AI infrastructure shifts toward liquid cooling and silicon photonics at OCP Summit

At the Open Compute Project (OCP) Global Summit 2026, industry leaders highlighted a fundamental shift in AI infrastructure design, moving away from traditional air-cooled server stacks toward integrated, high-density systems.

Nvidia introduced design contributions for its Vera Rubin platform via the MGX 1.2 specification. This architecture aims to provide a common base for interoperable AI systems, featuring rack-scale liquid cooling, fanless compute trays, and power densities reaching up to 250 kW per rack. This transition makes liquid cooling a critical requirement rather than an optional supplement.

Intel presented its vision for silicon photonics to address bandwidth and energy consumption limits in AI and high-performance computing. By using light instead of traditional copper interconnections, Intel aims to achieve interconnection energy levels below 5 pJ per bit. The company stated it is currently the only foundry capable of large-scale industrial production of 300 mm silicon photonics solutions integrating indium phosphide (InP).