Blackfuel today unveiled over $250 million in contracted revenue under multi-year customer agreements, announcing its ambition to build a global AI Token Grid, a worldwide network of inference capabilities.
A pioneer in AI Token Grid technology, Blackfuel is building a network of clusters, each dedicated to a specific accelerator architecture and connected by a common software and commercial layer. This network is designed to transform dedicated computing power into measurable and bookable inference output.
In collaboration with AMD and NTT DATA, Blackfuel is developing its inference capabilities operated in Europe, starting with a GPU cluster currently being delivered to the continent.
“The cloud was built around virtual machines. The next infrastructure layer will be built around tokens. We are developing Blackfuel to become one of the world’s leading AI computing companies, starting with our first deployments in Europe. This means bringing together infrastructure, software, and capital at an industrial scale so that customers can buy intelligence with the same predictability they have when buying electricity.” –
Charles Kantor, Blackfuel Co-founder
Building the infrastructure for AI in production
Blackfuel develops an integrated compute platform, encompassing everything from dedicated accelerator infrastructure to the software and services that make it productive. The company combines large-scale deployments for its customers with a broader inference offering for model developers and enterprises, including AI-native ones.
Its first deployment dedicated exclusively to inference, located in Barcelona, Spain, will be hosted by Digital Realty and will be based on liquid-cooled Dell PowerRack systems, equipped with Dell PowerEdge XE9785L servers featuring AMD Instinct™ MI355X i accelerators . The Dell AI Data Platform (Dell PowerScale) and Dell PowerStore will provide the foundation of the data infrastructure, with deployment and integration support provided by both Dell and NTT DATA.
Blackfuel is advancing plans to deploy additional clusters in Spain, Finland and France.
Blackfuel’s offering covers dedicated GPU clusters, managed inference, and reserved inference capacity. Customers can retain control of their compute environment or entrust Blackfuel with managing the serving layer, according to agreed-upon workload and performance requirements.
In parallel with its customer-facing deployments, Blackfuel is developing a separate inference pool for its Token Platform, with support from launch for certain models from Meta, Mistral, H Company, DeepSeek, Qwen, GLM, and Moonshot AI’s Kimi. Programming assistants and agentic applications are priority workloads.
Through a single enterprise interface, customers can select models and reserve inference capacity based on defined requirements for throughput, latency, and data location, without having to manage the underlying infrastructure. Reserved capacity offerings are specified model by model according to agreed-upon terms of service.
Split GPU capacity for production workloads
Blackfuel introduces fractional GPU capacity via logical GPUs: service units that allow customers to reserve a portion of a model’s inference capacity for a fixed monthly fee, instead of leasing a full multi-GPU deployment. Commitments specify the selected model, input and output token quotas, and applicable service levels, sized according to the customer’s workload. Customers can reserve a fraction of the inference capacity of a model replica. Blackfuel can scale the physical hardware and serving configuration while maintaining the agreed-upon capacity and service levels.
Optimizing the number of tokens per megawatt using software
Blackfuel’s inference stack combines open-source serving technologies, including vLLM and SGLang, with proprietary routing, allocation, usage measurement, and business control software. For its AMD-based infrastructure, the stack leverages AMD ROCm™ software. Customers can choose between pay-as-you-go serverless access or reserved capacity under defined service conditions. These offerings support model routing, organization- and project-level controls, observability, and cost tracking.
Beyond orchestration, Blackfuel is developing an agent-driven process to continuously create, evaluate, and optimize inference runtimes tailored to each model, accelerator architecture, and workload. Leveraging production workloads, this approach is designed to reduce inference costs and increase the number of useful tokens per megawatt while meeting defined quality and latency requirements, and extending the economic viability of successive generations of hardware.
“The opportunity isn’t simply about running models on more chips. It’s about continuously improving how each model runs on every architecture. We’re developing software agents to build, test, and refine these runtimes, with the goal of improving the economics of inference for our customers today and extending the productive lifespan of their infrastructure.” –
Dali Kilani, Co-founder of Blackfuel
Blackfuel is developing its platform to support multiple accelerator architectures and hardware generations, selecting systems based on workload requirements, performance, availability, and lifetime cost-effectiveness. Each architecture will have its own optimized runtime, connected by a common layer of services and business management. The goal is to maintain the hardware’s commercial viability through continuous software improvement. This common layer is designed to make capacity commercially transferable from site to site, customer to customer, and model to model, within the limits of each deployment and customer contract.
