A.I.’s Believers Don’t Want to Know About Trouble Ahead

Artificial intelligence promises to diagnose baffling illnesses, design new drugs, solve mathematical conundra, and unravel a long list of puzzlements that have proven difficult for humans to fathom.

The savants of Silicon Valley take these promises a good deal further. They speak of bringing humanity into a new age of enlightenment, seeming to revel in their god-like intentions to create a superintelligence destined to make humans obsolete.
Formerly a cornfield, Amazon's data center for A.I. at New Carlisle, Indiana.

not hiringSam Altman, the most prominent voice of the A.I. universe and CEO of OpenAI, the company that created ChatGPT, has written that a next step, artificial general intelligence, or A.G.I., will usher in “massive prosperity”.

But Altman also writes that the superintelligence that "humanity is close to building” will lead to "whole classes of jobs going away". Silicon Valley barges ahead nonetheless, making A.I. ever more powerful, unconcerned that their endeavors are likely to put millions out of work.

A.I. is already killing the entry-level job. "The unemployment rate for degree holders ages 22 to 27 hit 5.8% this spring", reports Bloomberg Businessweek. For computer engineering majors, the unemployment rate is now 7.5%. Companies are “going to need less and less people at the bottom”, said one CEO, a grim outlook for new college graduates. One said, possibly not kidding, it's harder "to land an entry-level role at one of the big banks than it is to get into Harvard University". One tech company completely stopped hiring any engineers below a mid-level position "because lower-level tasks could now be done by A.I." Newly graduated economists, normally enjoying 100% job prospects, are now having trouble finding positions as A.I. affects the market for even high-skilled workers.

Top executives at companies such as Amazon and JPMorgan say they plan to shrink their workforces considerably. CEO Andy Jassy sent shivers throughout Amazon with the comment that A.I. "will eliminate the need for certain jobs" as the company scrutinizes “revenue per employee” while examining what A.I. can do faster and cheaper than humans. Ford CEO Jim Farley said he expects A.I. will replace half of the white-collar workforce in the U.S.

Wall Street Journal columnist Peggy Noonan quotes another writer, John Ellis:

“I drive up and down ‘Old Post Road’ in Fairfield County (CT) almost every day. When I do, I pass office buildings and storefronts that are the workplaces of insurance brokers, local and regional bankers, mortgage brokers, lawyers, accountants, consultants, marketers, real estate agents, etc. And what I think about all those people as I pass them by is this: The companies they work for will employ 10%-25% fewer of them in (probably) two years, maybe three.”

What those people do for a living will be done by A.I. Noonan gets it right:

We natter on about what cable news natters on about: Is JD Vance next, can Gavin Newsom make the sale? But the biggest domestic political story of our time is happening now, a remaking of the employment field in America. Mr. Newsom doesn’t threaten Mr. Trump, A.I. does.”

But no fear! In their thrall of technology, tech titans such as Musk and Altman say that A.I. will, through cost savings and greater revenue, create such wealth that the evaporation of jobs will be replaced by a universal basic income, or maybe, Altman said on Theo Von’s podcast, that everyone will be given “an ownership share in whatever the A.I. creates”, an idea that he calls “universal extreme wealth”. It translates as, our future economic security will depend on the benevolence of Silicon Valley’s billionaires.

A GPU is a computer card or chip about three inches square that produces graphic displays on computers. Designed chiefly for video games, they proved ideal for A.I. when wired together in masses of thousands of chips. This, and the software running through them, is what makes a data center, as well as makes chipmaker Nvidia the U.S.’s largest company in market value at over $4 trillion.

There are already more than 5,000 data centers in the U.S., largely housing the internet, cloud computing, streaming, and so on. A.I. is in the process of exploding their size with hundreds of billions of dollars pouring into the building frenzy:

 In Indiana, Amazon will build 30 data centers on 1,200 acres of farmland. Its biggest customer will be Anthropic, a start-up in which Amazon invested $8 billion, which aims to develop an A.I. system that matches the capabilities of the human brain.

 Meta (Facebook, Instagram, WhatsApp) is building a 2-gigawatt data center in Louisiana.

 OpenAI is erecting a 1.2-gigawatt facility in Texas and another, nearly as large, in the United Arab Emirates.

 Elon Musk’s xAI data center in Memphis, dubbed “Colossus”, was built in 122 days of three eight-hour shifts and harnesses an array of 100,000 chips, which he plans to double. They skipped the time cost of planning and dealt with problems as they arose, says the builder.

The U.S. has a history of squandering an irreplaceable asset, its groundwater. Areas around the country are facing serious shortages. With A.I. we’ve come up with yet another industry that consumes vast quantities of a resource on which the future of the country depends.

