Frore Systems Unveils Advanced Cooling Technology to Tackle Escalating AI Data Centre Heat

2026-03-13 hardware

San Jose, Friday 13 March 2026
As AI infrastructure faces unprecedented thermal limits, Frore Systems’ newly launched liquid cooling technology cuts data centre power consumption by 10%, offering a vital efficiency breakthrough.

The Thermal Bottleneck in High-Density Computing

The operational demands of artificial intelligence are pushing rack densities to unprecedented levels, making thermal management a defining constraint on capacity delivery [2]. Goldman Sachs estimates that capital spending on AI infrastructure will reach $527 billion in 2026 [4]. By 2030, overall power consumption from AI data centres is forecast to jump by 175% compared to 2023 levels [4]. Mechanical cooling systems traditionally consume between 30% and 40% of a data centre’s total energy budget, creating a critical bottleneck for operators [3].

Shifting Architectures: Liquid Cooling Becomes Standard

As rack densities continue to rise, the industry is rapidly transitioning from traditional air cooling toward hybrid and liquid-based architectures [2]. On 28 February 2026, Vertiv released a reference architecture for the NVIDIA GB300 NVL72 platform, available as SimReady assets [5]. This 2.5 MW capacity design relies predominantly on direct-to-chip liquid cooling, which manages 77% of the thermal load, while air cooling handles the remaining 23% [5].

Geopolitical and Environmental Pressures on Infrastructure

The sheer scale of power required by modern AI infrastructure is forcing operators to look beyond conventional geography [3]. A single hyperscale facility can draw more than 1,000 MW of electricity, which is sufficient to power 800,000 homes in the United States [3]. To achieve preferred operating margins by eliminating mechanical cooling costs, developers are experimenting with Arctic installations, underground facilities, and subsea modules that utilise passive cooling [3]. However, these unconventional sites face significant hurdles regarding data connectivity, as physical fibre networks are expensive and slow to build, particularly in remote Arctic regions [3].

The Carbon Cost of Artificial Intelligence

The Elsham development starkly illustrates the immense resource requirements of next-generation infrastructure. Drawing 1 GW of electricity, the site’s power demand equates to 2.2% of the total UK electricity demand as of February 2026 [6]. The facility will span approximately 1.5 million square metres across 15 data halls, though it will only generate 50 MW of electricity on-site, meaning internal generation covers just 5 per cent of its total power draw [6]. In an attempt to mitigate its environmental footprint, heat from the Elsham site will be piped to nearby greenhouses for growing tomatoes [6].

Sources & Ecosystem Partners

Thermal management AI infrastructure