Data Centre Titans: Designing for Power Densities That Keep Changing – Gordon Johnson, Subzero

As power densities increase, cooling is becoming an increasingly important consideration in how data centres are designed and operated.

In this Data Centre Titans interview, Gordon Johnson, Senior CFD Manager at Subzero Engineering, discusses airflow management, liquid cooling and the practical challenges associated with supporting higher-density workloads.

– – – – – –

What is your role today at Subzero Engineering, and how did your journey into the data centre industry begin?

As Senior CFD Manager, I manage all of our computational fluid dynamics (CFD) modelling projects across the US, EMEA, and APAC regions, as well as deliver our internal and external training on data centre design, containment, cooling strategies, and CFD best practices.

My role focuses on improving efficiency and operational performance in legacy and next-generation data centres. In addition, I work closely with our global sales and engineering teams to ensure customers receive accurate, high‑performance containment solutions that are tailored to their facilities.

Also, I regularly speak at industry events on data centre cooling, containment, and emerging trends shaping the future of our industry.

You’ve spent more than 30 years working across data centres and cleanroom environments. What changes have had the biggest impact on how facilities are designed and operated today?

When I initially entered the industry, efficiency and energy savings were largely seen as an afterthought in data centre management and operation. As recently as 10-15 years ago, the prevailing mindset for data centre cooling was simply ‘the colder, the better’. Beyond driving massive energy bills, this approach wasted natural resources and expanded our carbon footprint.

Fortunately, the industry has since changed its perspective, with many facilities now operating supply temperatures near the upper end of ASHRAE’s recommended server‑inlet guidelines, ensuring efficiency and a closer connection to the industry’s ambition for a more sustainable future.

Looking back across your career, what have been some of the most important lessons you’ve learned about designing resilient infrastructure?

Looking back on my career in the data centre industry, one of the most important lessons I’ve learned is that simplicity is often what drives reliability and performance.

Data centres constantly evolve, and the demand for high-performance computing densities is growing, so those in the industry must design their facilities to adapt to changes without requiring complete redesigns.

Containment is a perfect example of simplicity at play, which is why I’m so passionate about the topic. Rather than consuming energy, effective containment reduces energy use and makes every other efficiency initiative more impactful and cost‑effective. It reinforces a principle I’ve come to fully embrace – the best energy saved is the energy you never consume in the first place.

As AI workloads push rack densities higher, how is the industry’s approach to cooling and airflow evolving?

As recently as 2020, a 20kW rack qualified as HPC (High Performance Computing). Today, AI and machine learning workloads routinely drive rack densities to 100kW and beyond, with demand continuing to climb. This rapid escalation is a key reason hybrid cooling, combining air and liquid systems to handle both low‑ and high‑density racks, has become the industry’s evolving standard for efficient cooling regardless of the workloads deployed.

The industry is recognising that future-ready data centres must combine liquid cooling with highly disciplined airflow management systems to support the evolving needs of the competitive market.

There’s a growing conversation around liquid cooling, immersion, and direct-to-chip technologies. How do you see these approaches fitting together within future data centre environments?

Future data centre environments must consider using a hybridised approach in their cooling technologies, so I expect to see this in next-generation environments. It will be common to see direct‑to‑chip, immersion, and air-cooling solutions operating side by side within the same facility, depending on the density requirements, operational priorities, and sustainability goals.

What is also important to note is seeing a more deliberate separation between high‑density and low‑density ITE, which is ultimately how operators will maximise efficiency.

What do you think operators still misunderstand about airflow management and thermal efficiency in modern facilities?

Some people don’t realise that the most common form of liquid cooling—direct‑to‑chip (DTC) — still relies heavily on air cooling strategies. DTC cold plates remove heat only from the highest‑load components, such as CPUs and GPUs. The remaining server and rack heat must still be removed by traditional air systems.

Since roughly 75% of the heat is captured by the liquid loop, a 100 kW rack still requires enough airflow to remove the remaining ~25 kW. To do this effectively, proper cold‑aisle or hot‑aisle containment is also essential.

At what point does cooling become a strategic infrastructure issue rather than simply an operational one?

Cooling has now become a strategic infrastructure consideration, directly impacting scalability, operational cost, sustainability ambitions, and the competitive market advantage. With AI and HPC, cooling is now shaping the entire data centre infrastructure, including power, layout, sustainability, and total cost of ownership, moving away from being seen as an operational issue, the moment it limits what can be deployed.

We’ve hit the point where cooling is no longer something we operate, but rather it’s something we must strategically design for the future industry.

How can operators balance rising compute demand with growing pressure around sustainability, energy efficiency, and resource use?

The key to balancing rising computing demand with growing pressures is to first pause and take a breath – not every rack moving forward will be running AI or ML workloads. At the same time, we must design and engineer next‑generation data centres to be sustainable, energy‑efficient, and relevant for the next ten to 15 years. Constant three‑year rip‑and‑replace cycles of both hardware and cooling aren’t viable. We need a balanced approach that prioritises smart, long‑term TCO decisions.

What role do modelling and simulation technologies like CFD now play in improving operational efficiency and long-term planning?

CFD has evolved from being a design validation tool into a strategic operational tool. It is still very much needed, even in data centres that deploy direct-to-chip liquid cooling. In many cases, this has become more important, not less.

For example, CFD is needed to validate that the airflow is still delivered correctly and that no hotspots form around non-liquid-cooled components, since CFD is the only tool that can accurately model how that remaining heat behaves in the white space. CFD is also the only way to simulate transient and failure scenarios safely.

As infrastructure becomes more complex, CFD modelling becomes more important as operators can no longer rely on previously used design approaches alone.

Looking ahead five years, what’s one bold prediction you would make about the future of cooling and thermal management in the data centre industry?

Data centre cooling and thermal management will continue to evolve over the next five years and well beyond. We expect ongoing adoption of hybrid cooling, through integrating air and liquid systems, to increase capacity, efficiency, and sustainability.

Moving forward, data centre design and cooling infrastructure will no longer follow a one‑size‑fits‑all model, so all of us in the industry need to evolve and adapt to this way of thinking.