Nvidia Unveils Modular AI Factory and Agent Security at GTC Berlin: what are the key facts?

On October 21, Jensen Huang confirmed at GTC Berlin that the Vera Rubin platform has entered full mass production.

Nvidia Unveils Modular AI Factory and Agent Security at GTC Berlin: what are the key facts?

Nvidia launched the DSX POD modular AI factory, built from 7 liquid-cooled racks, with delivery expected in early 2027.

Nvidia Unveils Modular AI Factory and Agent Security at GTC Berlin: what are the key facts?

DSX MaxLPS dynamically allocates GPU power, and the company claims up to 30% lower power consumption at the same output.

Nvidia Unveils Modular AI Factory and Agent Security at GTC Berlin: what are the key facts?

The OpenShell enterprise solution uses kernel-level sandboxes to isolate agents, while Sentry delivers millisecond-level control on the DPU.

Nvidia Unveils Modular AI Factory and Agent Security at GTC Berlin: what are the key facts?

The Bolt program will collect data in Madrid in 2027, targeting at least 25,000 L4 vehicles deployed in Europe.

Physix Frontier · News Briefing Card (enbrief · Oct 23, 2026)

Nvidia Unveils Modular AI Factory and Agent Security at GTC Berlin

KEY FACTS

On October 21, Jensen Huang confirmed at GTC Berlin that the Vera Rubin platform has entered full mass production.
Nvidia launched the DSX POD modular AI factory, built from 7 liquid-cooled racks, with delivery expected in early 2027.
DSX MaxLPS dynamically allocates GPU power, and the company claims up to 30% lower power consumption at the same output.
The OpenShell enterprise solution uses kernel-level sandboxes to isolate agents, while Sentry delivers millisecond-level control on the DPU.
The Bolt program will collect data in Madrid in 2027, targeting at least 25,000 L4 vehicles deployed in Europe.

KEY DATA

as fast as 14 daysDSX POD deployment time

up to 30%DSX MaxLPS power reduction

up to 40%DSX Flex load reduction

at least 25,000 vehiclesBolt European deployment target

PHYSIX OBSERVATION

Nvidia is no longer just selling chips; it is packaging and selling how to build, manage, and secure an AI factory. Europe has the industrial base but lacks compute, and this offering lands right in that gap. The signal worth noting is that the metric for an AI factory is shifting from card count to tokens produced per kilowatt-hour, making energy efficiency and speed of deployment the key battlegrounds.

Source: enbrief original report ↗