The warp speed developments of artificial intelligence are propelling exponential growth — but also ravenous power and cooling appetites — for data centers, which are driving the technology backbones behind the machine learning upheavals across the world.
The AI-powered data storage market is projected to grow from $28.7 billion this year to as much as $217.3 billion by 2033, according to a market analysis by Precedence Research.
That’s just one of scores of bullish and ambitious growth expectations for AI’s impacts on server farms and other technology infrastructure.
Dell Computer founder CEO Michael Dell said at this year’s South by Southwest (SXSW) conference that AI could drive 100 times more demand for data centers and their computing capacity over the next 10 years.
Dan Watts, CEO of US Signal (which operates data centers in Wisconsin, Michigan, Illinois and Indiana, and is adding others in Oregon, Colorado and Arizona via its recent acquisition of OneNeck) agrees.
“It’s going to be a massive growth driver for data centers,” said Watts.
He sees AI hyperscalers — big technology companies such as Microsoft, Google/Alphabet, Amazon, Oracle and IBM that provide cloud computing service and operate big data centers — driving approximately 60% of that growth in the short-term.
NUCLEAR OPTIONS
Those same big tech firms, along with private equity technology investors and data center developers, are also confronted with AI’s voracious energy and cooling needs, potential strains on power grids and water supplies and their climate-change inspired promises to curtail carbon emissions.
“AI and machine learning are very power intensive,” Watts said.
An analysis by Goldman Sachs projects AI driving a 160% increase in power demand from data centers by 2030.
“On average, a ChatGPT query needs nearly 10 times as much electricity to process as a Google search. In that difference lies a coming sea change in how the U.S., Europe and the world at large will consume power — and how much that will cost,” reads the Goldman Sachs analysis released in May.
All that worries environmental groups concerned about climate change, droughts and rising temperatures already straining power and water systems.
“American energy consumption is going up for the first time in 20 years. Data centers are a huge part of what’s driving that,” said Michael Khoo, a program director with Friends of the Earth.
Khoo foresees potential battles over power supplies and water allocations between high-intensity data centers and other users. Data centers require cooling — often in the form of water-based systems — to keep servers and processors from overheating.
“I see very real fights for power access for local communities versus data centers,” he said.
For utilities to keep up with data center needs, U.S. power grids could need a $2 trillion overhaul, according to commercial real estate firm JLL.
Private power grids — including some that could be powered by small-scale and micro nuclear reactors — could be developed to keep up with the capacity demands of AI data centers processing oceans of data and electronic transmissions.
“We’re just using so much technology. That is not going to change,” said Pat Lynch, executive managing director and global head of data center solutions for CBRE, a national commercial real estate firm.
Lynch said small modular nuclear reactors are being seriously explored by large data center operators with concerns about whether renewable sources and traditional grids can keep up with demand while also avoiding increased carbon emissions.
“The demand is migrating to the places where they can now find power,” he said, noting potential natural gas and hydroelectric paths to powering and cooling high-volume data centers.
Small nuclear reactors to power and cool data center still have to gain regulatory approvals and could face community concerns. But there is interest in their development from big money players, including Microsoft, as well as its founder, Bill Gates — who is developing a new sodium-cooled nuclear power plant in Wyoming.
The nuclear energy segment for data centers could ramp up to a $38.3 million marketplace globally by 2028, and hit as much as $278.4 million by 2033, according to BIS Research Inc.
SPACE INVADERS
There are more than 5,380 data centers currently in the U.S. and close to 11,000 worldwide, according to technology firms Cloudscene and Brightlio.
Data centers are currently located across the country. That includes near major internet and telecommunications hubs in Silicon Valley and Northern Virginia outside Washington, D.C., as well as other big cities and regions including the Pacific Northwest, Midwest and Mountain West (including Idaho and Wyoming).
Watts said three components to data centers are being monumentally impacted by the AI revolution.
“You’ve got floor space. You’ve got power, and you’ve got cooling,” said Watts, referring to space and power needs, as well as ways to keep servers and IT infrastructure from overheating.
POWER PLAYS
Data centers currently consume 3% of power nationally — a figure that could rise to 8% by the end of the decade.
That could be accompanied by a doubling or more of carbon dioxide emissions from data centers by 2030, according to the Goldman Sachs analysis.
Utilities (large and small) are trying to navigate the new AI universe.
“In some areas of the country, the demand for electricity is growing faster than the available capacity,” said Ben Fowke, CEO and president of American Electric Power, during a congressional hearing in May. AEP operates in 11 states, including Ohio, West Virginia, Tennessee, Texas and Louisiana.
“Just a few years ago, a large-scale industrial manufacturing facility might require 100 megawatts (MW) of electricity – enough to power about 100,000 homes. A facility that size would typically be one-of-a-kind in a region and would be a major source of economic activity for the area.
Now, it is common for a single data center to require three times to up to 15 times this amount of power for a single site,” Fowke said.
Lynch said AI-driven demand is driving limited supply for exponentially larger data center capacity, but sees technology infrastructure developments in Atlanta and Dallas. He also sees more projects landing in Ohio, Indiana, Nebraska and Nevada, as well as areas of Virginia and Maryland with ample power grids. Data centers avoid places with potential weather and other power disruptions.
“What we’re seeing today is growing demand and very little supply — mostly limited by utility capacity,” he said.
Lynch said some existing data center space could be retrofitted to accommodate AI needs, but said that can be challenged by the need to maintain existing operations, and some might eventually become obsolete.
The commercial real estate broker sees rural areas landing more data centers related to cryptocurrency (which have less intensive power needs than their AI counterparts).
“It seems like crypto is a more viable option for those communities,” Lynch said.
Local News
Comments / 0