Understanding just how the ground has shifted for data center construction it is important to understanding where AI is in its evolution. As microchip hardware continues to advance, the demand for power and thermal cooling continues to increase. These demands directly impact the design, building and testing of the facilities and infrastructure that are being built to support these advancing technologies. Strategies for smoothing power demand spikes, designing in-space flexibility to support alternative cooling technologies and integration of maintenance and reliability are all important in both new and retrofit of existing data centers to support the changing AI technology.

How AI Data Centers Compare to Traditional Data Centers

The differences in how AI data centers are designed and built starts with their primary function. Traditionally, many of the data centers built from the 1980s thorough about 2020 were designed for supercomputers, online access/presence along with data storage and on-demand recall. First, several were built by financial firms and other companies that needed to conduct accurate, secure, timely transactions or services. (Think credit card processing, airline ticketing or transmitting health care data.) Many were set up as on-premises solutions with racks filled with servers, software and switch equipment.

As technology progressed, colocation, cloud computing and Infrastructure as a Service changed the model for data centers. The need for secure access environments, connectivity reliability and larger-scale data centers were drivers. Social media platforms also contributed to the data center demands as they required additional data storage and recall (saving and sharing all of your photos and videos.)

AI, on the other hand, is about real-time processing. AI data centers are filled with power-hungry, heat-generating microprocessors. Much of the physical space inside a data center is designated for the electrical switches, power supplies and cooling equipment. They can consume as much power as a town, and processing demand is exploding with the new AI tools being rolled out by major tech companies – directly impacting where, how, and how fast they are built, is fundamentally different.

Six Ways AI Data Center Construction Is Different

Going a step deeper, here are six changes to data center construction that have emerged or accelerated with the AI boom, and how we and other construction firms are adapting to address them:

• The need for speed: The AI boom means firms need availability as quickly as possible to deliver on their product promises. Construction companies need to be dynamic and responsive to find ways to compress construction schedules, and quickly pivot when barriers (like labor or material shortages – more on that later) arise.
• Follow the power (or build your own): A new “land/power rush” is happening, with developers looking for land that is close to existing or in development power generation and transmission. They need construction partners equipped to work where the land/power is, often deep in rural areas. Developers are currently considering building their own generation plants, which can be a complicated process with lots of regulatory hurdles. Partnering with a construction company that has both data center and power generation plant experience can be a plus.
• Energy density is skyrocketing: Energy density in data centers has tripled in a few years, up to 100+ kW per rack, and that trend is still increasing. Electrical and mechanical engineers who can design to those densities are critical to project success. Plus the high demand for critical mechanical and electrical equipment has continued to impact supply chain delays.
• Liquid is the new air: Higher energy densities mean more heat, and only liquid cooling can handle these high levels of energy density. Construction teams need experienced engineers who can design liquid cooling systems such as direct-to-chip or immersion and have expert knowledge in the equipment and materials to support those systems.
• Uptime requirements are evolving: For some AI data center implementations, the requirement and need for high availability (five nines) may be different. Eliminating or deferring the high cost of full load continuous generational backup is minimized to battery support which allows for structured shutdowns of processes and equipment until the utility power can be restored. Provisions for electrical generation may be provided, to allow for future installation of equipment as that need or equipment becomes available.
• Plan ahead on procurement and be ready to pivot: With increasing demand for electrical and cooling equipment, securing those materials early is critical. Working with procurement teams in partnership with the designers and engineers is paramount to maintaining project delivery and installation schedules. Once the equipment has been ordered, management of the build and testing timelines are critical for the team to confirm quality and tracking for delivery to the project site.
• Prefab is your friend: With the labor and materials crunch, site construction happening in remote locations, and speed-to-market demands on the building construction, leads to continued push for off-site prefabrication. Often building large segments of the systems in warehouses and trucking them to the worksite is the only way to meet the project deadlines.

No Crystal Ball – So Find a Good Partner Who Can Evolve with You

We are still in the early days of AI and learning just what it is capable of – and what infrastructure will be needed to support it. Nobody knows how high energy density will rise, or how the electrical grid will evolve over the next several years.

In the absence of certainty, the best bet for tech firms and data center developers is to be willing to be flexible and find a construction partner who has not only kept up with the recent changes but has proven they have the strength and flexibility to evolve into the future. The story of AI is still being written – and so is the way to build for it.