Michael Fuller, Executive Chairman of Conflux Technology and former F1 engineer, explores how motorsport’s approach to thermal engineering is helping data centres improve cooling performance while lowering energy use and operating costs.
Formula One has always been defined by relentless engineering development. Between race events, engineers redesign components, refine simulations and optimise every system in pursuit of performance gains. Every race demands a better solution than the last and the design is never truly finished.
Today, a remarkably similar engineering rhythm is emerging in AI infrastructure.
Successive generations of GPUs are arriving at unprecedented pace, each bringing higher power densities, new thermal requirements and different physical footprints. Server architectures continue to evolve, rack configurations are becoming increasingly specialised and operators are under pressure to deploy new hardware quickly while maintaining reliability, controlling energy consumption and managing operating costs.
Although the applications are very different, both industries face the same fundamental engineering challenge: extracting ever more heat from increasingly compact systems.
After years solving thermal management challenges in Formula One, I founded Conflux to apply that expertise across automotive, aerospace and other high-performance industries. Today, I see many of the same challenges emerging in the design of next-generation data centres.
Cooling technology is evolving alongside processors
Liquid cooling is no longer viewed as a future technology for AI infrastructure. As power densities rise, the industry has recognised that air cooling alone cannot support the next generation of high-performance computing.
Although removing more heat, more efficiently is patently crucial, the challenge goes beyond that with the pace of development and the rapid iteration required to keep up.
Every new generation of GPUs introduces different thermal profiles, power densities and packaging constraints. Cooling systems must therefore evolve alongside the hardware they support.
Cold plates, once considered relatively standard components, are rapidly becoming highly engineered thermal devices designed around specific processors, server architectures and operating conditions. Their performance depends not only on coolant selection or flow rate, but on how effectively internal geometries distribute coolant precisely to where heat is generated.
Like Formula One, AI infrastructure is adopting a cycle of relentless engineering development, where thermal systems are continuously redesigned and optimised to evolve alongside each new generation of technology.
The overlooked cost of pressure drop
When operators assess liquid-cooling performance, the first question is often simple: can the system remove enough heat? While this is an excellent place to start, it is not the only question we should be asking. A more revealing question is how much energy the system must use to deliver that cooling.
That is where pressure drop becomes commercially important. As data-centre scale, marginal hydraulic inefficiencies can compound across thousands of servers, quietly adding pump power, energy use and operating cost.
This is where liquid cooling performance becomes a systems-level issue rather than a component-level specification. A cold plate that achieves excellent heat removal but creates excessive hydraulic resistance may simply shift the energy burden elsewhere in the system. The facility still pays for that inefficiency through higher pump energy, greater infrastructure demand and reduced operational flexibility.
In Formula One, engineers learn to think this way instinctively. No component is judged in isolation. A design that improves one metric but compromises another may not make the car faster. The same principle applies in data centres: the best cooling solution is not simply the one that removes the most heat, but the one that delivers the required thermal performance using the least total energy.
Balancing thermal performance and hydraulic efficiency
The challenge in liquid cooling is therefore not simply removing more heat but doing so with maximum efficiency. That requires balancing two demanding objectives simultaneously: pressure drop as we’ve discussed and thermal resistance.
Thermal resistance and pressure drop need to be engineered together. Optimising one at the expense of the other can create a system that performs well on a single metric, but less effectively in real operating conditions.
At Conflux, our latest single-phase cold plate designs are achieving thermal resistance as low as 0.0076°C/W at 6LPM and a 2500mm2, while maintaining exceptionally low pressure drop, placing them at the leading edge of what is currently achievable in liquid-cooling performance.
For data centre operators, this creates genuine operational flexibility. Higher thermal efficiency can be translated into lower energy consumption, reduced operating costs, higher rack density, or the optimal balance between all three depending on the priorities of the facility.
As AI infrastructure scales, cooling performance will increasingly be assessed not only by thermal capacity, but by how efficiently that capacity is delivered across the full cooling loop.
Better thermal performance starts with better engineering
Achieving this level of performance is not simply a consequence of additive manufacturing. It begins with first-principles engineering discipline. Drawing on decades of experience developing thermal management systems for Formula One, high-performance automotive and aerospace applications, Conflux’s engineering team uses computational fluid dynamics, thermal simulation and additive manufacturing to optimise coolant flow at a microfluidic scale.
The design objective is not just to increase turbulence, add surface area or push more coolant through a component. Those approaches can improve heat transfer, but they can also increase pressure drop and drive pump power. Effective design requires a more precise balance: directing coolant to the areas of highest thermal load, improving local heat transfer and removing unnecessary hydraulic losses wherever they appear.
This is where additive manufacturing becomes powerful. It enables internal channels to be shaped in ways that conventional manufacturing cannot easily achieve, allowing coolant to be guided through complex geometries with far greater control. The result is a cold plate capable of achieving both exceptionally low thermal resistance and very low pressure drop – two characteristics that are often difficult to optimise simultaneously.
The manufacturing process also improves reliability. Rather than assembling multiple machined layers together, additive manufacturing enables cold plates to be produced as single monolithic components. Removing bonded interfaces significantly reduces
potential leak paths while improving structural integrity, an important consideration in facilities where even brief interruptions can have significant operational and financial consequences.
It is a principle well established in Formula One engineering, where monolithic components are widely used to improve reliability by eliminating joints while withstanding extreme thermal and mechanical loads.
Engineering that evolves as quickly as technology
Perhaps the biggest lesson Formula One offers AI infrastructure is not simply how to engineer better cooling systems, it’s how to engineer them faster.
Traditional manufacturing relies on tooling, multiple production stages and assembled components, making design changes slower and more expensive. Because additive manufacturing produces components directly from digital models, engineers can rapidly adapt internal flow paths, channel geometries and mounting configurations without redesigning production tooling. Cooling solutions can therefore evolve alongside successive processor generations rather than constraining them.
Rapid iteration also depends on being able to validate performance just as quickly. At Conflux, in-house calorimetric testing allows single-phase cold plate designs to be thermally characterised and verified within days rather than weeks, while rack-scale testing capability enables system-level design verification under representative data-centre operating conditions.
By combining design, manufacture and performance testing under one roof, engineers can rapidly refine designs using real-world performance data, significantly accelerating development cycles.
For organisations deploying AI infrastructure, that agility is becoming almost as valuable as absolute thermal performance.
Engineering for an industry that never stands still
At first glance, Formula One and AI infrastructure appear to be worlds apart, yet the parallels are striking once you spot them. Both operate on relentless development cycles, where every new generation of technology demands a corresponding evolution in the systems around it. In both, thermal management is crucial.
For data centres, liquid cooling will become the standard, but competitive advantage will depend on how efficiently it is engineered. As processor power continues to climb, systems that balance heat removal, pressure drop, reliability and deployment speed will be better placed to reduce operating cost, improve flexibility and support long-term performance.
Perhaps Formula One’s greatest lesson is that thermal management is never a finished problem. It is an engineering discipline built on continuous optimisation, rapid iteration and the ability to adapt as technology evolves.
Fortunately, engineers have always performed best when the heat is on.
