Blackfuel has officially launched its first inference-only deployment in Barcelona, leveraging AMD Instinct™ MI355X GPUs and AMD ROCm™ open software. This marks a significant step in the company’s mission to build a global, inference-native cloud infrastructure aimed at delivering measurable and reservable AI output.
Barcelona Deployment: A New Model for AI Infrastructure
The Barcelona deployment is hosted by Digital Realty and features liquid-cooled Dell PowerRack systems equipped with Dell PowerEdge XE9785L servers powered by AMD Instinct MI355X accelerators. Deployment and integration support are being provided by Dell and NTT DATA. Blackfuel’s approach integrates compute, data-center capacity, storage, and software into a unified system, offering customers greater control over deployment location, models, and operating environments.
Blackfuel’s infrastructure is designed to optimize inference across multiple accelerator architectures and hardware generations. The AMD Instinct MI355X GPUs, based on CDNA 4 architecture, include 288 GB of HBM3E memory and up to 8 TB/s memory bandwidth, providing the hardware backbone for the platform.
AI Token Grid: A Shift Toward Output-Driven Infrastructure
At the core of Blackfuel’s strategy is its AI Token Grid, a network of architecture-specific accelerator clusters connected through a common software and commercial layer. This platform enables customers to reserve inference capacity aligned with specific throughput, latency, and data-location requirements without managing the underlying infrastructure.
Blackfuel is also introducing Logical GPUs, contractual service units that allow customers to reserve a fraction of a model’s serving capacity for a fixed monthly fee. This model ensures consistent service levels while allowing the underlying hardware and configurations to evolve.
Open Software as a Strategic Foundation
Blackfuel’s software stack builds on AMD ROCm™ and open serving technologies such as vLLM and SGLang. By combining open software with proprietary routing, allocation, and metering tools, the company aims to lower serving costs and increase useful tokens per megawatt. The platform is designed to support a wide range of models, including those from Meta, OpenAI, NVIDIA, and Google, among others.
Open software provides flexibility for developers and operators to adapt to rapidly evolving AI models and frameworks, ensuring that Blackfuel’s infrastructure remains future-proof.
Implications for the AI Ecosystem
Blackfuel’s Barcelona deployment highlights a broader shift in AI infrastructure from capacity-focused to output-driven models. By offering locally operated, reservable AI capacity, the company provides enterprises, model developers, and AI-native businesses with an alternative to traditional cloud providers. This approach aligns infrastructure capabilities with customer-specific performance and service commitments, setting a new standard for AI inference at scale.
