# Blackfuel Launches First Inference-Only Deployment in Barcelona By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/50500 Published: 2026-11-07T02:04:09.000Z Updated: 2026-11-07T02:04:09.000Z Section: Technology > The site uses AMD Instinct MI355X accelerators and open serving software. Customers can reserve capacity based on throughput, latency and data-location requirements. Blackfuel has opened its first inference-only deployment in Barcelona, letting customers reserve AI model-serving capacity on AMD hardware based on throughput, latency and data-location requirements. Digital Realty hosts the installation, which uses liquid-cooled Dell PowerRack systems with Dell PowerEdge XE9785L servers and AMD Instinct MI355X accelerators. The MI355X is based on AMD’s CDNA 4 architecture and has 288 gigabytes of HBM3E memory, with bandwidth of up to 8 terabytes per second. Dell and NTT DATA are supporting the deployment and integration. Blackfuel’s AI Token Grid connects accelerator clusters through a shared software and commercial layer. The company says the system is designed to make AI output measurable and reservable across multiple accelerator architectures and hardware generations. It combines compute, data-center capacity, storage and software in one system, giving customers choices over deployment locations, models and operating environments. The units Blackfuel calls “Logical GPUs” are contracts that let customers reserve part of a model’s serving capacity for a fixed monthly fee. Customers can set requirements for throughput, latency and data location without managing the underlying infrastructure. The contract covers a share of service capacity, while Blackfuel’s system is designed to maintain service levels as hardware and configurations change. The offering is aimed at enterprises, model developers and AI-native businesses. The software stack combines AMD ROCm with the open serving tools vLLM and SGLang, along with Blackfuel’s routing, allocation and metering software. The platform is designed to support models from Meta, OpenAI, NVIDIA and Google, among others. Blackfuel says the software approach aims to lower serving costs and increase useful tokens per megawatt, while giving developers and operators flexibility as models and frameworks evolve. The Barcelona site is Blackfuel’s first inference-only deployment as the company works toward global inference-native cloud infrastructure. Separately, [Cerebras Systems and Oracle have planned an AI inference service through Oracle Cloud Infrastructure](), with availability planned for calendar 2027. ## Links in this article - [Cerebras Systems and Oracle have planned an AI inference service through Oracle Cloud Infrastructure](https://www.tokenpost.com/news/business/42063)