Blackfuel Emerges From Stealth With Over $250 Million In Contracted Revenue

Blackfuel said it has emerged from stealth with more than $250 million in contracted revenue under multi-year customer agreements as it unveiled plans to build a global AI Token Grid. The company said the first deployment is scheduled to enter service in 2026. It is building architecture-specific accelerator clusters linked by a common software and commercial layer to turn dedicated compute into reservable AI output. Blackfuel said it is working with AMD and NTT DATA on European-operated inference capacity, starting with a GPU cluster that is currently being delivered in Europe.

Blackfuel says it is building an integrated compute platform that combines dedicated accelerator infrastructure with software and services aimed at making production AI more predictable for customers. The company says its model spans large customer deployments alongside a broader inference offering for model developers, enterprises and AI-native businesses. Its first inference-only deployment will be in Barcelona, Spain, where it will be hosted by Digital Realty and built on liquid-cooled Dell PowerRack systems with Dell PowerEdge XE9785L servers and AMD Instinct MI355X accelerators.

Dell AI Data Platform, via Dell PowerScale and Dell PowerStore, will provide the data foundation, with deployment and integration support from Dell and NTT DATA. Blackfuel said it is also advancing additional cluster deployments in Spain, Finland and France. Its offering includes dedicated GPU clusters, managed inference and reserved serving capacity. Customers can either keep control of their compute environment or work with Blackfuel to manage it.

Blackfuel says it manages the serving layer against agreed workload and performance requirements. In addition to dedicated customer deployments, it is building a separate inference pool for its Token Platform, with day-1 support at launch for selected models from Meta, Mistral, H Company, DeepSeek, Qwen, GLM and Moonshot AI's Kimi. The company says coding assistants and agentic applications are priority workloads. Through one enterprise interface, customers can choose models and reserve inference capacity against defined throughput, latency and data-location requirements, without handling the underlying infrastructure.

Reserved-capacity offers are set model by model under agreed service terms. Blackfuel is also introducing fractional GPU capacity through Logical GPUs, service units that let customers reserve a share of a model's serving capacity for a fixed monthly fee instead of renting an entire multi-GPU deployment. Commitments define the selected model, input and output token quotas and applicable service levels, sized around the customer's workload. Customers can reserve less than a full model replica.

Blackfuel can evolve the physical Blackfuel says its inference stack combines open serving tools, including vLLM and SGLang, with proprietary routing, allocation, metering and commercial-control software. For AMD-based infrastructure, the stack uses AMD ROCm™ Software. Customers can opt for usage-based serverless access or reserved capacity under defined service terms, with model routing, organization- and project-level controls, observability and cost tracking. Blackfuel also says it is developing an agent-driven process to build, benchmark and continuously optimize inference runtimes for each model, accelerator architecture and workload.

The approach is intended to reduce serving costs and increase useful tokens per megawatt within defined quality and latency requirements, while keeping successive generations of hardware economically productive for longer. "The opportunity is not simply to run models on more chips. It is to continuously improve how each model runs on each architecture. We are developing" Blackfuel is building its platform to support multiple accelerator architectures and hardware generations, with systems chosen to match workload needs, performance, availability and lifetime economics.

Each architecture will run in its own optimised execution environment, linked by a common service and commercial layer intended to keep hardware commercially productive through ongoing software improvements. The common layer is designed to make capacity commercially portable across locations, customers and models, subject to the limits of each deployment and customer agreement. Blackfuel says its model ties together customer demand, data-centre capacity, accelerator systems, operating partners and project and equipment financing in a coordinated deployment framework.

Its software connects that infrastructure to dedicated compute and contracted inference services, with the model designed to make AI output financeable through contracted token. Blackfuel says its software is intended to improve infrastructure economics across the asset lifecycle, lifting productivity in dedicated deployments while also monetizing uncommitted and residual capacity through shared inference services. The company says the model preserves customer reservations and agreed service levels while extending the commercial life of the underlying hardware. Blackfuel said it plans to start with European deployments and then replicate the model worldwide.

Xavier Fischer, co-founder of Blackfuel Sovereign AI at Industrial Scale, said Europe has the electrons and researchers it needs to own more of its AI capacity, turning what is often an operating cost paid abroad into a capital asset on European balance sheets. Fischer said Blackfuel is not an inference company that happens to own hardware, and not a landlord that happens to sell tokens. He said the company contracts demand first, builds against it, and requires every GPU it did not contract to earn its keep. Fischer said that order is what makes the asset financeable.

Blackfuel said its ambition is to make contracted token output as financeable as contracted power. The source text ends mid-sentence with: "Europe has the". Blackfuel says its AI-sovereignty model pairs European-operated infrastructure with a service structure that gives customers control over model selection, deployment location and capacity commitments. The company says the setup is intended to let enterprises decide where inference runs and how it is supplied, with deployment and support arrangements tailored to each customer's regional requirements.

Blackfuel describes its goal as global reach with local control, saying sovereignty needs are built into the technical, operational and contractual design of each deployment. It says its edge comes from combining that regional control with industrial-scale capacity, global technology partnerships and software designed to improve compute economics. Blackfuel says dedicated clusters are used for large commitments, while fractional capacity and the Token Platform extend the same operating expertise to a wider enterprise market. European deployments are the basis for that expansion.

AMD said its Instinct accelerators and ROCm open software stack are intended to deliver the performance, openness and flexibility customers need to build and run AI infrastructure at scale. Stephanie Dismore, senior vice president of EMEA at AMD, said the company looks forward to working with Blackfuel as it brings AMD technology to its initial deployment and expands access to high-performance AI infrastructure. Dell Technologies said organisations are moving from AI ambition to AI impact, and said that shift requires production infrastructure rather than experimentation.

Adrian McDonald, president of EMEA at Dell Technologies, said its full-stack AI platform gives customers a base to deploy inference at scale with the performance and reliability the moment requires. Digital Realty said AI capacity needs the right foundations, including power, cooling, connectivity and room to grow. Paula Cogan, managing director for EMEA, said Blackfuel's inference deployment in Barcelona brings those foundations together and supports AI infrastructure built for long-term customer demand. Digital Realty and NTT DATA said Blackfuel is an AI-native neocloud built for inference at scale, co-founded by Charles Kantor, Xavier Fischer and Dali Kilani.

The company said its founding team has experience across frontier AI, model quantization and cloud software, and that its platform combines dedicated accelerator infrastructure, proprietary software and long-term commercial commitments to build and operate AI compute at scale. Blackfuel said its service covers dedicated capacity and managed inference needs, with European deployments forming the basis for international expansion. The agreement with Dell Technologies was signed in Q2FY27.