Blackfuel Launched European AI Inference Platform
The company has initiated service using AMD hardware to offer fractional GPU capacity to major model developers.
Updated on Sept. 30, 2026 in Data Centers
Live Poll
Should companies prioritize building sovereign AI infrastructure within their own region?
Blackfuel has launched an AI inference compute platform across Europe, backed by over $250 million in contracted capacity. The deployment currently utilizes AMD Instinct MI355X accelerators in Spain, Finland, and France.
Why it matters
By providing reservable, fractional compute capacity, Blackfuel aims to shift AI infrastructure from open-ended scaling to predictable, performance-optimized output. This approach is designed to maximize token generation per megawatt of power.
The platform utilizes AMD Instinct MI355X accelerators configured to support models from Meta, Mistral, H Company, DeepSeek, Qwen, GLM, and Moonshot AI. It optimizes inference performance via vLLM and SGLang serving technologies.
The players
Blackfuel
An infrastructure company building an inference compute platform centered on optimized accelerator clusters.
Digital Realty
A provider of global data center, colocation, and interconnection solutions hosting the Barcelona facility.
Dell Technologies
An enterprise hardware firm providing the server and infrastructure stack utilized in Blackfuel deployments.
AMD
A semiconductor company providing the Instinct MI355X accelerators that serve as the primary compute engine.
The details
Blackfuel constructs architecture-specific accelerator clusters linked through a specialized software layer that benchmarks and optimizes runtime performance. The company employs an agent-driven process to align specific model requirements with hardware capabilities. This process enables fractional GPU access through Logical GPUs, allowing users to rent discrete slices of high-end processing power rather than entire hardware units.
Timeline
Q2FY27: Signed agreement with Dell Technologies.
September 30, 2026: Official news publication date.
The Tech Race
Blackfuel is positioning itself against general-purpose cloud providers by focusing exclusively on inference efficiency using vLLM and SGLang serving stacks. This represents an attempt to optimize hardware lifecycles in an era of rapid AI model evolution.
Enterprise users and developers can now access fractional GPU capacity for running high-performance models in European markets. This workflow change replaces the need to procure entire, dedicated clusters for smaller-scale inference tasks.
The takeaway
Blackfuel is shifting toward predictable token-per-megawatt pricing for AI compute, a critical shift for sustainable infrastructure scaling. Watch for the company's planned expansion from its European pilot to global markets to see if this model achieves parity with hyperscaler pricing.
Further reading
Learn more about evolving infrastructure requirements in the Data Centers section.
More information
View platform details and technical documentation on the Blackfuel company website.
Live Poll
Should companies prioritize building sovereign AI infrastructure within their own region?
