Blackfuel begins AI inference deployment at Digital Realty’s BCN1 data center in Barcelona
Редакция ЦОД:Главное
Blackfuel has begun deploying its AI inference platform at Digital Realty’s BCN1 data center in Barcelona, using liquid-cooled Dell Technologies AI Factory infrastructure and AMD GPUs. The platform will connect to Digital Realty’s ServiceFabric interconnection platform to provide dedicated inference capacity over private, low-latency connectivity.
- Blackfuel has begun deploying and scaling its AI inference platform at Digital Realty’s BCN1 facility in Barcelona.
- The infrastructure includes liquid-cooled Dell Technologies AI Factory systems and AMD GPUs.
- ServiceFabric is intended to provide private, low-latency access to dedicated inference capacity.
- Digital Realty estimates that a 50 ms TTFT reduction could save 1 to 2.5 seconds in agentic workflows with 20–50 sequential inference calls.
- No power capacity, CAPEX, GPU count or commissioning date has been disclosed.
Blackfuel has selected Digital Realty’s BCN1 facility in Barcelona as the site for deployment and scaling of its AI inference platform.
The deployment uses liquid-cooled Dell Technologies AI Factory infrastructure and AMD GPUs. Blackfuel is integrating the platform with Digital Realty’s software-defined ServiceFabric interconnection platform, with the stated aim of delivering dedicated AI inference capacity through private, low-latency connectivity.
For inference workloads, particularly agentic applications involving multiple sequential API calls, network latency can materially affect end-to-end response times. Digital Realty says a 50-millisecond reduction in time-to-first-token (TTFT) could shorten an agentic workflow comprising 20 to 50 sequential inference calls by 1 to 2.5 seconds versus public-internet connectivity. This is a Digital Realty estimate rather than an independently verified performance result from the Blackfuel deployment.
Digital Realty states that ServiceFabric spans more than 800 data centers worldwide. The BCN1 deployment therefore combines AI compute infrastructure, liquid cooling and private interconnection in a carrier-neutral data center setting.
For data center operators and interconnection providers, the project could indicate that the competitiveness of distributed AI infrastructure will increasingly depend on predictable latency and data locality alongside available compute capacity. The source does not disclose the deployment’s power capacity, GPU or server count, capital expenditure, commercial terms, or operational start date.
The deployment illustrates demand for AI inference infrastructure located close to enterprise data and users, where controlled low-latency connectivity can be as important as GPU availability and cooling. It also points to a broader convergence of liquid-cooled AI infrastructure and private interconnection within carrier-neutral data center platforms.
- Data Centre Magazine / 1 октября 2026 г. · АнглийскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.