This isn’t a paradox, but a sharp snapshot of the phase we’re going through: AI hardware is pushing the market toward ever more sophisticated cooling and green energy, while traditional electrical components are seeing demand slip. Gem Terminal’s June figures – with overall sales dipping just as orders for AI-dedicated cooling systems and renewable energy supplies surged – tell the story of an infrastructure adapting unevenly.

The thermal density of racks hosting GPUs like the NVIDIA H100 or the upcoming B100 is reaching levels where conventional air cooling falls short. A single server can dissipate over 2–3 kW, and modern racks are now heading toward 30–50 kW, forcing operators to invest in liquid loops, heat exchangers, and, in extreme cases, immersion cooling. This isn’t just a problem for hyperscalers: any organization evaluating on-premise deployment of LLMs must now factor thermal management into their TCO calculations.

At the same time, the demand for green power goes beyond ESG box-ticking. Data centers, especially local ones that can’t spread costs across thousands of nodes, are exposed to volatile electricity prices and grid bottlenecks. Powering inference workloads with renewable sources becomes a factor of economic competitiveness even before it’s an environmental one. That’s why orders for clean energy solutions are growing while those for generic electrical components decline.

Structurally, a convergence is emerging between compute power, thermal management, and energy sustainability. For those keeping infrastructure on-premise, the advantage will no longer be solely data sovereignty or latency, but the ability to design an optimized thermal and energy cycle, perhaps reusing waste heat for other industrial purposes. This game rewards suppliers capable of vertically integrating these technologies and risks sidelining those producing traditional components without clear diversification.

Gem Terminal’s signal should be read against the light: the growth of AI no longer lifts the entire sector indiscriminately; it sorts companies by their ability to respond to new physical and energy constraints. For system integrators and IT decision-makers, it becomes essential to assess whether their facility can thermally sustain the next generations of accelerators. Otherwise, the economic appeal of the cloud may prevail even in scenarios where data sovereignty would be paramount.