NVIDIA GTC Berlin 2026 spans 20–22 October, but the program has distinct dates within that window. Workshops are listed for 20 October, while the conference and expo run on 21–22 October. Jensen Huang's keynote is scheduled for 21 October.
For developers, that distinction helps with planning. A training session, a keynote, and an exhibition are different ways to learn about a technology. Decide which one answers your question before trying to follow every part of the event.
Dates to keep separate
The overall event window is useful for a calendar headline. For a specific session, use the organizer's schedule and local time. A broad conference date does not identify when a particular announcement or workshop will happen.
NVIDIA lists AI infrastructure, agents, open models, robotics, and physical AI among the program's topics. Those are confirmed areas of coverage. They should not be turned into predictions that a particular product will launch.
Choose a topic based on the project you have
If you are serving a model in an application, begin with the question that is limiting your project: cost, response time, memory use, or reliability. Use that question to choose sessions and documentation to read afterward.
If your work involves robotics or another physical system, focus on how a demonstrated approach is evaluated outside a controlled setting. For an AI agent, look for information about tool access, task boundaries, and how failures are detected.
These are ways to evaluate the material, not claims about a specific system that has yet to be announced. The goal is to leave with an experiment you can run, rather than only a larger collection of slides.
How to read a performance demonstration
Record the workload and the conditions. Which model or task is being used? What input size is shown? Is the comparison running on the same type of system, and is the result measured for one device or a larger configuration?
Also ask what happens after the first successful run. A production application needs to handle repeated requests, unusual inputs, and failures. A headline speed figure is a starting point for investigation, not a complete deployment decision.
For a useful comparison, keep your own target task constant. If a session provides a reproducible example, try it with a small sample that resembles your actual workload.
What to look for after an announcement
Find the accompanying documentation and check release status. A preview is an invitation to evaluate; general availability carries a different expectation about who can use the product and under what conditions.
Look for access requirements, pricing information, and the parts of the demonstration that are publicly reproducible. If those details are absent, keep the story in the announcement category rather than presenting it as a tool anyone can use immediately.
A practical way to follow GTC
Pick one main topic and one secondary topic. Save the official session links, write down the questions you want answered, and take notes against those questions. That keeps the event useful even if you have limited time to watch.
This preview uses the organizer's published schedule. It does not name hypothetical launches. The next useful step is to check the live agenda, then follow the keynote and relevant sessions for confirmed details you can test in your own work.
