Spike-1 provides SOTA sovereign LLMs access to all TU/e researchers
TU/e Supercomputing Center (SCC) and EAISI now host sovereign LLMs on Spike-1 — an NVIDIA DGX B200 system owned by TU/e — and open access to all TU/e researchers over an internal API, no vendor in between.
Every researcher can start on the light tier models (requiring less than 8 GPUs) and a weekly budget of 10M prompt tokens and 1M completion tokens, by sending an email to supercomputing@tue.nl and stating:
"I would like to get access to Spike-1 hosted LLMs for my research"
Currently Hosted Models
| Model | GPUs |
|---|---|
| Kimi-K3 | 16 |
| GLM-5.3-FP8 | 8 |
| GLM-5.3-Flash-FP8 | 4 |
| qwen3-8-flash-next-fp8 | 2 |
| qwen3-8-27b-fp8 | 1 |
This model list updates over time, typically including Day-0 support for the latest open-weight models.
Access to Heavier Models & Higher Budgets
If you as a researcher need access to heavier models (GLM-5.3-FP8, Kimi-K3) or a larger budget, an application can be done by sending a concise project summary to supercomputing@tue.nl describing:
- What you are trying to achieve
- Why it is new
- Why it requires Spike-1
An example of such an application can be provided upon request.
Data Sovereignty & Privacy
Your prompts never leave our machines. We never store or inspect prompt or completion content: the hosted models write no prompt text to disk or logs; content lives in RAM only for the duration of the request. Your data and IP stay sovereign.
Experimental Service & Data Sensitivity
Currently, this service is experimental. Spike-1 is classified for low/medium data sensitivity. The LLM hosting is experimental with no GDPR/NIS2 guarantee. By requesting an API key, you acknowledge these risks.
