Hyperfusion Joins Smart Inference: Fast Endpoints in the Middle East

We are adding Hyperfusion as another provider to Smart Inference. They run their own GPU data centers in the UAE, and their API is OpenAI compatible, so nothing changes on your side. You keep the same endpoint and the same code, and the router gets one more place to send your request.

Almost every provider behind Smart Inference so far sits in the US or in Europe. That is fine if your users are in Frankfurt or Virginia. It is a lot less fine if they are in Dubai, Riyadh or Mumbai, because every request then travels to Europe or the US and back before anyone sees a token. Hyperfusion is our first provider with endpoints in the Middle East, and they quote sub-50 ms latency across the UAE, India, the wider MENA region and Eastern Europe.

Their catalog covers open-weight models like GPT-OSS, Qwen3, Gemma 3 and DeepSeek-R1-Distill. Their API lists the models and the prices in two separate places, so our adapter reads both and matches them by exact model id. After that, Hyperfusion is just another candidate. On every request the router drops whoever cannot serve the model or is not healthy right now, then picks on price.

I like this one because a router can only pick from what it has, and until now almost everything it had sat in the US and Europe. Hyperfusion adds a third region.

If you run inference somewhere we do not cover yet, the provider page says what we need from you.

Welcome aboard, Hyperfusion.