CHIPPED: Part 3
data center NIMBY, power play
It has been a century since physicist Julius Edgar Lilienfeld, an immigrant to the United States, patented the idea of using a semiconductor material to make a transistor. A hundred years later, silicon microchips, some with tens of billions of transistors, are in everything from computers to cars to coffeemakers. They make our modern world possible. Now, with Artificial Intelligence, they are poised to run the world.
CHIPPED is divided into five parts. Each can be read independently, but there is an arc to the order.
There is no digital without physical and no bits without electrons flowing through the grid. A single H100 Nvidia Graphics Processing Unit (GPU), a chip that makes AI possible, uses 700W of electricity. Two chips use more energy than a typical American household with an air conditioner running 24/7. Nvidia has sold millions of these chips.
Elon Musk’s company, xAI, recently announced plans to supersize its supercomputer, Colossus, which trains models for its AI service, Grok, from 100,000 H100 chips to a million.
It won’t be the only one in the million-chip club. Broadcom, an Nvidia competitor with its own AI chip—XPU—reports it has several clients building million-chip supercomputers, too.
Despite improvements in chip efficiency, the explosive growth in the sheer number of chips has swamped any energy-saving gains. The data centers that house these chips will need more and more energy to operate and more and more water for cooling.
BUILDING BOOM
In the two years since AI went from a science fiction trope to everything on display “now with AI!” at the Consumer Electronics Show (CES), hundreds of billions of dollars have poured into building data centers. Not only are there more of them, and not only are they are getting bigger (“hyperscale”), but advances in chip performance have dramatically increased data center capacity. There is lot more computing happening.
Over the last five years, the number of hyperscale data centers has doubled to about 1,100, with at least another 400 in development, according to industry analyst Synergy Research Group. In another five years, hyperscale data center capacity is expected to triple, largely due to AI.
Private equity giant Blackstone considers data centers “one of the most exciting and critical investment themes of our time.” But what was until recently seen as a boon for local economic development has become a pariah blamed for driving up electric bills, competing for local water supplies and leaving landscapes blighted with large, vaguely ominous buildings without any windows.
The amount of energy used by data centers is astonishing. According to Lawrence Berkeley National Laboratory, between 2018 to 2023, data center energy use in the US more than doubled from 1.9% to 4.4% of total energy consumption. AI, which burst onto the scene at the end of 2022, is thought to have contributed significantly to the bump. In another three years, data centers could account for as much as 12% of US energy use.
It isn’t hard to see why. An AI search query requires 10x the energy to process than a standard Google search because of all the computations involved. Everything “now with AI” uses more energy.
NIMBY
For years, Ireland lured tech companies with generous tax incentives, a strategy that worked like a lucky charm. The island’s moderate climate, which makes it easier to keep data centers cool, was an additional draw. But a good thing became too much of a good thing when AI sparked a frenzy in data center construction. Data centers now account for 21% of Ireland’s electricity use, a figure that could hit 30% by the end of the decade.
With electric bills soaring and the specter of rolling blackouts looming, the South Dublin County Council recently rejected Google’s plans for to build yet another data center. Dublin has become a cautionary tale. Data centers are now a hard sell anywhere on the island.
Increasingly, they are a hard sell anywhere, including in the US.
Fifteen years ago, Virginia began offering a combination of “pioneering tax advantages” and the promise of cheap electricity to attract data center developers. It worked. According the Virginia Economic Development Partnership, the state now “hosts the largest data center market in the world…home to more than 35% of all known hyperscale data centers worldwide.”
But for many, what was sold as a tech miracle has turned into Sorcerer’s Apprentice-level nightmare. Data centers in just three counties—Loudoun, Fairfax and Prince—collectively use 4,410 megawatts (MW) of electricity. That’s enough to power at least 882,000 homes. To keep up with demand, which shows no signs of slowing, the push is on to quickly approve a large gas-fired power plant, along with new gas pipelines, to generate more electricity.
The promise of a shiny digital future is binding us ever more tightly to a dirty fossil past.
Data centers can be noisy neighbors, too, as the residents in Loudoun County discovered. The source of a persistent “eerie hum” that could heard for miles turned out to be the whir of giant fans at a nearby data center taking advantage of “free air cooling” when outdoor temperatures ranged between 35° and 70°F.
In interviews and in over 40 noise complaints to the county obtained through a public-records request, residents described the noise as sounding like a propeller, a "loud drone hovering above 24/7," a "big fan," a construction site, a "low-frequency sound," an airplane engine, a helicopter, a freight train, a leaf blower, and a lawn mower.
