The AMD Instinct MI350P with 144 GB HBM3e is AMD’s newest inference GPU - strong price-performance, fully managed by GEKI with SLAs in the right data center.
Non-binding · reply within 1–2 business days
AMD Instinct MI350P, operated by GEKI.
AMD’s newest generation, 144 GB HBM3e, built for inference. That makes the MI350P a very economical base for open models.
The full technical specs: AMD Instinct MI350P →
Strong price-performance
Plenty of HBM3e memory and high bandwidth - for inference workloads often the more economical choice.
AMD’s newest generation
MI350 series with 144 GB HBM3e, built for AI inference.
Sovereign & open
Run open models in the EU, under your control and without hyperscaler lock-in.
Fully managed
GEKI plans, delivers and operates with SLAs. You don’t need your own GPU team.
Fully managed AI factories in different configurations, matched to your models.
2× MI350P
288 GB total VRAM
A compact entry into your own sovereign AI factory. Fits e.g. DeepSeek V4 Flash, Qwen 3.6 or Gemma 4 for a high number of concurrent users.
8× MI350P
1,152 GB total VRAM (≈ 1.1 TB)
More capacity for larger models and many parallel users. Fits e.g. GLM 5.2 or current models around 1T parameters.
Suitable for companies and SaaS providers that want to run open models under their own control. Get in touch →
Open models on your own factory cover a large share of enterprise tasks.
A good fit for
Less suited for
An honest assessment of which configuration fits - or what to do instead.
A clear offer plus operations as managed pricing.
GEKI operates with SLAs in the right data center and supports your workloads and agents.
A short call with Matthias — an honest assessment of configuration and operating model.
