AI, the Newest Technology, Is No Help With Climate Change
Apr 26 2024There are those who say artificial intelligence will lift humankind to unimagined heights. "It's more profound than fire or electricity”, Sundar Pichai, CEO of Alphabet Inc. (parent of Google) actually said. Others fret that it will possibly run amok and do us in.
Just how is not explained, but possibly it will get rid of us by quickening climate change because we are learning that AI is a
ravenous consumer of energy. Rene Haas, chief executive of the U.K. chip design company Arm, thinks that by the end of the decade AI data centers will need as much as 20% to 25% of U.S. power needs.
Estimates vary widely. Some suggest hysteria is setting in. Maryland-based Constellation Energy, which has contracted to supply Microsoft with nuclear energy for its data centers, projects that AI's insatiable need for power could exceed by five or six times the amount needed to charge America's electric vehicles.
Driving demand is that the bigger the data sets that "train" AI, the better the results, but ever bigger data sets consume ever increasing power. The International Energy Agency said in a January report that a request to ChatGPT uses ten times the amount of a Google search, and that the power demanded by AI will increase ten times by just 2026. Digiconomist Is a company that tracks world digital use and exposes its unintended consequences. Founder Alex de Vries estimates that, thanks to AI alone, the amount of electricity required to power the world’s data centers could jump 50% by 2027. A peer-reviewed study by his company calculates that AI servers could use between 85 to 134 terawatt hours annually, which is about what his own country the Netherlands uses in an entire year, and which amounts to the entire output of about 15 average-size nuclear power plants.
not just aiAI data farms are a subset of data and computational centers (for search such as Google, Bitcoin mining, etc.) that are being created at the rate of a new one every three days, Amazon Web Services vice-president for engineering Bill Vass told The Wall Street Journal. Electricity demand to power data centers is forecast to rise by 13% to 15% compounded annually through 2030.
In the face of worldwide efforts to mitigate the emissions causing climate change, whatever gets built to supply that power will far outpace the rate that new renewable energy can be brought online. That means increased use of fossil fuels. In that same article, former U.S. Energy Secretary Ernest Moniz said, “We’re not going to build 100 gigawatts of new renewables in a few years. [We're] kind of stuck."
Before the AI explosion, attention to waste energy use was focused on crypto-currencies. Millions of servers around the world compute numbers non-stop day and night in a quest that one of the numbers from one of the servers at one of the crypto farms will be accepted by the Bitcoin algorithm a game of 'guess the number' as de Vries puts it. As he says, those millions of Bitcoin mining devices are…
"generat[ing] 500 quintillion random numbers … every single second of the day. Only once every 10 minutes [does] one participant guesses correctly…"
to be rewarded with new Bitcoins. The payout can be huge, but it is a game designed to waste huge amounts of electricity calculating the quintillions of numbers that are of no use whatever.
A few years back de Vries's firm "conservatively" estimated that the Bitcoin network was consuming 87.1 terawatt-hours (TWh) of electrical energy annually, equaling the energy usage of a country the size of Belgium. With the price of a single Bitcoin having risen inexplicably to over $63,000, uncounted numbers of hopefuls have undoubtedly piled into mining since, adding that many more TWhs of energy waste and CO2 emissions.
not just fossil fuelsThrough overconsumption by humans, the world's supply of fresh water is rapidly diminishing, a threat that, absent drastic action, will lead to mass migrations and wars. So what are we doing? Building those data centers every three days that use vast quantities of water.
When Bitcoin transactions are laid against water consumption needed to cool the servers, the average Bitcoin transaction equates to some 16,000 liters. U.S. consumption could be on par with the average annual water consumption of approximately 300,000 U.S. households, equivalent to that of a city like Washington, DC.
A Google data center consumes about 4.3 billion gallons of water annually, about 450,000 gallons of water every day.
A $1.1 billion data center that began construction last July by Meta Platforms Inc (Facebook, et al.) is expected to use 665 million liters of water a year to cool the atmosphere where stacks of servers run and never turn off – as much as 195 liters a second.
everything electricGoing beyond AI, Bitcoin, and data centers, the problem worsens. A principal mission of President Biden's administration has been an assault on climate change, with almost $400 billion in the Inflation Reduction Act (IRA) allocated to clean energy, upgrading the power grid, incentives for consumers to switch to electric cars, and more.
