Inference infrastructure, built like a utility.
BXD Inference runs its own SOC 2 Tier III datacenter in Măgurele, Ilfov — 20MW of critical power, racks we own outright, and the GPUs to serve AI inference on top of them.
A datacenter we own,
not one we rent.
Every rack, every GPU and every megawatt behind BXD Inference sits in our own building outside Bucharest. No hyperscaler in the middle, no cloud markup on the hardware, and no queue for capacity that already exists on our floor.
-
01
Power
20MW of critical power capacity on site, so scaling up is a provisioning decision rather than a waiting list.
-
02
Density & cooling
Racks laid out for modern GPU thermal loads instead of retrofitted from general-purpose hosting.
-
03
Security & compliance
SOC 2 controls and Tier III operations, with physical access to the floor under our own control.
-
04
Operations
Our team runs the room. Hardware is owned and operated end to end, so escalation stops with us.
At a glance
| Location | Măgurele, Ilfov, Romania |
| Operator | BXD Import Export SRL |
| Critical power | 20 MW |
| Classification | Tier III |
| Compliance | SOC 2 |
| Hardware | Owned & operated |
| Deployment | Bare metal, dedicated racks, clusters |
| Inference | Hosted open-weight & custom models |
Serve any model
Open-weight, fine-tuned, or your own — deployed on our hardware behind one inference endpoint.
Rent the GPUs underneath
Dedicated or on-demand, straight from our own racks — no cloud markup.
01
Frontier-scale hosting
Serve the largest open models in production without waiting on shared cloud capacity — your traffic gets dedicated GPUs in our own facility.
02
Fine-tuned & proprietary models
Bring a fine-tune or a model your team built in-house. We containerize, secure, and scale it on hardware that stays yours for the term.
03
Dedicated GPU clusters
Rent bare-metal GPU capacity by the rack or the cluster, backed by 20MW of critical power and SOC 2 Tier III operations.
curl https://api.bxdinference.sbs/v1/chat \ -H "Authorization: Bearer $BXD_API_KEY" \ -H Content-Type: application/json \ -d '{ "model": "kimi-k3", "stream": true, "messages": [{"role": "user", "content": "..."}] }'
One API, every model
REST and streaming endpoints, shared auth, usage logs, and rate limits across every model and cluster you run — hosted or dedicated.
- OpenAI-compatible request shape, so existing clients point at us with a base-URL change
- Server-sent event streaming on every hosted model
- One key across hosted inference and your dedicated racks
Talk to our infrastructure team.
Tell us what you're running and we'll scope the right mix of hosted inference and dedicated hardware — usually within a day.
Măgurele, Ilfov
Romania Facility tours by appointment
Bl. C5, Sc. 2, Et. 1, Ap. 20
Sector 4, Bucharest 040975 BXD Import Export SRL · CUI RO39336570