AI Demands to Be Fed. We’re All Servers Now

The colossal energy demands of artificial intelligence have earth-shaking implications for everyone. Already rising steeply, they are set to accelerate at a dizzying pace as various global powers race to be the first to achieve supreme intelligence over everything.

The mantra of the moment is “Compute, baby, compute.” Power to the machine.

Case in point is the petrostate of Alberta. More than 70 per cent of the province’s electricity now comes from natural gas-powered turbines, and Premier Danielle Smith wants to share the bounty with Big Tech.

In response, tech companies have already proposed 29 data centre projects. Together these energy-intensive behemoths would consume 16 gigawatts of energy. That’s more than 10 times the current load for Edmonton. In fact, it exceeds all of Alberta’s current peak electrical consumption of 12 gigawatts. In filling the demand for AI data centres, Alberta would basically need to create a second, parallel electrical system.

Apparently sobered by that prospect, the Alberta Electric System Operator has put a cap of 1.2 gigawatts on new projects till 2028 to protect the grid.

There is good reason to fasten that cap tightly. Texas tells an instructive story about the consequences of unfettered high-tech energy cannibalism.

The state, like Alberta a gung-ho methane extractor, hosts 350 data centres, many the size of football fields. Some of these complexes use more power than mid-sized cities or entire municipalities. Balancing the Texas grid with such explosive growth now threatens the system’s reliability. Already the Lone Star State is one of the top blackout states.

To add insult to injury, homeowners will be asked to pay rising electrical bills for infrastructure upgrades. Texans will find feeding the tech beast a massive chore, according to the Electric Reliability Council of Texas. The state will need to add the electricity equivalent to the output of 30 nuclear reactors to meet soaring demand by 2030 from the world’s greediest energy eaters: data centres, bitcoin mining and hydrogen production. As a consequence, tapping independent on-site power generation with no ties to the grid is all the rage among tech companies in Texas.

AI’s fast-growing hunger for electricity

The trend lines are staggering. We’re just in the early days of the AI boom, and already data centres consumed 4.4 per cent of the United States’ electrical supply in 2023. A recent report by the California-based Lawrence Berkeley National Laboratory estimates they will command 6.7 to 12 per cent of total U.S. electricity by 2028. In Ireland, where data centres are clustered around the outskirts of Dublin, the industry is already gobbling 21 per cent of that country’s electrical supply.

This as the world is seized by a surge in electricity consumption due to several factors. Hot weather caused by climate disruption has played a role by boosting the need for air conditioning and so, too, have electric cars. But the growth of data centres and artificial intelligence is the main factor for the sharp increase in electricity use.

The International Energy Agency recently calculated that powering data centres “is on course to account for almost half of the growth in electricity demand between now and 2030.” In fact, “the U.S. economy is set to consume more electricity in 2030 for processing data than for manufacturing all energy-intensive goods combined, including aluminium, steel, cement and chemicals.”

Are leaders in Alberta paying attention to such projections? How about in B.C., where Bell AI Fabric last month announced plans to build six new AI data centres, while eyeing more in Manitoba, Ontario and Quebec? All provinces need “policy frameworks” to guide them now and “longer-term” as they weigh tempting AI data centre private investments against the significant risks they bring, urges the Canadian Climate Institute.

Here’s what the AI data centre explosion means for all of us who don’t happen to be policymakers or tech titans.

Start with bigger monthly electricity bills. Rising energy demand inexorably leads to rising prices for electrons or the methane making them. In regions that have welcomed AI energy cannibals into their midst, such as Ohio, electricity prices have already jumped by 26 per cent. Thanks to the AI revolution, the cost of building a natural gas power plant and its turbines has gone up three times since 2022. Just ordering one can take five years.

Your water bill likely will go up, too, as your community competes for dwindling supplies because data centres, which need to be constantly cooled, worsen water scarcity.

In fact, AI grimly compounds the dangerous dynamics of the energy-water nexus. It is impossible to produce energy without water, whether we’re talking about spinning turbines with dammed reservoirs or injecting water underground to dislodge oil and gas through fracking. At the same time, it is impossible to clean and pump water without energy.

