Blackfuel Emerges From Stealth with More Than $250 Million in Contracted Revenue to Build a Global AI Token Grid

PARIS, September 30, 2026 – Blackfuel today emerged from stealth with more than $250 million in contracted revenue under multi-year customer agreements, unveiling its ambition to build a global AI Token Grid.

Pioneering the AI Token Grid, Blackfuel is building a network of architecture-specific accelerator clusters connected through a common software and commercial layer, designed to turn dedicated compute into measurable, reservable AI output.

By collaborating with AMD and NTT DATA, Blackfuel is developing its European-operated inference capacity, beginning with a GPU cluster currently being delivered in Europe.

“The cloud was built around virtual machines. The next infrastructure layer will be built around tokens. We are building Blackfuel to become one of the world's defining AI compute companies, starting with our first deployments in Europe. That means bringing industrial-scale infrastructure, software and capital together so customers can buy intelligence with the same predictability with which they buy power.”

Building the Infrastructure for Production AI

Blackfuel is developing an integrated compute platform, from dedicated accelerator infrastructure to the software and services that make it productive. The company combines large customer deployments with a broader inference offering for model developers, enterprises and AI-native businesses.

Its initial inference-only deployment in Barcelona, Spain, will be hosted by Digital Realty and built on liquid-cooled Dell PowerRack systems featuring Dell PowerEdge XE9785L servers with AMD Instinct™ MI355X accelerators¹. The Dell AI Data Platform (Dell PowerScale) and Dell PowerStore will provide the data foundation with deployment and integration support from both Dell and NTT DATA.

Blackfuel is advancing additional cluster deployments in Spain, Finland and France.

Blackfuel's offering spans dedicated GPU clusters, managed inference and reserved serving capacity. Customers can retain control of their compute environment or engage Blackfuel to manage the serving layer against agreed workload and performance requirements.

Alongside its dedicated customer deployments, Blackfuel is developing 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. Coding assistants and agentic applications are priority workloads.

Through a single enterprise interface, customers are able to select models and reserve inference capacity against defined throughput, latency and data-location requirements, without managing the underlying infrastructure. Reserved-capacity offers are specified model by model under agreed service terms.

Fractional GPU Capacity for Production Workloads

Blackfuel is 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 specify 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 hardware and serving configuration while maintaining the agreed capacity and service levels.

Maximizing Tokens per Megawatt Through Software

Blackfuel's inference stack combines open serving technologies, including vLLM and SGLang, with proprietary routing, allocation, metering and commercial-control software. For its AMD-based infrastructure, the stack builds on AMD ROCm™ Software. Customers can choose usage-based serverless access or reserved capacity under defined service terms, supported by model routing, organization- and project-level controls, observability and cost tracking.

Beyond orchestration, Blackfuel is developing an agent-driven process to build, benchmark and continuously optimize inference runtimes tailored to each model, accelerator architecture and workload. Informed by production workloads, the approach is designed to lower 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 software agents to build, test and refine those runtimes, with the goal of improving customer economics today and extending the productive life of the infrastructure.”

Blackfuel is developing its platform to support multiple accelerator architectures and hardware generations, selecting systems around workload requirements, performance, availability and lifetime economics. Each architecture will have its own optimized execution environment, connected through a common service and commercial layer. The aim is to keep hardware commercially productive through continuous software improvement. The common layer is designed to make capacity commercially portable across locations, customers and models, within the constraints of each deployment and customer agreement.

Connecting Customer Demand to Industrial Expansion

Blackfuel brings customer demand, data-center capacity, accelerator systems, operating partners and project and equipment financing into a coordinated deployment model. Its software connects that infrastructure to dedicated compute and contracted inference services.

The model is designed to make AI output financeable through contracted token supply and recurring cash flows. Blackfuel's software is designed to improve infrastructure economics throughout the asset lifecycle, increasing the productivity of dedicated deployments while enabling the monetization of uncommitted and residual capacity through shared inference services. This approach preserves customer reservations and agreed service levels while extending the commercial life of the underlying hardware. Starting with its European deployments, the company intends to replicate this deployment model worldwide.

“Europe has the electrons and it has the researchers. What it needs today is to own more of its AI capacity, turning what is often an operating cost paid abroad into a capital asset on European balance sheets. We are not an inference company that happens to own hardware, and we are not a landlord that happens to sell tokens. We contract the demand first, we build against it, and every GPU we did not contract has to earn its keep. That order is what makes the asset financeable.”

Sovereign AI at Industrial Scale

Blackfuel's ambition is to make contracted token output as financeable as contracted power. Europe has the engineering talent and power resources. The challenge is mobilizing capital at scale to turn those strengths into AI capacity it operates, rather than simply rents from elsewhere.

Blackfuel's approach to AI sovereignty combines European-operated infrastructure with a service model designed around customer control of model selection, deployment location and capacity commitments. The aim is to give enterprises greater control over where inference runs and how it is supplied, with deployment and support arrangements defined around each customer's regional requirements. The ambition is global reach with local control, with sovereignty requirements addressed in the technical, operational and contractual design of each deployment.

The company sees its competitive advantage in combining that regional control with industrial-scale capacity, global technology partnerships and software that improves compute economics. Dedicated clusters serve large commitments; fractional capacity and the Token Platform bring the same operating expertise to a broader enterprise market. European deployments form the foundation for that expansion.

Partner Perspectives

AMD

“Enterprises and developers are looking to build AI infrastructure that delivers high performance, scalability and greater control over where and how AI workloads are deployed. AMD Instinct™ accelerators and the AMD ROCm™ open software stack are designed to provide the performance, openness and flexibility customers need to build and operate AI infrastructure at scale. We look forward to working with Blackfuel as it brings AMD technology to its initial deployment and expands access to high-performance AI infrastructure.”

Dell Technologies

“Organizations are moving from AI ambition to AI impact. That shift demands infrastructure built for production, not just experimentation. With Dell's full-stack AI platform, they have the foundation to deploy inference at scale with the performance and reliability the moment requires.”

Digital Realty

“AI capacity requires the right foundations: power, cooling, connectivity and room to grow. Blackfuel's inference deployment in Barcelona brings those foundations together, supporting AI infrastructure built for long-term customer demand.”

NTT DATA

“As a sovereign full-stack AI infrastructure integrator, NTT DATA brings together compute, storage, networking, cybersecurity, operations and managed services to deliver production-ready AI platforms. Our work with Blackfuel demonstrates how we help clients deploy and scale secure, resilient AI infrastructure at industrial scale.”

About Blackfuel

Blackfuel is an AI-native neocloud built for inference at scale, co-founded by Charles Kantor, Xavier Fischer and Dali Kilani. Its founding team brings experience across frontier AI, model quantization and cloud software. The company combines dedicated accelerator infrastructure, proprietary software and long-term commercial commitments to build and operate AI compute at scale. Its platform serves dedicated capacity and managed inference needs, with European deployments forming the foundation for international expansion.

Media contact: comms@blackfuel.ai
Website: www.blackfuel.ai

1. The agreement with Dell Technologies was signed in Q2FY27. ↩