AI · Software · Technology
GTC Berlin 2026 local AI announcements: distinguishing available features from upcoming releases
NVIDIA’s local AI announcements at GTC Berlin, held October 20–22, 2026, are not all available immediately. They include items already offered, items whose supply begins October 23, and items coming later this month or in December. This article organizes the announcements by timing and lists what to check first when considering a purchase or deployment. It is not an on-site report or product review; the information below is based on NVIDIA’s official materials.
1. Check the event name first
This event is GTC Berlin, not the March GTC. NVIDIA’s official announcement stated that Jensen Huang’s keynote was scheduled for October 21 at Tempodrom, with sessions and hands-on labs continuing at STATION-Berlin. If you see an article with different dates, first compare the event name and dates.
2. Why the three product lines should not be treated as the same equipment
NVIDIA describes RTX Spark as a product for running personal agents on Windows PCs, DGX Spark as a product for development and scaling on Linux, and DGX Station as a product for large teams’ workloads. Their operating systems and purposes differ, so you should not conclude that one device is better based only on a single memory figure.
Local AI means running models on the device. However, connecting email, search, or external APIs can send data through those connection paths. You must separately check device selection and whether data is transmitted.
3. What is available now
NVIDIA’s official blog states that Underdog Desktop Agent is available on RTX Spark. The company explains that Underdog runs models on the device, so you should personally check connection settings and privacy options before actual use. RTX Spark’s release date and announced specifications are summarized separately in What to check first in the RTX Spark Windows PC release date and announced specifications .
4. What is coming on October 23 and later this month
Supply of the DGX Spark 64GB system was announced to begin through partners on Friday, October 23. NVIDIA explains that this system can run models such as Nemotron 3.5 Bolt 74B (24GB), Qwen3.8 27B, and Gemma 4 26B-A4B.
NVIDIA Sync Model Launcher is scheduled to become available later this month. It was introduced as a tool that downloads models and containers suited to the configuration and connects them to OpenCode. Configurations combining multiple DGX Spark systems reportedly support up to eight units based on manual setup.
5. What to look at from December onward
DGX Station for Windows was announced for release in December. The announcement also lists DGX Station’s 748GB of coherent memory and up to 20 petaflops of AI performance, but these are likewise specifications presented by the company. How fast it is in actual work can only be determined by measuring it with the same model and input conditions.
6. A checklist for applying the announcements to your situation
Writing down the five items below makes it easier to apply the announcements to a purchasing decision.
1) Availability status: Distinguish whether it can be purchased now, has a supply start date, or is merely planned.
2) Operating system: Check whether it is a Windows or Linux device.
3) Model and usage conditions: Check whether the model you will use is on the list and whether the license terms are suitable.
4) Data paths: Separately record which apps or APIs communicate with external servers apart from model execution.
5) Performance conditions: Check which model, engine, device, and input size produced the announced figures.
7. Recommended related reading
Before quoting announcement statements, it is safest to compare the original link and date. The method is summarized in How to verify the sources of AI answers .
Sources
NVIDIA GTC Berlin information: https://www.nvidia.com/en-eu/gtc/
NVIDIA blog announcement: https://blogs.nvidia.com/blog/local-ai-gtc-rtx-dgx-spark-station-sync/
Date official materials were checked: October 22, 2026. Announced figures and availability dates are based on the company’s announcement and may vary by region, retailer, and software updates. The equipment was not tested directly.
Sources and original article
Read the Korean original on Naver
This article reflects public information available when it was written. Specifications, prices and schedules can change; check the linked official documentation.
This English edition was written and translated with AI assistance, with sources provided. For errors or additional information, please request a correction. Editorial policy