RITS runs a small cluster of GPU machines on campus that serves AI models locally. It exists so that teaching and research at NYU Shanghai can use modern models without the material ever leaving hardware we control.

What it does

The stack hosts open-weight models behind internal APIs, so a course tool, a research script, or a demo can call them the same way it would call a commercial service:

• Language models — text generation, summarisation, and question answering.
• Speech to text — transcription of recordings and interviews.
• Text to speech — synthetic narration for teaching material.
• Image generation — illustration and visual prototyping.

Why run it locally

Two reasons. The first is data: some research material and some student work should not be sent to a third party at all, and running the model on our own machines removes the question. The second is cost — an experiment that would be expensive to run against a metered API is simply free here, which makes it possible to try things that would otherwise not get tried.

What it is good for

• Course tools that need to call a model many times without a per-token bill.
• Research pipelines processing material that should stay on campus.
• Comparing open-weight models against each other on your own data.
• Prototyping something before deciding whether it warrants a commercial service.

Getting access

Access is arranged by consultation rather than self-service, because the right answer is often a different tool. Describe what you are trying to do — the task, the kind of data, and roughly how much of it — and write to shanghai.genai@nyu.edu.

If your work involves sensitive research data, note that NYU also operates the Pythia platform for exactly that case; we will point you there when it is the better fit. See NYU AI Policies & Resources.