re going to run out of power, run out of electricity, in the coming years.”
According to engineering professors at the University of Pennsylvania, electricity consumption by computers doubled or tripled from 1-2% of global supply in 2018 to 4-6% only two years later in 2020. “If we continue at this rate, by 2030, it’s projected to rise between 8-21%, further exacerbating the current energy crisis,” said Assistant Professor Deep Jariwala in Penn Today.
According to the International Energy Agency (IEA), electric vehicles (EVs) will require a huge increase in electricity, between 525 and 860 terawatt-hours (TWh) globally in 2030, up from 80 TWh in 2020.
Add in the fact that mining for crypto uses up to 2.3% of all electricity in the US….all that is a lot of electricity being consumed by technology that has gone mainstream in just the past few years.
An AI Picture is Worth a Thousand Watt: Let’s take a closer look at Artificial Intelligence. According to this MIT Technology review, creating an image using Dall-E on ChatGPT for example, will require the same amount of energy needed to fully charge your iPhone. If you spent a few days working on a project and generated, say, a thousand images using a generative AI engine, you’d have released “as much carbon dioxide as driving the equivalent of 4.1 miles in an average gasoline-powered car.”
This blows my mind.
After all, I’ve generated hundreds of pictures using AI tools, some of them really dumb and unnecessary. I have asked the tool to make incremental changes to images I didn’t really need or care about, each time unintentionally contributing to this massive surge in energy usage. So if I multiply me times the billion-plus people who are doing the same thing…..it makes me wonder: Who’s paying for all this?
As Nate B. Jones states in this TikTok post, the energy bill is paid by the tech companies and VCs. And while access to this AI resource is relatively inexpensive (free for most people), he hints that it may not stay that way forever.
It’s actually not free to use ChatGPT. It’s subsidized by venture capital, just like Uber used to be subsidized for like three-dollar taxi rides back in the day. And now it’s extremely expensive. It’s like $30 or 35 or whatever, and the market has come back to where it is because it’s no longer subsidized. So ChatGPT is subsidized right now, and people have gotten used to this idea that there is intelligence and compute that’s cheap, and it’s true.
For now.
But the AI ball has only started rolling up the S Curve. As it gathers momentum, the capital required to pay for the ginormous amount of chips and energy needed to feed the growing AI ecosystem will be humongous. Open AI founder, Sam Altman has been oft-quoted saying he is trying to raise USD 7 trillion dollars to manufacture the chips OpenAI needs to power AI in the coming years. Altman actually denies saying “7 trillion,” but he does not deny the task ahead to amass the resources required. As he explains in this recent interview with Fridman:
I think compute will be the currency of the future. I think it will be maybe the most precious commodity in the world. And I think we should be investing heavily to make a lot more compute. Compute is – I think it’s going to be an unusual market – intelligence is going to be more like energy, where the only thing that I think makes sense to talk about is, at price X, the world will use this much compute, and at price Y, the world will use this much compute. If it’s really cheap, I’ll have it reading my email all day, giving me suggestions about what I maybe should think about or work on, and trying to cure cancer. And if it’s really expensive, maybe I’ll only use it, or we’ll only use it, to try to cure cancer.
The economics of compute, and the ability to pay for this phenomenal demand for energy, has many implications. Here are two that I find particularly interesting:
Human Jobs Won’t All Disappear: Assuming demand for electricity skyrockets for an extended period of time into the future, AI will continue to be expensive, and cost effectiveness may actually turn out to be better for humans than AI. Here is Jones again, explaining how that might be.
We’ve never had cheaper intelligence from a machine perspective. But people are right there and people can do the work and people have extended planning that, right now, ChatGPT doesn’t. And even if they invent extended planning to go with ChatGPT, which I know they’re working on, that still won’t mean it’s cheap enough to do all of our jobs for us. So in a sense, there’s a very economic reason why people should feel confident that they will still have jobs even if there is a super powerful intelligence somewhere. That intelligence will be used for very specific applications where it makes sense for it to be used. Probably the applications that require the most intelligence to get the job done, where there’s differentiated benefit for that degree of hyperintelligence. And for most of the other things it will be people, just like it’s always been.
