How AI Workloads Are Reshaping Edge Data Center Demand

The short answer: edge data centers are getting denser, hotter, and more distributed

AI workloads are no longer confined to hyperscale campuses. Inference, computer vision, and real-time analytics are moving closer to the user, and that's driving a fundamental shift in edge data center demand. Instead of a few large facilities, we're seeing a growing need for many smaller sites—each capable of handling high-density racks, often 30–50 kW per rack or more, compared to the 10–15 kW typical of traditional edge deployments. This isn't just a trend; it's a response to latency requirements and bandwidth costs that make centralized processing impractical for many AI applications.

Consider a factory using AI for defect detection. The cameras generate gigabytes per minute, and sending that to a cloud region adds seconds of delay. That's too slow. The answer is an on-prem edge node with GPU accelerators, but that node draws far more power and throws off more heat than a standard IT closet. The industry update here is that edge data center design is being rethought around AI workloads—from power distribution to cooling to physical footprint.

What's driving the shift: latency, bandwidth, and data gravity

Latency is the obvious driver. Autonomous vehicles, AR/VR, and remote surgery need single-digit millisecond response times. But there's also the cost of moving data. The IEA estimates data centers consumed around 415 TWh in 2024, and a big chunk of that is networking equipment. Keeping data near the source reduces backbone traffic and can cut operational expenses. Then there's data sovereignty—many countries require that certain data stay within their borders, which pushes processing to local edge sites.

AI inference is the main workload at the edge. Training happens in hyperscale clouds, but inference—running the trained model—often needs to be close to the user. That's why we're seeing edge data centers pop up in places like manufacturing plants, hospitals, and retail distribution centers. They're not massive, but they're dense. A typical edge AI node might have 4–8 GPU servers, each drawing 2–3 kW, plus storage and networking. That adds up quickly.

Power and cooling: the real engineering challenge

The biggest headache for edge deployments is power. A 50 kW rack isn't unusual for AI inference, and that's per rack. Multiply that by 10 racks and you're at 500 kW in a space that used to house 100 kW. That changes everything upstream: the utility feed, the UPS, the switchgear, and the cooling system. You can't just swap in a bigger AC unit. You need to plan for the heat density.

Cooling is where most projects get tripped up. Traditional CRAC units pushing 18°C air under a raised floor won't cut it. For 30 kW+ racks, you're looking at liquid cooling—either direct-to-chip or rear-door heat exchangers. That's a big shift for edge sites, which often lack the plumbing for chilled water. Some operators are turning to self-contained cooling units that use refrigerant or pumped two-phase systems. The key is to match the cooling method to the actual density, not to a generic rule of thumb.

Redundancy is another issue. Edge sites are often unmanned, so you need remote monitoring and automatic failover. But you also need to balance cost. A 2N UPS for a 500 kW edge site is expensive. Many operators are going with N+1 or even N, relying on the cloud for backup. That's a trade-off between availability and budget, and it's a conversation we have with every client.

What this means for edge data center design

If you're planning an edge data center for AI workloads, here are the numbers to think about. Power density: plan for 30–50 kW per rack, with a path to 80 kW. Cooling: liquid cooling is no longer optional for high-density racks. PUE: aim for 1.3 or lower, but don't sacrifice reliability for a pretty number. Space: a 10-rack edge site might need 100–150 m², including cooling and electrical rooms. And don't forget network: you'll need multiple fiber paths and low-latency connectivity to the nearest cloud region.

The industry is still figuring out the best form factor. Some are using prefabricated modular units that ship as a complete pod. Others are retrofitting existing telecom shelters. The common thread is that AI workloads demand more power and cooling per square meter than anything we've seen before. That's why VERHI has been focusing on integrated power and cooling solutions for edge sites—UPS systems that can handle high inrush currents from GPU servers, and precision cooling units that can maintain stable temperatures even when the IT load swings wildly.

A quick look at the market forces

The memory shortage is complicating things. DRAM and NAND prices have spiked because manufacturers are prioritizing AI accelerators, so building an edge site today costs more than it did a year ago. That's pushing some operators to delay projects or use older GPUs. But the demand isn't going away. The IEA projects data center electricity use could double by 2030, and a lot of that growth will be at the edge.

Local opposition is another factor. Data Center Watch found that $64 billion in projects were blocked or delayed between May 2024 and March 2025. That's mostly hyperscale, but edge sites face similar zoning and permitting issues. The good news is that edge data centers are smaller and can often be tucked into existing buildings, which reduces the footprint and the pushback.

Practical steps for your next edge AI deployment

1. Start with a power audit: measure the actual draw of your AI gear, not the nameplate rating. GPU servers often draw 80% of peak during inference.
2. Choose a cooling architecture early. If you're going above 20 kW per rack, talk to a cooling engineer before you pour the slab.
3. Plan for growth. A site that runs 30 kW per rack today might need 50 kW in two years. Make sure your UPS and breaker panels have headroom.
4. Think about remote management. Edge sites are often unattended, so you need sensors for temperature, humidity, and water leaks, plus a way to shut down non-critical loads.
5. Work with a partner who understands both power and cooling. That's where VERHI comes in—we provide precision air conditioning and UPS systems designed for edge environments.

The bottom line

AI workloads are reshaping edge data center demand, and the change is happening faster than most people expected. The days of a simple 10 kW rack in a closet are over. If you're building or upgrading an edge site, focus on power density, cooling, and scalability. And don't wait too long—the industry is moving, and the cost of retrofitting later is always higher than building it right the first time.