Why Are Data Centres Getting So Heavy?
In data centres, weight is becoming a defining factor in design, construction, and location strategy for companies across the globe. As artificial intelligence (AI), machine learning, and high-performance computing (HPC) drive demand for denser, more powerful infrastructure, the equipment inside data centres is growing significantly heavier.
The result? Floors are being reinforced. Raised platforms are being reconsidered. Single story is a must. And some buildings simply can’t take the strain.
This article explores the physical weight of modern AI data centres—what’s contributing to it, why it’s increasing, and how these changes are forcing the industry to rethink everything from structural engineering to site selection.
Contents
- What makes a data centre heavy?
- Why data centres are getting heavier
- Designing data centres for AI infrastructure
- Site selection challenges for AI data centres
What makes a data centre heavy?
Several core components contribute to the physical weight of a modern data centre – and with the rise of AI and machine learning, many of these are getting heavier.
- Server racks: At the heart of the data centre, racks can weigh close to two tonnes each.
- Cooling systems: Liquid cooling infrastructure, including chillers, pumps, and pipework, adds significant weight both on and beneath the floor of a data centre.
- Power Distribution Units (PDUs): These systems regulate and distribute electricity across rows of racks, with heavy-duty models often floor-mounted.
- Backup power units: Large battery banks and UPS systems are essential for redundancy, and contribute to the overall weight of a data centre.
- Switchgear and transformers: These industrial-grade components manage power conversion and safety, often housed in separate heavy enclosures.
- Containment systems: Hot and cold aisle containment are often made from metal and reinforced plastic or glass.
Why data centres are getting heavier
The growing weight of equipment inside data centres isn’t incidental—it’s a direct result of the (pretty recent) technological shift toward AI, high-performance computing, and onsite power solutions. Every advancement in capability comes with added physical load.
AI and high-density hardware
AI infrastructure is the primary driver behind today’s heavier data centre racks. Compared to traditional IT equipment, AI hardware is significantly denser, packing more compute power—and hence more physical material—into the same footprint.
A standard 42U server rack with conventional compute, networking, and storage typically weighs between 680kg to 1100kg, but AI racks can weigh up to two tonnes.
Density and cooling
Greater compute density means more performance per square foot, but it also means significantly more heat. In response, data centres are adopting liquid cooling and direct-to-chip solutions, which add additional weight through plumbing, coolant, and support infrastructure.
Onsite power and battery storage
Beyond compute, facilities are increasingly incorporating onsite energy systems like lithium-ion battery storage and backup generation. These installations add considerable structural load and must be factored into overall weight distribution across the facility.
Designing data centres for AI infrastructure
As the weight of AI and high-density equipment continues to rise, the design of data centres are changing to accommodate the new physical limits of modern infrastructure.
Flooring
One of the most immediate design implications in AI data centres is the decline of traditional raised floors. Reinforced raised floors are increasingly viewed as cost-inefficient, and instead, many operators are opting to build directly on slab foundations to better handle the weight of AI server racks.
However, raised floors haven’t disappeared entirely. Some facilities still use them, but in a different way. They’re lower and primarily used to house water pipes and cabling.
Single story vs. multi-story
In addition to rethinking what’s beneath the racks, data centre designers are rethinking what’s above them, by eliminating upper stories altogether.
The cost of designing multi-story facilities to accommodate the weight of AI equipment is proving to be a major deterrent; it’s simply too expensive to build that second story. This means that sprawling, single-level layouts are rising in popularity because weight can be more easily managed and evenly distributed.
Project timelines
While adapting to heavier infrastructure does require changes in materials, such as stronger concrete and larger steel beams, evidence shows that these kinds of modifications are not really affecting construction timelines.
What is changing, however, is the wider approach to long-term planning. While data centres have historically been built with a 10–20-year lifespan in mind, the speed of AI advancement means that there is a bigger focus on data centres being future-proofed as they could become obsolete much quicker than that.
Read more: The Data Centre Life Cycle
Strategic layouts
Some operators are adopting different layout models, distributing heavier racks across the facility to avoid overloading any single area. Others are implementing pod-based or modular designs, where high-density, high-weight AI equipment is isolated in specially designed rooms. This helps limit structural stress and provides more flexibility for future upgrades.
Site selection challenges for AI data centres
The rise of high-density AI infrastructure is reshaping not only how data centres are designed—but also where they can be located. Traditional assumptions around site feasibility are being challenged as weight restrictions grow more demanding in addition to power and structural requirements.
Read more: What Goes Into Choosing a Data Centre Location?
Retrofitting
These pressures are especially evident in retrofits of existing buildings. While repurposing older structures has long been a strategy to save time and cost, the physical requirements of AI server racks are making many retrofits difficult… if not impossible. Height restrictions are as much of a challenge as weight restrictions.
Before any equipment is installed, building teams must conduct rigorous structural assessments to determine whether the floor loading can safely accommodate modern, ultra-dense racks. Older multi-story buildings, which were once seen as ideal for retrofits, often fall short.
Brownfield sites
Despite these challenges, retrofitting isn’t off the table entirely. In fact, there’s growing interest in brownfield sites – previously developed land that may include disused industrial buildings, shopping centres, or hospitals. These sites can offer major advantages, particularly when it comes to power infrastructure and weight restrictions. Brownfield sites like former factories often come with heavy-duty flooring and existing power connections—saving developers both time and money.
In contrast, greenfield sites – undeveloped plots of land – require extensive infrastructure buildout before construction can even begin, particularly when it comes to power access.
As AI continues to reshape the design standards of data centres, it’s also reshaping their real estate strategy. Location decisions now hinge not just on space and cost, but on the physical readiness of a site to support the next generation of hardware.
As AI and high-performance computing redefine what goes into a data centre, they’re also reshaping the physical realities of building one. From heavier racks and cooling systems to onsite power infrastructure, today’s facilities carry more weight—literally—than ever before. This shift is forcing developers, engineers, and operators to rethink traditional approaches to design, construction, and location strategy. As the industry moves forward, structural considerations will be just as critical as compute capacity in shaping the data centres of the future.
Sources
