How natural refrigerants provide a practical path to effective data center retrofits

AI is reshaping data center design and fundamentally changing the makeup of modern facilities. Traditional data centers were built around the capabilities, workloads and constraints of a vastly different technological landscape. These legacy data centers utilized air-cooled environments with thermal management systems that were not designed for the heat levels or unpredictable demands of today.

Why AI workloads are challenging traditional data center infrastructure:

• Increased rack densities: In conventional data center environments, rack densities ranged from 5-15kW per rack. However, AI workloads can range from 50kW to 300+ or even 600+ kW per rack with 1MW per rack expected in the coming years.
• Air cooling has limits: Traditional air-cooling systems have certain limits – once workloads pass 20-30kW per rack, conventional systems struggle to keep up.
• Unpredictable loads: At AI Inference Factories – which are data centers that produce the actual output responses to user input prompts – there are also significant and unpredictable spikes in usage resulting in equivalent surges in power consumption and heat production.
• The emergence of liquid cooling: Under these new conditions, liquid cooling has become the primary method of heat extraction. New infrastructure is required across the entire thermal chain from chip-level heat capture to the final rejection to ambient to support liquid cooling strategies.

Retrofitting vs purpose-built AI data center

Because of the specialized cooling, power and structural elements required to run AI, many organizations have concluded that it’s easier to build new facilities than retrofit existing ones. And a huge number of hyperscalers are doing just that – building large-scale AI Factories that are optimized for higher-density workloads.

There are significant benefits of designing purpose-built AI Factories from the ground up and leaning on the many blueprints that have been released by industry leaders. However – for many organizations – building new facilities isn't always the most viable, cost-effective or energy-efficient option.

Retrofit projects: the opportunities and challenges facing colocation providers

The way that a data center is set up often lends itself to retrofitting opportunities. This is especially true for colocation providers who often have access to additional buildings, are set up to take on new customers and can adapt to additional workloads as a matter of routine. For many colocation providers, retrofitting can be a practical alternative to investing in a new purpose-built AI facility.

When it comes to AI readiness however, colocation providers face a unique set of challenges. Many colocation operators actively acquire existing facilities and inherit legacy infrastructure. But this does not stop customers from demanding AI-ready environments.

When a customer wants to run AI in a colocation setting, it can present a challenge as the request triggers a major shift in technical and operational requirements. For example, they might need to migrate from traditional air cooling to liquid or two-phase cooling technologies that can handle higher thermal loads.

As colocation providers evaluate AI readiness, there are three challenges that frequently emerge and should be kept in mind:

1. Speed to market
AI hardware cycles move at an extraordinary pace. The lifecycle of IT equipment has reduced significantly as technological advancements outpace hardware limitations. It could be the case that, by the time one infrastructure project is completed, a newer generation of technology may have already entered the market.

Added to this time pressure, the end users that are running IT equipment in a colocation setting want the infrastructure to adapt as soon as possible so that they can start training and deploying their AI models.

2. Reliability and resilience
AI infrastructure represents a significant investment. Cooling and power systems must provide dependable performance while supporting increasingly demanding workloads. Reliability and resilience are not negotiable.

3. Adaptability and flexibility
Colocation providers want to ensure that they are set up to adapt to what’s to come. This is not about future gazing and predicting the next innovation. Instead, it’s about making sure that the infrastructure is adaptable and flexible to future requirements and advancements.

Why refrigerant selection can play an important role in the long-term viability of retrofit projects

As colocation providers confront these challenges, refrigeration selection can play an important – yet often overlooked – role.

The data center industry is increasingly placed under the microscope as data centers face legal and social pressures around sustainability and long-term operational continuity. This includes minimum efficiency and economization requirements. It also includes refrigerant phase downs such the recently mandated transition away from high Global Warming Potential (GWP) refrigerants to those of no more than 700 GWP.

As the industry moves towards sustainable cooling technologies, the adoption of lower GWP alternatives can play a huge role in achieving operational continuity. This is especially important for those who are considering retrofit projects and want to ensure that further adjustments aren’t immediately required to keep up with evolving regulatory requirements.

As colocation providers evaluate both cooling and retrofitting strategies, refrigerant selection is becoming an increasingly important consideration. Regulations continue to evolve and operators across the industry face uncertainty about what refrigerants will be phased down or phased out in the years to come. Selecting solutions that can remain viable over the long term can help reduce the risk of costly replacement projects down the line.

How natural refrigerant solutions can provide a more practical path to AI readiness

CO2 is a future-ready refrigerant that combines low GWP with long-term regulatory certainty and broad deployment flexibility. It’s a natural option with zero ozone depletion potential and a GWP of 1. By contrast, traditional refrigerants can have GWPs of more than 2,000, though many of these high-GWP refrigerants are being phased out by the United States Environmental Protection Agency (EPA). CO2 can also offer operational and efficiency benefits depending on application requirements and system designs.

M&M Carnot – a subsidiary of Johnson Controls – provides a chiller package that’s ideal for colocation providers who are looking to retrofit existing facilities. There are several reasons why it has become such a suitable and effective option:

• Designed for seamless legacy data center retrofits
• Integrates with existing heat rejection infrastructure
• Supports direct cooling for high-density AI workloads
• Future ready with natural CO2 refrigerant (GWP of 1)
• Patented Rain Cycle™ technology that maximizes free cooling efficiency and extends the number of free cooling hours

Industry conversation often focuses on new builds. However, the reality is that many operators will continue to rely on existing facilities for years to come. The organizations that succeed are those that can adapt quickly and deploy flexible cooling strategies that enable AI workloads while preserving speed, resilience and long-term sustainability.

As AI continues to push data center infrastructure into new territory, retrofit-friendly approaches – like those offered by Johnson Controls and M&M Carnot – may prove to be one of the industry's most valuable tools for unlocking capacity where it already exists.