The demand for data centers has drastically increased over the past five years with the proliferation of artificial intelligence. As technology companies compete for the future of AI, they are also investing heavily in the construction of new facilities that will power the notoriously computation- and resource-intensive technology.
According to MIT Technology Review, there are currently over 3,000 confirmed data centers across the U.S. that hold the servers and accessory cooling systems run by cloud computing companies and technology companies. A report from McKinsey forecasts a $5.2 trillion investment in AI infrastructure by 2030, with 15 percent of that flowing to builders for land, materials, and site development.
Unsurprisingly, large architecture firms have rapidly developed sector-specific expertise in data center construction. Though many practices have long maintained devoted data center divisions, these teams have ballooned in profitability as technology and real estate interests coalesce to build bigger.
The recent rise in AI usage has made data centers a profitable enterprise for architects. In 2020, Corgan and HDR topped Building Design + Construction’s annual list of architecture firms in the data center sector, respectively earning approximately $61 million and $34 million from their data center projects. In 2024, the same list reported that Corgan had more than doubled its profits, reaching roughly $135 million, while HDR quadrupled its, with $130 million in earnings.
AI’s Manifest Destiny
Tom Widawsky, a data center designer and technical principal at HDR, explained to AN that the impetus behind market growth lies with hyperscalers, or companies like Google, Meta, Microsoft, and Amazon, which provide cloud computing and data management for clients that require immensely powerful operations. As its name suggests, hyperscalers provides cloud infrastructure that allows software to scale and grow according to demand, a process that requires data centers that span at least 10,000 square feet, with the capacity to hold more than 5,000 individual servers. “AI is still in what they call the inference gathering stage,” Widawsky said. These companies are “competing to launch into the next phases of what it will do and how it will actually get into application mode.”
Because of their energy and connectivity demands, data centers have typically been built in places with extensive fiberoptic networks and reliable energy sources, like the suburbs of Northern Virginia, which to this day maintain the highest density of data centers in the world. However, shrinking land availability has diminished owners’ ability to upgrade the computational power of their existing facilities, prompting developers to look to more unconventional sites, which often lack the robust electrical infrastructure needed to support the operation.
Traditionally, data centers have maintained a low profile as largely unremarkable, windowless warehouses. But the push to alternative sites is reshaping how architects are approaching design. “The demand that they’re having to serve is changing so quickly that we do have to reevaluate almost the entire building as part of each delivery,” said Dan Drennan, managing principal and data center leader at Corgan, where 300 of its 1,200 employees work on this project type. “Whether that’s the way the envelope responds to a different location or it’s the way the structure needs to respond to a different configuration based on the availability of land or the piece of land we’re working on. So it’s not as cut-and-paste or repetitive as I think folks might wish.” He also shared that Corgan is doing more multistory configurations than in past years.
As data centers remain sites of increasing contention, their design has shifted toward making the buildings more palatable to their local communities. “One of the big things that we’re seeing is that a higher level of design is being required on the exterior of these projects so that they blend in as much as possible with their surroundings,” reflected Joy Hughes, global accounts director at Gensler, in a recent conversation about data centers with Ryan Cook, Gensler’s critical facilities leader for the Southwest region. “We try to work with the jurisdictions and have some dialogue with the community about how these facilities are designed and built to ensure that we are sensitive to their concerns.”
Offsetting Environmental Impacts
It is estimated that a user’s prompt to OpenAI’s ChatGPT consumes ten times as much energy as a Google search. Technology’s appetite for electric power is projected to skyrocket as artificial generative intelligence becomes further integrated into our day-to-day lives. Recent reports already confirm that 4.4 percent of all energy in the U.S. goes toward data centers, and that number could nearly triple to 12 percent by 2028.
In some cases, corporations are so desperate to open a facility that they temporarily power the building using fossil-fuel generators as they await permitted utility access or infrastructure upgrades. These generators are also employed as a redundancy in case of an outage. Many data centers, particularly those hosting cloud storage, operate 24/7 and demand constant power.
The strain on the power grid from data centers can be significant and elevate prices for nearby residents. Though many utility companies are actively working to upgrade their service to support the energy consumption of data centers, reporting from The New York Times suggests providers will also pass many of these costs on to everyday consumers through hiked energy bills. Though renewable energy is growing quickly, the power grid in the U.S. is still largely powered by fossil-fuel generation, which can pollute local communities close to power plants or chemical production facilities.
Largely unable to address the energy supply issues the data center sector is facing in the era of AI, climate-sensitive architects have instead focused their efforts on aspects of projects that they can control, such as reducing emissions through materials sourcing and the specification of efficient mechanical systems. Drennan shared that as processing power has increased, the requirement for cooling, whether through air or liquid systems, has increased, which is one of the limiting factors in upgrading existing data centers. While water usage remains a concern for some facilities, many systems are going to closed-loop liquid cooling, which reduces water needs.
“The good news is that a lot of the times the things that make projects more efficient also align with sustainability aspirations,” said Drennan. “Whether you’re using local resources to meet a sustainability requirement or just because it can get to the site faster, it helps [the client] either way.”
One example is the use of mass timber in data center construction. In Northern Virginia, Gensler and Thornton Tomasetti have designed the first hyperscale facilities in the country to feature a hybrid timber and steel structure. Other firms hope to quickly follow suit.
“We think that data centers actually have a natural ability to use mass timber,” Widawsky offered. “Unfortunately, clients are more interested in mass timber for their administrative buildings and not so much for data centers. We also try to push improvements in low-carbon concrete,” he added.
Unfortunately, mass timber’s carbon sequestration properties will have diminishing returns as the energy consumption of data centers increases. It is predicted that soon some of the largest facilities will surpass the energy consumption of small cities. A 2024 report from Morgan Stanley forecasts that the sector’s emissions will reach 2.5 billion metric tonnes of carbon dioxide by 2030. One solution that seems to have some promise is the arrival of small modular reactor technology—basically, tiny nuclear power plants.
Until our energy grids run on renewable energy, the proliferation of data centers across the country portends a worrisome rise in carbon emissions. Because a majority of data centers’ emissions are created through operation and not construction—the inverse of a typical building project—design professionals are limited in their ability to control the typology’s performance, which is currently subject to the whims of consumer demand.
Trevor Schillaci, formerly AN’s senior programming associate, is a master of architecture student at Rice University.
→ Continue reading at The Architect's Newspaper