Linking customer demand to industrial expansion
Blackfuel brings together customer demand, data center capacity, accelerator systems, operating partners, and project and equipment financing within a coordinated deployment model. Its software connects this infrastructure to dedicated compute and contracted inference services.
This model is designed to make inference production financially viable through contracted token provisioning and recurring cash flows. Blackfuel’s software is designed to improve infrastructure profitability 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-upon service levels while extending the commercial lifespan of the underlying hardware. Starting with its European deployments, the company intends to replicate this deployment model globally.
“ Europe has the electricity and the researchers. What it needs now is to own more of its own AI capabilities, transforming what is often an operating cost paid abroad into a fixed asset on European balance sheets. We are neither an inference company that incidentally owns hardware, nor a lessor that incidentally sells tokens. We first secure demand through contracts, then build the corresponding capacity. Every GPU not covered by a contract must be made profitable. This order is what makes the asset financeable.”
Xavier Fischer, co-founder of Blackfuel
Sovereign AI on an industrial scale
Blackfuel’s ambition is to make contract-based token production as financially viable as contracted electricity capacity. Europe has the necessary engineering talent and energy resources. The challenge lies in mobilizing capital at scale to transform these assets into AI capacity that it operates itself, rather than simply leasing it from elsewhere.
Blackfuel’s approach to AI sovereignty combines infrastructure operated in Europe with a service model designed around customer control over model selection, deployment location, and capacity commitments. The goal is to give businesses greater control over where inference is performed and how it is delivered, with deployment and support terms tailored to each customer’s regional needs. The ambition is to achieve global reach while maintaining local control, with sovereignty requirements integrated into the technical, operational, and contractual design of every deployment.
The company believes its competitive advantage lies in combining this regional control with industrial-scale capability, global technology partnerships, and software that improves the profitability of compute resources. Dedicated clusters meet large-scale commitments; fractional capacity and the Token Platform bring the same operational expertise to a broader range of businesses. European deployments form the foundation of this expansion.
Partner statements
AMD
“Businesses 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 open AMD ROCm™ 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 collaborating with Blackfuel, which is integrating AMD technology into its initial deployment and expanding access to high-performance AI infrastructure.”
Stephanie Dismore, Senior Vice President EMEA, AMD
Dell Technologies
“Companies are moving from AI ambitions to tangible results. This evolution demands an 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 required today.”
Adrian McDonald, President EMEA of Dell Technologies
Digital Realty
“AI capabilities require a solid foundation: power, cooling, connectivity, and scalability. Blackfuel’s inference deployment in Barcelona provides these conditions to support an AI infrastructure designed to meet long-term customer demand.” –
Paula Cogan, General Manager EMEA, Digital Realty
NTT DATA
“As a sovereign end-to-end AI infrastructure integrator, NTT DATA combines compute, storage, networking, cybersecurity, operations, and managed services to deliver production-ready AI platforms. Our collaboration with Blackfuel exemplifies how we help our customers deploy and scale secure and resilient AI infrastructure at industrial scale.”
Anne-Sophie Lotgering, Managing Director Europe, NTT DATA
About Blackfuel
Blackfuel is an AI-native neo-cloud, designed for large-scale inference, co-founded by Charles Kantor, Xavier Fischer, and Dali Kilani. Its founding team brings cutting-edge experience in AI, model quantification, and cloud software. The company combines a dedicated accelerator infrastructure, proprietary software, and long-term commercial commitments to build and operate compute capacity for large-scale AI. Its platform addresses the need for dedicated capacity and managed inference, with European deployments forming the foundation for its international expansion.
Site Web : www.blackfuel.ai
The agreement with Dell Technologies was signed in the second quarter of Dell Technologies’ fiscal year 2027.
The text of this press release, being a translation, should not be considered official in any way. The only authoritative version of the press release is the original language version. The translation should always be compared to the source text, which remains the authoritative version.
Source link:https://www.businesswire.com/
