AI Is Breaking the One-Size Data Center

For years, data centers were built around a simple rule: maximize availability. The default target was often 99.999% uptime, with layers of redundancy across power, cooling, and networking to keep services running even when parts failed.

That model is now under pressure. In this TechRadar Pro Perspectives piece, the CTO and Co-Founder at InfraPartners argues that AI workloads have exposed how different infrastructure needs can be. Training a large language model, serving real-time inference, running enterprise software, and handling business-critical transactions do not demand the same levels of resilience, latency, or geographic distribution.

How AI training and inference change data center design

The article says operators are increasingly separating AI training from AI inference when they design facilities. For training, the main constraints are often:

• energy supply
• cooling capacity
• speed of deployment
• compute density

In some cases, that means building training sites without backup generators, complex redundancy systems, or high-tier architecture. These facilities are increasingly being placed wherever power is available, with less emphasis on traditional resilience standards.

Recommended reading

UK data centre boom hits a water reality check

Inference is different. Because those workloads often sit closer to users and support services people interact with daily, latency, availability, and customer experience matter more. That creates a stronger case for resilient infrastructure and geographically distributed architectures.

Precision resilience and upgradable facilities

Rather than applying the same design standard everywhere, the piece argues for “precision resilience” — matching redundancy to how workloads actually behave instead of relying on legacy assumptions.

That matters because operators are dealing with labor shortages, demand that is outpacing supply, pressure to deliver capacity faster, and an ongoing power gap. Overengineering every AI deployment adds cost and complexity at a time when both are in short supply. Every extra redundancy layer consumes capital and can slow projects down.

The proposed answer is flexibility. As AI demand shifts and future compute densities and cooling requirements remain uncertain, facilities need to be upgradable. The article points to off-site building blocks assembled later on site as one way to create data centers that can evolve over time.

According to the piece, the result will be a mix of specialized sites: energy-optimized training campuses near power sources, distributed inference facilities where uptime and latency directly affect user experience, and hybrid environments that support both AI and traditional workloads.