Architects who design data centers are grappling with one of the biggest challenges their field has seen: How to create buildings that can handle the intense power and cooling demands of artificial intelligence.

These warehouse-sized properties house rows upon rows of computer servers and other equipment, machines that serve as the backbone of the internet by processing vast amounts of data — think tens of thousands of movies streaming simultaneously — and generating excessive heat in the process.

Now artificial intelligence, or AI, is requiring data centers on steroids to process the growing amount of data being generated. Just like e-commerce drove unprecedented demand for industrial warehouses, AI development is expected to more than double the need for global data center storage capacity by 2027, according to JLL’s Data Centers 2024 Global Outlook.

The market for generative AI — an advanced version of artificial intelligence that generates new content and learns data patterns without human direction — is expected to grow to $1.3 trillion in the next decade from $40 billion in 2022, according to Bloomberg Intelligence. That surging demand is already rippling across the U.S. economy as the soaring shares of Nvidia, the maker of computer chips used in AI, drove the S&P 500 stock index to records in the past week after reporting its quarterly revenue more than tripled.

The more powerful servers used in AI that are densely clustered produce even more heat than traditional data centers. Generative AI’s greater energy requirements — ranging from 300 megawatts to 500 megawatts per data center, an amount that can power 300,000 to 500,000 homes — are expected to also require more energy-efficient designs and locations, which in turn will require data center operators to increase efficiency and work more closely with local governments to find sustainable energy sources to support data center needs.

To compensate, architects are trying out multiple designs. One such approach is designing and building data centers with higher ceilings to create pathways for liquid-cooling equipment on top of servers, according to architecture firm Page Southerland Page.

Other data center operators may not want water flowing near digital equipment. So raised floors may come back in vogue to hold cooling tubes. Architects in recent years had switched to concrete slabs to support servers that had become more heavy, according to AnD Cable Products, a maker of data center equipment. Now, the extreme heat generated by AI servers is expected to require the development of floors that are raised to make room for cooling tubes but are also capable of supporting heavy machines.

The additional cooling equipment, plus the extra servers needed for AI processing, means buildings housing these types of data centers may require more square footage.

As a result, AI-specialized data centers are expected to be taller, cover more ground and perhaps be located in more rural areas.

Some governments are stepping up to help with the construction costs. The state of Georgia last year extended its sales and use tax exemption for high-technology data center equipment through 2033 while adding a requirement that a data center create at least five jobs in counties with less than 30,000 people and at least 10 jobs in counties between 30,000 and 50,000 in population. That rises to at least 25 jobs in counties with more than 50,000 people.

Each data center must also have a minimum investment of $25 million in counties with less than 30,000 people, rising to a $75 million minimum investment in counties with populations between 30,000 and 50,000 and climbing further to at least $250 million in counties with more than 50,000 people.

In Ohio, the state's Data Center Tax Abatement Program offers full exemption from the sales tax on construction materials, computer equipment, mechanical and electrical equipment, cooling systems and power infrastructure once a company invests at least $100 million over three years and maintains an annual payroll greater than $1.5 million.

Architects and construction managers are trying to figure out the best way to adapt to this profound change, Tom Earp, principal engineering director at Page, told CoStar News. Designers have ideas, but no accepted industry standard has emerged yet because the technology is expected to constantly be changing, he said.

“It makes the design process incredibly challenging because no one can be certain that a specific data center design can handle all the future hardware,” Earp said. “It’s short-sighted to do design work to handle only one or two generations of server hardware.”

Part of that design challenge involves AI requiring data centers to use more sustainable power sources such as solar and wind, according to JLL’s Data Centers 2024 Global Outlook. They may need to be two or three stories tall, instead of the current standard one-story model, to compensate for the denser placement of server racks, according to Page. And, there may be a need for more data centers, though it's not known how many, according to Digital Realty, a real estate investment trust that owns and develops data centers.

“AI stands to be the biggest transformation in the data center industry since the sector burst onto the scene,” JLL said in the report.

Already driving the change and design challenges is spending by Meta, Amazon, Microsoft and Oracle, technology companies that spent nearly $160 billion in 2022 to boost their cloud computing power an average of 30% over the past five years, according to Newmark's U.S. data center market report published in January. That spending trajectory — an indicator of data center demand — is likely to continue because of the accelerating pace of AI breakthroughs.

A good portion of that spending on data centers has ended up in central Ohio’s large swaths of commercial land in the past few years. Data center inventory in the Columbus, Ohio, area grew 146% between 2012 and 2021 to 4.6 million square feet, according to Newmark.

Other major data center users such as Google and Facebook are spending billions of dollars to expand or set up shop in the area to satisfy the "gargantuan projected" demand associated with AI, making Columbus into a mature data center market, according to Cushman & Wakefield's latest Americas data center report.

The average size for data center land deals in central Ohio over the past 18 months is 218.9 acres, according to Newmark. Land prices are expected to rise as sellers recalibrate to meet the appetite of hyperscalers, Cushman & Wakefield said, referring to the largest types of data centers that are typically built to allow major expansions.

Companies are also expanding on large parcels elsewhere around the world. Blackstone, the world's largest alternative asset manager, recently partnered with Austin, Texas-based Digital Realty to invest $7 billion to develop 10 data centers across four massive campuses over the next few years in Northern Virginia, Paris and Frankfurt, Germany.

Blackstone acquired data center developer and owner QTS Data Centers in 2021 for $10 billion and, just last year, QTS signed at least $8.5 billion of development deals pre-leased to major global technology companies that need more AI capabilities.

AI is also pushing companies to convert or expand some existing facilities to handle the new technology or perform data center retrofits where possible, according to architecture firm Gensler's 2024 Design Forecast report. It points out the conversions are worth the effort because former corporate and research campuses on the periphery of urban centers offer attractive acreage and existing architecture for data centers.

In Ohio, Aligned Data Centers bought a former General Motors plant built during WWII on 129 acres last year with plans for a four-building data center campus. The site at 2509 Hayes Ave. in Sandusky has its own electric substation with 80 megawatts of power that can be expanded up to 200 megawatts.

Equinix, a Redwood City, California-based real estate investment trust that specializes in data centers, is experimenting at an Ashburn, Virginia, facility with different ways to handle AI’s demands. Equinix is testing “liquid cooling technologies, such as direct-to-chip, immersion and rear door heat exchangers,” the company said in its recent earnings report.

Rural locations also hold appeal for AI-enabled data centers, with low energy costs, abundant sustainable energy sources and colder temperatures than what's found in densely populated cities, JLL found.

More data centers are now located in urban areas that are closer to users and to speed things such as on-demand video streaming. But AI processing does not need to be located near the end-user, Earp said. That could lead to more AI-enabled data centers in remote areas.

Another benefit of rural areas is that cooler air temperatures found outside cities, and in colder climates, reduce the need for cooling operations. Google, Meta and Amazon have all established data centers in the Nordics over the past several years and more growth is expected in the region in coming years, according to JLL.

Applied Digital, a Dallas-based data center developer, recently began construction of an AI-enabled data center in Ellendale, North Dakota. Even with the cool North Dakota climate, Applied Digital still uses liquid cooling for the server equipment in the 342,000-square-foot data center.

“I just want to say thank you to the team in Ellendale, North Dakota,” Applied Digital CEO Wes Cummins said in a conference call last month to discuss earnings. “We’re sitting here in January and it’s not the most pleasant climate for them to be continuing construction on our site.”