THERMAL MANAGEMENT
The thermal chain: how heat is reshaping electronics design for AI infrastructure
By Michael Poto, product manager, global Chilled Water Systems at Vertiv
A
I hardware is pushing electronic components into operating regimes that were once considered exceptions. Accelerators, high-bandwidth memory and advanced power delivery circuits now run at intensive utilisation levels that generate continuous thermal stress. In this context, heat is no longer a secondary consideration addressed after electrical design is complete. Instead, it is a defining factor that influences performance, reliability and system architecture from the silicon level upwards. For electronics engineers, this shift requires a broader perspective. Thermal behaviour must be understood not only at the component or board level, but as part of a wider thermal chain that extends from the chip to the data centre plant.
Silicon performance and thermal limits
At the heart of AI systems are highly integrated devices operating close to their physical limits. As transistor densities increase and clock speeds remain high, power density at the die level continues to rise. While process improvements have delivered gains in efficiency, those gains are often offset by the scale and intensity of modern AI workloads. Thermal constraints directly affect achievable performance. Junction temperature influences leakage current, timing margins and long-term device reliability. For AI accelerators running extended training cycles, even small increases in temperature can reduce effective throughput or trigger throttling mechanisms designed to protect the silicon.
Board-level challenges and power delivery
Thermal behaviour at the board level introduces additional complexity. High- current power delivery networks generate heat within voltage regulators, inductors and interconnects, adding to the thermal load imposed by processors and memory. As power densities rise, traditional air cooling approaches struggle to remove heat evenly across densely populated boards. Localised hot spots can form around
20
power components, affecting efficiency and component lifespan even when average temperatures appear acceptable. This has implications for layout, component selection and power stage architecture. Engineers must balance electrical performance with thermal paths, allowing heat to be conducted away effectively without compromising signal integrity or manufacturability.
Moving heat off the board Once heat is generated, it must be transferred away from sensitive components as efficiently as possible. This is where the interface between data centre design and system-level cooling becomes critical. Direct-to-chip liquid cooling has gained attention because it addresses the thermal bottleneck at its source. By coupling cold plates directly to high-power devices, thermal resistance is reduced and temperature gradients across the die are minimised. For electronics engineers, this can translate into more predictable operating conditions and tighter control over thermal margins.
The role of plant-side systems Beyond the electronics enclosure, waste heat must ultimately be reused or rejected. The characteristics of this final stage of the thermal chain influence earlier design decisions more than is often recognised. AI-focused cooling strategies typically result
FEBRUARY 2026 | ELECTRONICS FOR ENGINEERS
in higher coolant return temperatures and more variable thermal loads. Cooling plants must therefore operate efficiently across a wide range of conditions rather than around a fixed operating point.
Where lower supply temperatures are required or guaranteed capacity must be maintained regardless of ambient conditions, centrifugal chiller technologies, in both air- cooled and water-cooled forms, provide a stable and scalable solution.
Control and observability What connects the thermal chain end to end is observability and control. Sensors embedded at the component, board and system levels feed data into control platforms that adjust cooling delivery in real time. This feedback loop enables cooling systems to respond to actual electronic behaviour rather than static assumptions. It also provides valuable data for engineers.
Designing with the thermal chain in mind
As AI continues to drive critical digital infrastructure into higher power and density regimes, thermal management must be treated as a core design parameter rather than an afterthought. Decisions made at the silicon, board and system levels are increasingly interdependent, linked by the movement of heat through the thermal chain.
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50
