How big has the AI buildup been? What major problem with that is on the way? If AI proves to be a bubble, what of value would stay?
The oldest
piece here, “AI energy demand in US will surge but also provide opportunity to
manage energy” (Aislinn Murphy, Fox Business, April 18th) told
us that “the world, particularly the United States, is projected to see a
massive jump in data center and artificial intelligence demand for electricity
by 2030, per a recently released International Energy Agency (IEA) report.” That happened not only in five years but
within six months, though we can’t yet vouch for the prediction that “renewable
energy sources will meet nearly half of the additional demand, followed by
natural gas and coal, with nuclear starting to play an increasing important
role.”
With that,
let’s look at “What AI’s insatiable appetite for power means for our future”
(Kurt Knutsson, Fox News, June 20th). Even less than five months ago, “the modern
AI boom” was “pushing our power grid to its limits,” as “the energy needed to
support artificial intelligence is rising so quickly that it has already
delayed the retirement of several coal plants in the U.S., with more delays
expected,” and “energy is becoming the next major bottleneck.” As the previous author also wrote, power is
going for “running” it “at scale,” for current use of the technology, not for
creating models for future releases. Perhaps
unexpectedly, 30% to 55% “of a data center’s total power use” goes to “keeping
AI servers from overheating,” and, overall, “the demand for AI is growing
faster than the energy grid can adapt.”
Despite pledges to use renewable energy, much of that may be nuclear
instead of wind, solar, or hydro, and even if not, “because the grid is shared,
fossil fuels often fill the gap when renewables aren’t available.”
In “At
Amazon’s Biggest Data Center, Everything Is Supersized for A.I.” (June 24th,
The New York Times), Karen Weise and Cade Metz reported that “a year
ago, a 1,200-acre stretch of farmland outside New Carlisle, Ind., was an empty
cornfield. Now, seven Amazon data
centers rise up from the rich soil, each larger than a football stadium.” The company plans to build about 23 more
there “over the next several years,” which “will consume 2.2 gigawatts of
electricity – enough to power a million homes,” along with “millions of gallons
of water to keep the chips from overheating.”
When fully constructed, this facility “will be the largest power user in
the state of Indiana by a country mile.”
People
connected with rural areas may not mind the jobs and money such projects bring,
but per Ivan Penn and Karen Weise in the August 14th New York
Times, “Big Tech’s A.I. Data Centers Are Driving Up Electricity Bills for
Everyone.” Even though “Amazon, Google,
Microsoft and other technology companies” are moving “into the energy
business,” “the average electricity rate for residents has risen more than 30
percent since 2020,” and as “recent reports expect data centers will require
expensive upgrades to the electric grid,” “A.I. could turbocharge those
increases,” “unless state regulators and lawmakers force tech companies to
cover those expenses.”
Similarly,
“AI Isn’t Free. The First Costs Are on
Your Bill, and More Are Coming” (Kay Rubacek, The Epoch Times, September
24th). With rising electric costs
common nationwide, “despite the technological advancements, computing power is
not getting more efficient in terms of power usage. It is becoming ever more energy-hungry.” As such, “the Department of Energy now warns
of a hundred-fold increase in blackout risk by 2030 if data center growth
continues and plants keep closing on schedule,” yet “experts cannot accurately
predict (AI’s) future costs because the technology is changing too fast.”
General-public
reactions to AI power and water use are coming in. They are often not positive, as “AI Data
Centers Create Fury from Mexico to Ireland” (Paul Mozur et al., The New York
Times, October 20th). “In
country after country, activists, residents and environmental organizations
have banded together to oppose data centers,” but “there are few signs of a
slowdown,” as, per bank UBS, “companies are expected to spend $375 billion on
data centers globally this year and $500 billion in 2026.” In Ireland in particular, where “a third of
the country’s electricity is expected to go to data centers in the next few
years, up from 5 percent in 2015,” the “welcoming mood has soured,” and it has
now “become one of the clearest examples of the transnational backlash against
data centers,” as “a protest movement has grown.” “Impoverished small towns” in Mexico near
where data centers have appeared have “began experiencing longer water
shortages and more blackouts.”
It is clear
from all this that the rubber of increased AI infrastructure is meeting the
road of damage to residents. There will
be vastly more conflict next year, much of it, even in the United States as
protests multiply, preventing data centers from being built. That will become yet another problem for the
technology to overcome, and will push costs even higher.
I have been
reading about the possibility of a severe artificial intelligence downturn, and
comparisons and contrasts with what happened almost 200 years ago with railroads. Then, the failed companies left behind track,
bridges, and stations that were later used when the industry reconstructed
itself. What would AI abandon? Failed companies’ data center buildings would
remain, but the chips would, as now, be worthless well under a decade
later. While the news that it is not
upgrades driving current resource usage is heartening, and the chance of what
is now a vast number of profitable and worthwhile applications disappearing is
almost nonexistent, companies going bust could mean the end of tens of
trillions in market capitalization. It’s
easy to imagine effects such as a 50% NASDAQ-index fall. Yet those gigantic physical structures will
still be useful. How, we don’t know, but
they will be, one way or another.
No comments:
Post a Comment