NVIDIA launched DSX POD, a factory-integrated AI factory reference design based
on the Vera Rubin architecture, targeting faster European AI infrastructure
rollouts. Partners build modular container units that are fully integrated and
validated in manufacturing; once on site they require only connection to
pre-provisioned power, cooling and networking to operate, NVIDIA said, cutting
deployment time from about six months to as little as 14 days. The design moves
compute, networking, cabling, rack-level power and liquid cooling installation
into the factory, reducing on-site planning, construction, skilled labor and
logistics and lowering risks from component delays and error-prone field work.
NVIDIA said factory-tested containers shorten the path to model training and
production inference, enabling customers to start generating tokens and revenue
earlier. JENSEN HUANG: AI factory creates value when it starts producing tokens,
not when construction finishes.