The Amazon facilities in Indiana, for example, will each year use millions of gallons of water to prevent chips from overheating. Globally, researchers at the University of California Riverside and the University of Texas Arlington tell us that by 2027 A.I. water usage will equal the total amount of water withdrawn by the entire country of Denmark. Or half that of the U.K.

Data centers typically help themselves to a municipal water supply, at least until there’s an uproar over diverting drinking water to industrial use, or when pressure drops to a trickle, or the funny taste reveals contaminants, or the water table is in danger of running dry. The taps went dry for people living near Meta’s $750 million data center in Georgia. Sediment caused one family to replace ruined appliances more than once. The county is on track for a water deficit by 2030. All was foreseeable but Meta probably figured the plant would be obsolete by 2030 and would leave Newton County to cope with the problem.

This comes about when tech companies identify a desirable area, swoop in with enticements to the local officials who are made along with construction workers to sign non-disclosure agreements. By the time residents learn what is happening, it’s too late.

Faced with the crucial need to modernize the nation’s grid for the sake of national security, to bring wind and solar power in from distance, to strengthen against cyber attacks, the power industry is diverted to this new demand of spending to accommodate commercial companies.

That expansion is because a ChatGPT query consumes 10-times the electricity of a Google search. Which explains why the five data centers OpenAI plans will collectively draw as much electricity as the three million homes of Massachusetts.
The innards of a data center..

OpenAI and Meta, operating on their large language models GPT-4 and Llama 3.1, use around 30 megawatts of electricity at a time, according to the nonprofit research group Epoch AI. That’s roughly as much power as 30 Walmart stores use at any given moment. That Amazon sprawl in Indiana is projected to consume at a level of 2.2 gigawatts — enough to power a million homes.

Analysts’ projections vary, but there are forecasts that data center electricity consumption will rise by 2030 to as much as 17% of all the electricity produced by the United States. Globally, McKinsey & Company projects that to support A.I. at its current rate of expansion, the world by 2030 will need to add around two to six times the energy capacity it took to power the entire state of California in 2022.

Remarkably, the A.I. companies simply expect the power industry to meet its demands in whatever state they choose to set up their enormous power-guzzling centers. As public utilities, electric companies may be required to come up with a customer’s request, but who bears the costs of transmission lines, equipment, and even new power plants? This is especially true when A.I. companies choose states which have few or no data centers and will need to make costly upgrades to meet the demand. Wood MacKenzie, an energy research firm, analyzed 20 large power users and concluded that what they would pay to power companies would not be enough to cover their costs of expansion. Only Ohio, as far as we know, has demanded that A.I. companies pay more of the costs of meeting power demands that in that state will exceed 50-times the usage of existing data centers.

To cover the cost of this unbridled industry, utilities will need to seek approval of higher rates, and that means all their customers will find themselves paying heftier monthly electricity bills, subsidizing the noisy and unwanted data centers and the few jobs they provide once construction is done. U.S. power companies are already charging more to cover a surge in spending to upgrade the aging grid, harden facilities against extreme weather, power electric vehicles, and now comes the voracious appetite of data centers.

How can the grid handle this without more power plants? Wind and solar have produced almost all of the increase in the nation's power supply over the last decade, and A.I. data centers that run non-stop
Gas turbines can be as big as a large plane and cost millions of dollars.

cannot rely on these intermittent sources anyway. But a new natural gas-powered plant starting from scratch would probably not go on line until 2030, and one reason for that are the giant turbines at their core. They each take around four months to build, to which add trans-shipment of huge units and installation. A utility placing an order today faces a wait of three to four years.

The story for transformers is much the same. The average age of large power transformers is 38 years and the backlog across the industry runs to over two years.

There seems to be scant awareness that fulfilling the steep rise in demand for power is more likely impossible than possible. How, in less than five years, can the massive U.S. power industry grow by the 17% mentioned above? The panjandrums of Silicon Valley may need to stow their dreams of rearranging how we live our lives because there’ll be a power outage.

There was once a quest to reduce carbon emissions and mitigate climate change, but instead the world has chased after two new fields — A.I. and bitcoin mining — both reliant on round-the-clock computing powered by natural-gas power plants. Tech companies have quietly let their emission pledges go up in smoke: Microsoft acknowledges that its emissions are 30% higher than in 2020 owing to its data centers. Google’s are up 50% over the same period. Eric Schmidt, who made his first billions running Google and is of course invested in and on the board of various A.I. companies, said artificial intelligence is too important to let climate worries get in the way. Quoted in The New York Times he said:

”We’re not going to hit the climate goals anyway, I’d rather bet on A.I. solving the problem.”