Mitsubishi Heavy and Nvidia partner on 10MW cooling for AI factory-scale data centers

Mitsubishi Heavy’s 10MW chiller and MCP announcement might sound like HVAC news, but it speaks directly to the physical constraints of frontier AI. As GPU clusters scale from megawatts to tens of megawatts and beyond, conventional data center designs start to buckle under thermal and water constraints. By aligning a modular cooling and power architecture explicitly with Nvidia’s DSX ‘AI factory’ platform, MHI is positioning itself as a critical enabler of the next wave of hyperscale AI build‑outs.

This matters for AGI timelines because infrastructure is increasingly the gating factor. If power, cooling and permitting bottlenecks slow the deployment of Blackwell‑ and Rubin‑class clusters, model scaling naturally decelerates. Conversely, if companies like MHI can make it turnkey to drop 10MW‑scale, liquid‑cooled AI modules into industrial settings with better PUE and lower water use, the ceiling on economically viable compute rises. That supports more and larger training runs, more experiments with agentic systems, and more geographic diversification of AI factories.