Yet national security has dictated that the U.S. inshore the production of semiconductors. The administration's CHIPS Act provides $50 billion for subsidies and loans so far to Micron, Qualcomm, GlobalFoundries, Intel and Taiwan's TSMC.
But chip factories called "fabs", for fabricators are voracious consumers of electricity. They can draw 100 times more power than a typical industrial business. Micron, for example, is building a new fab in upstate New York that is expected to require as much power by the 2040s as the states of New Hampshire and Vermont combined.
And, of course, there is Biden's program to do away with the internal combustion engine and cause Americans partly by subsidy and partly by regulation to convert to electric. To the degree successful, this will add an enormous burden on the power grid.
There will be 26 million EVs on America's roads by 2030 according to an estimate by the Edison Electric Institute. That would require an extra 100 TWhs of electricity to handle the load, which is about 2.5% more than today's U.S grid produces. The Environmental Protection Agency (EPA) thinks its vehicle rule will result in EVs being 67% of automakers ' output by 2032 but it will be late in that decade before two-thirds of America's 400 million vehicles will be electric. For that to happen, there must be something on the order of 1,000 TWhs more of electricity, which is 25% more than what the U.S. grid produces today.
Those estimates don't mention whether trucks are included in "vehicles" trucks for which Biden's EPA has just issued a rule that expects 60% of new urban delivery trucks and 25% of long-haul semis (18 wheelers) to be electric by 2032. That will add millions of "zero emission" trucks at charging stations where a semi will soak up seven-or-so times the amount of electricity that a typical house uses in a day.
That is, if the charging stations get built. Some 1.4 million charge units will have to be installed in those stations by 2032 to accommodate the EPA mandate, about 15,000 a month. A report from S&PGlobal in January 2023 concluded that the nation will need more than 2 million would be needed by 2030. Neither is happening. Those wide differences show the difficulty in estimating: how often must the average vehicle pull into a charging station, how long on average does charging take for different vehicles in different climes, how many homes will install their own units, etc. And one neglected subject is transformers. There are as many as 80 million distribution transformers that convert bring electricity into neighborhoods. Designed for today’s loads, millions will need to be replaced and – no surprise – most are made in China.
overloadAll these demands for electricity raise the question of whether it is possible.
In March, The Guardian reported that according to filings made to the Federal Energy Regulatory Commission …
"The nation will need the equivalent of about 34 new nuclear plants, or 38 gigawatts, over the next five years to power data centers and manufacturing and electrify buildings and vehicles."
If solar, a very different estimate says that 500 gigawatts will be needed by 2035, which would require “on the order of 1,000 utility-scale solar plants” adding up to the size of New Jersey, says an analysis in National Review magazine. To find the space and the sun most of the bigger projects would have to be built on Bureau of Land Management terrain in the western states. The process should begin now, because the red tape of the permit-review process in the U.S. is interminable.
Even if that were possible, how would we get the electricity to where it is wanted?
New transmission routes would have to be built to bring electricity from solar and wind areas that have none. Much of the existing six hundred thousand miles of high-transmission lines in this country is in need of upgrade, having been built just after World War II. Biden’s IRA has funding for transmission but nowhere near enough. We covered the problems of the grid a year ago in “Biden Wants an All-Electric America. The Grid Has a Problem”, a bracing report that you can read here.
In all, it looks like Mr. Biden is in store for disappointment.
Please subscribe if you haven't, or post a comment below about this article, or
click here to go to our front page.
The electrification of transportation has already hit a wall, the cost of EVs, the lack of charging stations and the inconvenience. AI creates other problems in terms of replacing human labor, or replacing humans, and the threat to our system of government of by creating unreliable sources of information or false identities, just as social media has already had a negative effect on our society and culture. The demand on energy from existing data centers, bitcoin and now AI will make any reduction in emissions impossible. Why haven’t the political leaders and scientists emphasized this fundamental conflict and given the failure of legislators to regulate social media or technology giants, what hope is there that they can get in front of AI before it is too late? The horse is out of the barn.