Nuclear Power Makes a Come Back: Another major implication of the skyrocketing demand for electricity will be a resurgence in nuclear power. In fact, according to the International Energy Association (IEA), nuclear capacity in 2022 grew 40% globally. Many believe nuclear power is needed to counteract the increased demand for fossil fuels, particularly coal, to meet the surging demand. That is why people like Altman are not only investing in the chips that are needed for AI compute, they are also investing in energy production – particularly nuclear. According to this article,
Altman… previously invested in US nuclear fusion startup Helion Energy, providing $375 million to the company in 2021. Microsoft, the biggest financial backer of Altman’s company OpenAI, which is currently providing massive cloud resources for its ChatGPT signed a power purchase agreement last year for 50MW of electricity from Helion, to be delivered by 2028.
Which brings me to a final word of caution, thanks to the insightful journalist Ward.
Is this the climate friendly solution we need, or is this us turning back to a technology that American more or less abandoned 40 years ago. Nuclear energy has stalled in the US. Do we need to re-start it? Does this serve humanity purposes or just the purposes of someone like Sam Altman who needs power, both literally and figuratively, in order to make his companies possible?
Great questions to ponder, while snacking on an energy bar, typing prompts into ChatGPT, in the backseat of a Tesla.
ARTICLE FAQS
1. Why is global electricity demand rising so quickly?
Electricity use is being driven up by three fast-growing sectors: artificial intelligence data centers, electric vehicles, and cryptocurrency mining. Each of these requires enormous amounts of energy, and together they are straining power grids that were not designed for such rapid growth.
2. How much energy do technologies like AI and Bitcoin really consume?
Bitcoin mining in the U.S. alone consumes as much electricity as three million households. Generating a thousand AI images with a model like DALL-E releases as much CO₂ as driving four miles in a gasoline car. EVs are expected to require up to 860 terawatt-hours globally by 2030, a tenfold increase from 2020.
3. Will AI and automation eliminate most human jobs?
Not necessarily. AI is expensive to run at scale, and its economic value is maximized in high-intelligence, high-leverage applications. For most work, humans remain more cost-effective. Analysts suggest this economic reality means people will continue to do much of the day-to-day work, even as AI advances.
4. Why are experts calling for a return to nuclear power?
As electricity demand accelerates, many worry that fossil fuels like coal will be used to fill the gap. Nuclear power, which produces large amounts of energy with relatively low emissions, is being reconsidered as a necessary option. Global nuclear capacity grew 40% in 2022, and major investors in AI, like Sam Altman, are also backing nuclear startups.
5. Who pays for the growing cost of AI and compute power?
Today, much of the cost is subsidized by venture capital and large tech firms, keeping services cheap or free for users. But this is unlikely to last. As energy and chip costs mount, the true price of compute will rise, reshaping how widely and frequently people can afford to use advanced AI.
6. What are the bigger risks and questions society faces in this shift?
The surge in electricity demand forces hard choices about energy sources, climate goals, and infrastructure investment. Society must weigh whether expanding nuclear power is a sustainable solution, or whether it mainly serves the needs of powerful companies driving the AI and crypto boom.
Amazon also just bought a nuclear-powered data center!
As we think of the value that AI will bring to organization we are not think of the carbon foot print. We are not thinking of the fling term price we have to pay for short term benefits. Should this be part of the corporation sustainability agenda. For every prompt we type we plant a tree?
Microsoft just made a deal that will lead to the re-opening of a reactor at infamous Three Mile Island in Pennsylvania (Film reference: China Syndrome). They will be the exclusive recipience of this reactor’s energy for 20 years.
https://www.technologyreview.com/2024/09/26/1104516/three-mile-island-microsoft/