Answering the perceived trade-off that keeps data center operators up at night

Key Takeaways

• Standardization and flexibility are not opposites—they are enablers of each other
• Modular infrastructure can deliver both: repeatable components that can be adapted to changing facility needs
• AI workloads demand the predictability of standardization and the agility of modular design

Do standard designs or standardization limit innovation?

Answer: No. Standardization reduces operational risk and cost, freeing up resources and capacity for innovation.

Data center teams often believe it is a trade-off: standardization reduces cost and operational risk, while flexibility accommodates evolving workloads like AI. In reality, modern infrastructure delivers both. When every data center uses the same rack, the same busway, and the same cooling platform, innovation shifts from "keeping the lights on" to "optimizing for AI."

That said, standardization does not equate to a completely "one size fits all" setup. It means repeatable building blocks that include racks, busways, cooling units, and power distribution systems engineered to work together. When every component follows the same design rules, teams can:

• Deploy faster – No site-by-site re-engineering
• Train less – One set of procedures works everywhere
• Stock fewer spares – Common parts across all facilities
• Reduce errors – Familiar designs mean fewer surprises and more logistical clarity

This also means that when it comes to infrastructure management, teams can readily automate provisioning across all sites, predict capacity needs from real usage data, test new cooling strategies on a repeatable baseline or even deploy agentic AI with consistent telemetry everywhere.

Are there organizations already implementing standardized builds for their data centers?

Answer: Yes, industry leaders like Google, Meta and Oracle have implemented or spoken in support of standardization in data center infrastructure, and its benefits.

The interest these major companies have in standardization is proof that it enables instead of constraining flexibility. Equinix's Flexible Data Center (FDC) design is a catalogue of repeatable power, cooling, and rack elements, allowing the company to execute nearly 60 concurrent builds across 34 markets with consistent quality and rapid deployment.

Similarly, the Open Compute Project (OCP) has launched the "Open Systems for AI" strategic initiative to develop standardizations for power, cooling, mechanical infrastructure, and management telemetry. This initiative enables AI infrastructure to be deployed as flexibly as traditional compute, with the goal of establishing common physical infrastructure standards that make data centers fungible (interchangeable or replaceable), built with modular, interoperable components that can accommodate rapidly evolving AI hardware without requiring complete overhauls.

Notable contributions include Project Mount Diablo (a 400V disaggregated power architecture co-authored by Google, Meta, and Microsoft that supports racks from 100kW to 1MW), and Project Deschutes (Google's advanced liquid cooling specification for up to 2MW heat loads). As Oracle's SVP and chief technical architect Mr Pradeep Vincent stated, open standards are "a huge enabler for the scale and speed at which we need to build AI infrastructure", enabling operators to source from multiple vendors and manufacturing pipelines.

By standardizing interfaces and modular architectures, OCP prevents supply chain fragmentation, accelerates deployment timelines, and ensures data centers can adapt to the rapid pace of AI hardware innovation without being rebuilt from scratch for each generation. 

A Quick Look at Some OCP Projects

So, standardization of data center infrastructure does not sacrifice flexibility and innovation; it creates a more conducive environment for both.

How can infrastructure remain flexible?

Answer: Through modular components that can be added, removed, or reconfigured without redesigning the entire facility.

Flexibility in a standardized environment comes from modularity. Consider these examples:

This is not theoretical. Starline’s Track Busway, for example, allows operators to start with moderate capacity and scale to 1,250 amps per tap-off without replacing the underlying infrastructure. Likewise, the rear door heat exchangers by USystems, a brand of Legrand, have a small footprint, no requirements for specialist infrastructure, and sizes to suit racks between 42U to 52U, enabling facilities to simply add more units as needed with increasing cooling needs.  

What does modularity really mean?

Answer: Modularity means designing systems where components are independent, interchangeable, and upgradable without affecting the overall setup.

True modularity has three characteristics:

• Hot-swappable – Replace a component without shutting down the system
• Scalable – Grow capacity by adding modules, not replacing the entire unit
• Standardized interfaces – A power tap-off from one vendor fits the same track as another

For AI data centers, modularity is a necessity, as workloads that could double in intervals as short as  seven months cannot afford to have infrastructure that requires a forklift upgrade each time.

Legrand's Keor FLEX UPS exemplifies this: internal scaling to 1.2 MW, with hot-swap power modules and dual communication buses. As rack densities ramp up, operators can house up to 12 modules, each with 100kW capacity, per UPS and reach a maximum power rating of 4.8MW by configuring up to 4 systems in parallel, without the need to replace the backup power system.

The Bottom Line

For APAC operators planning AI-ready facilities, choosing standardization is not a sacrifice of flexibility; it’s about choosing standard components that are modular to make room for adaptation and innovation.