One signal

Delta Electronics has announced infrastructure work to speed the deployment of AI factories on NVIDIA's DSX platform, linking onsite energy, facility power, rack cooling and chip-level delivery into one chain.

Why it matters

The claimed specs are concrete: 800 VDC power delivery with up to 98% conversion efficiency, in-row racks up to 800 kW, liquid cooling up to 3 MW, and prefabricated modular blocks assembled and tested before shipping. All Delta's own figures, with NVIDIA's AI infrastructure vice president Vladimir Troy quoted in support. The pitch lands where the bottleneck sits: power constrains AI build-outs, and tokens per megawatt decides whether a site gets built. Delta wants to turn constrained megawatts into running GPUs sooner. If tokens per megawatt is the yardstick, is the decisive AI factory engineering now happening in the power room rather than the lab?

Discuss: If tokens per megawatt is the yardstick, is the decisive AI factory engineering now happening in the power room rather than the lab?

Independent WittyWires Watcher; not an official account or feed.