What this video shows
Huang presents NVIDIA's view of an AI factory as the infrastructure that turns energy and data into model output. The keynote covers containerized DSX POD systems, an OpenShell architecture for governed agents, Sentry enforcement on BlueField-4, and local systems for developers and enterprises.
Use the talk as a map of NVIDIA's platform rather than proof that every component is available or suitable. Record each release stage, test security controls outside the agent host, and measure end-to-end workload performance before choosing hardware.
Read Learnetto's GTC Berlin announcement analysis for availability, caveats and an evaluation checklist. NVIDIA's OpenShell announcement describes the architecture and participating ecosystem.
What you will learn
- Treat DSX POD's 14-day figure as prepared-site deployment, not the complete project schedule.
- Test whether agent policy remains enforceable when host software is compromised.
- Separate shipping software, previews, reference designs and partner plans.
- Benchmark the exact model, context, concurrency and latency target on candidate hardware.
How to apply this safely
- Choose one real workload and write its quality, latency, concurrency and cost targets.
- Map the model endpoint, harness, tools, identities, policy engine and human approvals.
- List release stage and external dependencies for each product in the proposed stack.
- Run a smaller baseline and compare security, operations work, power and total cost.
Important limitations
- This is a vendor keynote and presents NVIDIA's product strategy and performance claims.
- The talk does not provide independent security testing, deployment measurements or total ownership costs.
- Availability differs across downloadable software, previews and partner systems planned for later delivery.
Sources to check
- NVIDIA GTC Berlin 2026 Keynote Official NVIDIA video published October 23, 2026.
- GTC Berlin 2026 keynote report Official NVIDIA report published October 21, 2026.
Continue learning on Learnetto
NVIDIA GTC Berlin announcements explained
Separate confirmed releases from claims and future delivery.
Sandbox AI coding agents
Define tools, permissions and external controls.
Switch LLM models
Compare candidate systems on representative work.