Edge AI and local models: what your own AI hardware costs in 2026
AI · not reviewedWritten with AI from the sources listed, not yet editorially reviewed.
Open models now run on a laptop or a single GPU — but the memory shortage has made edge modules up to twice as expensive.
The essentials in four points
- In July 2026, NVIDIA raised prices for Jetson modules and developer kits by up to 101%; the Jetson AGX Thor Developer Kit went from USD 3,499 to 5,499.
- Small modules are affected too: the Jetson Orin Nano Super Developer Kit now costs USD 399 instead of 249.
- Open models with permissive licences cover many sizes: small Qwen3 variants (Apache 2.0) run with 16 GB of memory, gpt-oss-20b (Apache 2.0) needs about 16 GB, gpt-oss-120b about 80 GB.
- Data protection remains the biggest hurdle: 66% of German companies using AI name data protection requirements as their biggest obstacle; 45% worry that company data could fall into the wrong hands.
What this means for businesses
For many tasks — searching documents, drafting text, checking images — open models can run in-house without handing data to third parties. Plan hardware budgets at current, higher prices and check the terms of each licence, such as OpenAI’s usage policy for gpt-oss or Meta’s Llama licence.
Sources · 3
- CNX Software · 22 Jul 2026NVIDIA increases the price of Jetson modules and devkits by up to 101%
- Hugging Face (Community-Blog) · 13 May 2026Open-Source- und Open-Weight-Modelle für den lokalen Betrieb 2026 (Übersicht)
- Bitkom · 14 Sept 2026Erstmals nutzt die Mehrheit der Unternehmen KI
Links open the original source in a new tab. Figures from company statements are labelled as vendor or company figures in the text.
Matching hardware
Private AI on premises
Run language models in-house so that confidential data never leaves your organisation.
More from this branch
All in Edge AI →The next OROBO Report comes to you.
Planned from December 2026, every Thursday. Free, unsubscribe in one click.