What exactly are data centers?

The number of data centers in Pennsylvania has doubled over the past few years, with the state investing tens of billions of dollars in the technology.

Data centers are physical locations that house and maintain infrastructure-supporting cloud computing and the training of artificial intelligence models. As AI becomes more prevalent in everyday use, Pitt professors explain how these invisible backbones of the internet work. 

There are currently more than 85 data centers in Pennsylvania, with about 40% of the centers located in Pittsburgh.

Stephen Lee, an assistant professor in the School of Computing and Information, said a data center is essentially where computing and Information Technology infrastructure systems are kept.

“These systems refer to the physical machines and wiring that use electricity to process and store digital information, like photos and messages,” Lee said. “Beyond this, data centers consist of servers, networking components, storage and anything that allows you to interact with the computing system.”

Masoud Barati, an assistant professor in the Swanson School of Engineering, said data centers are essentially large-scale factories for data, with computers and storage devices doing the heavy lifting instead of assembly-line machines.

“When you walk into a data center, it often feels like entering a futuristic warehouse of technology,” Barati said. “You would hear a constant whirring noise from hundreds of cooling fans, feel cool air blowing, and overhead, you might see cable trays carrying bundled network cables.” 

According to Barati, networking equipment is an essential component of data centers.

“Data centers contain a maze of network switches and routers, as well as miles of cabling [in overhead trays or under the floor] linking everything together,” Barati said. “This is the component that ensures data can move in and out.” 

Barati said there are usually very few people in the server halls because the operations are monitored remotely or from a control room. 

“The day-to-day operation of a data center is about proactive care and monitoring,” Barati said. “The staff — often working in shifts, since these facilities run continuously — focus on watching for anomalies, conducting regular tests and following strict procedures for any work that needs to be done on the live systems.”

Increased artificial intelligence usage is changing the requirements for data centers in several major ways compared to traditional computing, Barati said.

“AI-focused data centers need to handle greater power density, more heat, specialized hardware and intense data flows between machines,” Barati said. “They are often built with cutting-edge cooling, because traditional air cooling struggles beyond a certain point.” 

Lee said a significant amount of energy goes into cooling infrastructure, which keeps the servers at an optimal temperature so they can function properly.

“These systems produce a lot of heat, so the centers require large amounts of cooling infrastructure to keep the servers cool,” Lee said. “This ensures they don’t melt down and cause heating or power-related issues, which could cause serious problems.” 

According to Lee, there is a substantial difference between small and large — or hyperscale — data centers. Hyperscale data centers currently account for approximately 44% of the worldwide capacity of all data centers and are expected to account for 61% of the total capacity by 2030. The incoming data center in Springdale will be a hyperscale data center, spanning 565,000 square feet. 

“Whereas small data centers have fewer servers, hyperscale data centers are very large,” Lee said. “They can house tens of thousands of servers and a ton of equipment.”

Brad Messner, a clinical assistant professor in the School of Business, said cloud-based data centers, which support the internet and store private corporation data, became more common in the late 90s to early 2000s. 

“Cloud computing means accessing something shared from anywhere, like Gmail or Microsoft OneDrive,” Messner said. “When you go on Amazon or any website, the servers for these sites are typically sitting in a cloud-based data center.” 

Messner said the cloud-based services and data are hosted at multiple data centers, which have various benefits.

“If there is a fault at one location, users immediately get re-routed to a data center somewhere else and things remain functional,” Messner said. 

Messner said there is an important business angle to data center design. They need to remain fast and responsive to stay competitive and therefore must design more efficient systems, which requires more energy. A large data center is capable of using as much electricity as an entire small city, enough to power 40,000 houses.

“The energy required to run some of these systems is astronomical, and this causes strain on the current energy grid,” Messner said. “However, in many cases, data centers are being built and designed with as much energy efficiency as possible, with some even built underground or at a below-ground level to reduce the amount of impact.”