Why AI Data Centres Should Think More Like Formula One Teams

Updated October 2026: This article has been updated to reflect the latest test results for Conflux single-phase cold plates. The original version cited 0.0076°C/W at 6 LPM across a 2,500 mm² footprint (tested as of August 2026).

As AI infrastructure scales, most conversations around data centre performance still focus on compute capacity: faster GPUs, denser racks and increasing power availability.

However, a critical constraint is quietly emerging alongside these advances: cooling efficiency.

While the industry has largely acknowledged that liquid cooling is necessary to support next-generation hardware, less attention has been paid to how efficiently those systems operate at scale.

That distinction, between simply cooling hardware and engineering an efficient thermal system, is one Formula One has been refining for decades.

Engineering That Never Stands Still

Perhaps Formula One’s greatest lesson isn’t simply how to engineer an efficient cooling system, but how to engineer one that can evolve.

Every race weekend brings new components, revised aerodynamic packages and evolving thermal demands, leaving engineers with little time to redesign, test and validate solutions.

That relentless cycle has fostered an engineering mindset built around rapid iteration, where cooling systems are expected to evolve alongside the hardware they support.

AI data centres are now entering a similar phase.

Successive generations of GPUs are arriving faster than ever, each with different thermal profiles, power densities and packaging constraints, meaning cooling solutions can no longer remain static while the technology around them advances.

From Thermal Capacity to System Efficiency

That evolution also changes the engineering challenge. It is no longer enough for cooling systems to remove the required amount of heat. They must do so with maximum efficiency.

AI workloads are pushing rack power densities to unprecedented levels, and simply meeting thermal requirements does not guarantee an optimal system. Instead, operators must consider how much energy is required to deliver that cooling capacity.

This shifts the focus from peak performance to system efficiency, a change that has significant implications for both operational cost and infrastructure design.

The Hidden Cost of Pressure Drop

One of the most important, and often overlooked, aspects of cooling performance is pressure drop.

In liquid cooling systems, pressure drop has a significant influence on the energy required to move coolant through the system. While small, perhaps even negligible, at the component level, its impact becomes significant when multiplied across thousands of servers.

Inefficient flow paths can increase pump power requirements, raise energy consumption and reduce overall system flexibility. In extreme cases, they can become a limiting factor in scaling infrastructure.

Why Component-Level Thinking is No Longer Enough

Components such as コールドプレート are often assessed in isolation, focusing on metrics like heat transfer performance and flow rate.

However, as systems become more complex, this approach can be misleading. Formula One engineers have long understood that a component delivering impressive standalone performance can still compromise the performance of the car as a whole.

A component that delivers strong thermal performance but induces high hydraulic resistance may degrade overall system efficiency.

The net result is higher energy use, even if individual components appear optimised.

To address this, cooling must be treated as a system-level engineering challenge, where thermal and hydraulic performance are considered together rather than separately.

Balancing Competing Requirements

Effective liquid cooling requires managing a fundamental trade-off between improving heat transfer and maintaining low resistance to flow.

Design approaches that increase turbulence or surface area can enhance thermal performance, but they often do so at the cost of higher pressure drop. The challenge lies in achieving both low thermal resistance and low hydraulic resistance at the same time.

This balance demands precision engineering at both macro and micro levels. The most effective solutions focus on directing coolant exactly where it is needed, rather than applying uniform cooling across an entire component.

At Conflux, for example, our latest single-phase cold plate designs achieve thermal resistance as low as 0.006°C/W* at 6 LPM across a 2,500 mm² footprint, while maintaining exceptionally low pressure drop, demonstrating that high thermal performance and hydraulic efficiency can be achieved together, rather than traded off against one another. *Tested as of October 2026.

It’s the same engineering philosophy that underpins Formula One: the best solution isn’t the one that excels in a single metric, but the one that delivers the greatest overall system performance.

The Formula One Engineering Mindset

One of Formula One’s defining characteristics is that engineers do not evaluate components in isolation.

Every design decision is based on its impact on overall system performance. A component that performs exceptionally well in isolation can still reduce the overall performance of the car if it compromises weight, reliability or aerodynamic efficiency.

This principle is directly relevant to data centres, where cooling systems must balance competing objectives under tight constraints.

This is why cooling design is increasingly moving beyond component optimisation towards system-level engineering, where every design decision is judged by its impact on overall performance.

Delivering that level of optimisation demands equally sophisticated design tools and manufacturing techniques.

The Role of Advanced Design and Manufacturing

Achieving high levels of efficiency in liquid cooling begins with how fluid flow is controlled within components.

Modern engineering approaches use simulation tools such as computational fluid dynamics to analyse and refine internal geometries. These tools make it possible to optimise coolant distribution, minimise losses and improve heat transfer where it matters most.

It’s an approach familiar to Formula One, where advanced simulation and rapid manufacturing allow engineers to continually refine components in pursuit of marginal gains. In liquid cooling, those same principles are enabling a new generation of highly optimised cold plate designs.

Additive manufacturing makes this possible by enabling complex internal channels and geometries that would be difficult or impossible to produce using conventional manufacturing methods. This gives engineers far greater control over coolant flow, allowing thermal performance and pressure drop to be optimised together rather than traded against one another.

これにより、コールドプレートを複数の部品の組み立てではなく、単一のモノリス部品として製造することも可能になります。接着界面をなくすことで、潜在的な漏れ経路が排除されると同時に構造的整合性が向上します。これは、信頼性が性能と同等に重要視されるミッションクリティカルなデータセンター環境において、大きな利点となります。.

冷却の役割の再考

フォーミュラワンは常にわずかな改善の積み重ねが意味のある競争優位性をもたらす、限られた利得のスポーツであった。AIデータセンターも同様の時代に突入しつつある。.

プロセッサーの性能が向上し続けるにつれて、もはや問題は冷却システムが十分な熱を除去できるかどうかだけではない。システム全体を通して、熱性能、圧力損失、エネルギー消費、信頼性のバランスを保ちながら、最大の効率でそれを実行できるかどうかである。.

その結果、データセンターイノベーションの次の段階は、より高速なプロセッサによってだけでなく、それを冷却するためのよりスマートで効率的な方法によって定義される可能性がある。.