As demand for AI-ready data centres accelerates, Ryne Friedman, Associate at hi-tequity, explains why retrofitting existing facilities is becoming a serious alternative to greenfield development.
AI’s rapid rise has made speed-to-market a major driver in data centre development. While compute capacity remains a critical constraint, the need for speed is increasingly shaping the choice between greenfield builds and facility retrofits.
The traditional focus on balancing cost and design has shifted, with speed-to-market now taking on greater importance alongside compute capacity. Greenfield builds are no longer the automatic choice for providing scale, as retrofitting is now being given serious consideration. Retrofitting can offer a pragmatic alternative, helping to address friction around land, power and permitting, especially in Tier-1 markets.
The time advantage
One of the key advantages of retrofitting is time. Greenfield developments tend to take years, with large site assembly, protracted permitting timelines, uncertain local regulations and grid delivery lead times making them a longer-term play.
A retrofit project, however, can often be measured in months. Because the project uses an existing industrial building shell and its established pathways, construction time can be reduced, and permitting may be less complex and therefore faster to secure. For players serving the commercial AI expansion, a time advantage can also translate into a contract advantage.
Modernising for AI density
Retrofitting means modernising the mechanical and electrical infrastructure of an existing footprint to handle the significant loads associated with AI. The best facilities to retrofit are often those that were used in the CPU era of computing. These facilities operated at lower densities. To meet AI’s demands, comprehensive modernisation is required.
Power and electrical: Legacy power equipment, such as older UPS systems and batteries, will need to be replaced. Electrical distribution systems are usually undersized and unable to support high-density GPU demand in AI clusters. Retrofitting requires replacing or rebuilding the electrical train, including switchgear, distribution and protection, to support higher-density environments.
Cooling systems: AI hardware produces high heat loads. The cooling systems in older facilities often need to be replaced and modernised to handle higher power consumption and heat loads. Coolant distribution units (CDUs) are key components for properly handling this power and heat. These CDUs must be integrated. Without this equipment, it will be difficult to support the advanced cooling approaches needed for GPU heat loads. With it, the facility can support the high-density ecosystems that AI increasingly requires.
Cost
While retrofitting can reduce property and land costs, it can also increase expenses due to the required modernisation of electrical and mechanical systems. However, the benefit of retrofitting is that it can dramatically shorten the timeline and increase schedule certainty, which is becoming increasingly valuable in the race for AI capacity.
The bifurcating market
The retrofit-versus-greenfield decision is increasingly shaped by market type. For Tier-1 and saturated markets, new construction can pose significant challenges. There is a scarcity of land and often friction around entitlements to contend with.
Retrofitting uses existing power infrastructure footprints rather than procuring new power delivery infrastructure. Because these are dense, high-demand areas, retrofitting is often the fastest, and therefore one of the most desirable, ways to get capacity online. Operators in these markets need speed and certainty.
Greenfield can still be a viable option, but it is becoming clear that this is not always the case. For Tier-2 or expansion markets, it could very well be the best option, as land, permitting and interconnection paths are more easily achievable in these scenarios. Greenfield cannot be eliminated as an option. It is a longer-horizon play that can be optimal for building custom AI clusters. In this case, a multi-year timeline may be permissible given the scale.
Execution will decide the winners
Being a realist in today’s AI market means recognising that speed is a major factor in data centre success. The choice between retrofitting and a greenfield build depends on the project’s timeline. High-level coordination and stakeholder alignment are essential. Utility standards, lead times and approvals can be hurdles that require tight coordination.
By sourcing owner-furnished, contractor-installed (OFCI) equipment, such as transformers and breakers, and coordinating procurement to mitigate delays, developers can improve schedule control around high-voltage equipment. Relationships, good partners and clear communication can all help reduce schedule risk, especially if utilities are already backed up. Power is in such demand that some customers are building their own substations to accelerate timelines.
The AI boom has forced the data centre industry to recognise that the fastest path to compute capacity may also be the most valuable. Increasingly, the choice between greenfield and retrofit comes down to which path can deliver usable capacity on the timeline the market demands.