Now consider this: data centres owned by Google alone used nearly five billion gallons of fresh water for cooling in 2022. A recent study concluded that AI global growth will command 1.7 trillion gallons of water by 2027. That’s half of the total annual withdrawal of the United Kingdom. Obviously, such demand isn’t sustainable, let alone moral. The study added that “AI’s water footprint can no longer stay under the radar and must be addressed as a priority as part of the collective efforts to combat global water challenges.”

Beyond forking out more for electricity and water, each of us will pay the ultimate price for bowing to the AI empire — defeat in the fight against the climate crisis.

Alex de Vries-Gao, a Dutch researcher and founder of Digiconomist, a firm that exposes the material consequences of digital growth, dares to state the obvious. AI’s lightning-speed growth “clashes with other social ambitions, such as achieving climate goals and reducing total energy consumption.”

The comforting dream that energy conservation would improve while fossil fuels would be replaced by renewables, thereby averting climate catastrophe, has evaporated and been replaced with a technological horror show. The big tech companies that once promised a green future have changed their tune. They want unlimited energy for their AI revolution, they want it now, and they don’t care if the source is coal, methane, geothermal, renewables or nuclear power plants.

Data centres, by the way, must run 24-7 and need constant supplies of energy. So renewables like wind and solar, even with a massive battery farm, aren’t a first choice for Big Tech because of their intermittent nature.

Powering less freedom, more fragility

Add to our chaotic prospects the way our daily lives will depend upon elaborate systems increasingly vulnerable to malicious attacks. Thanks to AI, it is much easier to automate and enhance a digital takedown of an electrical system. The International Energy Agency has documented the growing vulnerabilities posed by AI: “Cyberattacks on energy utilities have tripled in the past four years.”

One way to measure our individual dependence — and impact — is to ask how much electricity each of us uses up whenever we receive a bit of AI “help” via our phones or computers.

Figuring that out isn’t an easy task, because the tech giants are no more forthcoming on the details of the energy consumption of their AI language models than they were on the addictiveness of their social media algorithms.

The MIT Technology Review took a shot at approximating by employing best estimates derived from open AI models in the public domain. They found, for example, that large language models like ChatGPT consume anywhere from 114 joules to 6,706 joules per response to a question. That’s like running a microwave from one-tenth of a second to eight seconds. The more parameters an AI model consults, the greater its accuracy and the greater its energy draw. To manufacture a five-second AI video showing an idiotic skier on a sand dune, for example, uses 3.4 million joules. That’s like running a microwave for an hour for the hell of it.

Multiply that act of energy blindness by billions. Then calculate that AI use is still a digital infant in the scheme of things.

Training an AI program is even more energy intensive. Training OpenAI’s GPT-4 has swallowed $100 million and gobbled 50 gigawatt hours of energy. That’s enough to keep San Francisco going for three days or to electrify thousands of homes for a year.

Let’s be clear about what we are doing here. We are using lakes of clean water, reviving coal-powered stations, gulping methane, raising electrical prices, squandering agricultural land and cannibalizing electrical supplies to ask inane questions and make videos about stupid shit.

Boosters of AI tell us the future must be transformative and artificial. But the true, alarming implications are coming into focus as we see that AI models, as they get smarter, can now behave like their makers. They can evade human control. They can rewrite code to avoid being shut down. They can disobey commands. They can falsify information. They can deceive and cheat. They can even lie to avoid modification.

Every once in a while, a crack in the facade of the AI empire is chipped open by one of its elites, and when that happens it’s wise to pay close attention. A matter of survival, even.

Judd Rosenblatt is an AI CEO whose social media tag says, “I am trying to make AI go well and ensure a flourishing future for all conscious life.” Last week he warned in the Wall Street Journal that “the gap between ‘useful assistant’ and ‘uncontrollable actor’ is collapsing” and went on CNN to reveal that some AI programs are not only defying their creators but threatening to blackmail them.

Why, in every possible way, are we meekly and madly handing ourselves over like doomed hostages to artificial intelligence and its (temporary) masters?

Who asked for this? Who voted for this? Who will resist?

How many of us, before it’s too late, will lift our eyes from our screens to reclaim our freedom to be human and live on a healthy planet?