••••••••••••••••••••••••
Meanwhile, in Tennessee, Musk’s expansion plans for Colossus have hit a snag over electricity needs. “It’s a physics problem, not a political problem about how much energy can be provided,” says Doug McGowan, CEO of Memphis Light, Gas and Water (MLGW).
The issue is two-fold: The challenge of generating enough energy and the challenge of delivering it. There is a limit to the number of electrons the grid can handle locally, but also nationally. The US grid, an aging network of transmission lines, transformers and substations that has been described as “creaky” and “decrepit,” is in urgent need of a $2 trillion upgrade.
Under the Inflation Reduction Act, $5.7 billion has been spent on grid upgrades over the last two years. The US would need to increase annual spending more than forty-fold to catch up with China, which recently announced plans to spend $800 billion to upgrade its grid over the next six years.
Parts are another problem. Transformers are in short supply, with an order backlog that stretches for years. Without them there is no way to convert the electricity that flows through high voltage transmission cables into the low voltage version that flows into buildings.
Extreme weather and wildfires, often linked to climate change, have also taken a toll, dramatically increasing the number and severity of power outages. Instead of building a grid for the 21st century, money has been funneled into short-term emergency fixes.
THE NUCLEAR OPTION
Over the last year, climate pledges by Google, Microsoft, Meta and Amazon to reduce greenhouse gas emissions have taken a back seat in the fiercely competitive race for AI dominance. To meet net zero commitments while also addressing the need for vast new supplies of electricity to run data centers, all four companies recently announced plans to add emissions-free nuclear energy to their power portfolios, positioning it as better than solar or wind because power generation would be 24/7, rather than limited to when the sun shines or the wind blows.
But the nuclear option comes with another serious environmental concern: radioactive waste. Nearly 90,000 tons of spent radioactive fuel is currently in temporary storage at reactors in the US because there is no permanent place to put it. Critics see this as a disaster waiting to happen. Nuclear-boosters note that a permanent storage facility wouldn’t need to be very big, roughly the size of a 10’ deep football field. But locating the right spot has been a political non-starter for decades.
...I think what people don’t realize is that it (storage) is actually a serious technical challenge. The half-lives of some of these elements stretch into tens, if not hundreds of thousands of years. We’re asked to design solutions that will last as long as the risk. That’s not something we usually do. The technical and scientific challenge for nuclear waste is, whatever our solution, that we will never see whether we were correct or not. Designing a system where you don’t have feedback is very difficult…”
Until there is a permanent solution, the federal government pays nuclear operators about $2 million per day to cover the costs of onsite storage. Any new nuclear plants would add to the taxpayer-funded tally, which is now stands at roughly $9 billion.
RATE HIKES
The AI power grab is just getting started, according to a recent report from investment bank Goldman Sachs: “AI is poised to drive 160% increase in data center power demand.” US utilities will need to spend $50 billion in new generation capacity by 2030, advises Goldman, along with investing in new gas pipelines to handle an extra 3.3 billion cubic feet per day.
Following a decade when US power demand was basically flat—even with a growing population and an expanding economy—demand is now soaring. Energy efficiency gains that took decades to achieve have been quickly gobbled up by an insatiable, power-hungry AI Pac-Man. Coal plants that were scheduled for retirement as part of the clean energy transition are being kept in operation indefinitely because of AI.
Utility bills are soaring, too. To cover an eight-fold increase in the cost of supplying electricity during periods of peak demand, rate-payers in the Northeast and around Chicago will see their electric bills rise by as much as $20 per month ($240 per year) by mid-2025. PJM, a Regional Transmission Organization (RTO), is responsible for making sure there is enough power in the system. Power providers bid for business at PJM auctions. The last auction broke records.
“…For families already squeezed by inflation, this is bad news. And this isn’t a onetime bump. It’s a shift that will impact electricity prices for years to come. The next PJM capacity auction is scheduled for 2025, which will set the price of peak electricity for 2026-2027, and experts think the electricity prices could increase even further, causing increases of as much as $40-50 per month for a typical household…”—Fast Company
An extra $240 spent on electric bills won’t be available to spend on anything else. There are 65 million people living in PJM’s territory. Assuming four people to a household, that’s 16.25 million households. $240 x 16.25 million is nearly $4 billion that won’t be spent on clothes, cars, consumer goods, food, education or entertainment.
CHIPPED: Part 4 | water wars, 12 years from rock to e-waste