Blackfuel launches with a Barcelona inference build as cloud revenue growth meets hardware

Blackfuel launched on 30 September, according to DataCenterDynamics. Its first deployment goes into Digital Realty's BCN1 site in Barcelona, a colocation campus with 161,500 sq ft (15,000 sqm) of space that opened earlier this year. The company is small, European and deliberately unglamorous.

The Barcelona build runs liquid-cooled Dell PowerRack systems carrying Dell PowerEdge XE9785L servers with AMD Instinct MI355X accelerators. Blackfuel says further deployments are planned elsewhere in Spain, plus Finland and France.

That is a capex-heavy way to enter a market where revenue growth is being measured in tokens, megawatts and backlog rather than seats.

The commercial model is where Blackfuel tries to differentiate. It will sell dedicated GPU clusters, managed inference and reserved serving capacity, plus an "inference pool" for its token platform. It is also introducing fractional GPU capacity, which lets a customer reserve a share of a model's serving capacity for a fixed monthly fee instead of renting an entire multi-GPU deployment. The platform is intended to support multiple accelerator architectures and hardware generations, so customers can pick systems by workload.

Co-founder Xavier Fischer framed the pitch around ownership and financing rather than raw compute. "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," he said in the launch material. "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."

Digital Realty CTO Chris Sharp, quoted in the same announcement, tied the deal to the landlord's ServiceFabric interconnection platform and called Barcelona "the first proof point." Blackfuel was founded by Fischer, Dali Kilani and Charles Kantor.

Cloud growth is real, but the delivery model is shifting

The timing lands inside a broader reallocation of enterprise computing. Arelion's report, published on 28 September, found that 58% of senior business leaders have delayed, scaled back or added safeguards around strategic initiatives in the past two years because of concerns about their network provider. AI and data-driven projects were the most affected, cited by 49% of respondents who reported disruption, ahead of security and compliance at 44%, digital transformation at 42% and cloud migrations at 38%.

That is a demand-side caveat, not a demand collapse. Data Center Knowledge reported on 21 September that enterprises are increasingly choosing colocation for AI inference because inference performs best at cabinet densities most corporate data centers cannot support: 35 kW cabinets with air cooling, and 70-150 kW cabinets with optional liquid cooling. The same piece cites VMware's Private Cloud Outlook 2026 finding that 83% of enterprises have completed or are planning to repatriate workloads from the public cloud, and Foundry's 2026 Cloud Computing Study showing 74% of enterprises accelerated cloud migrations last year.

Read together, those numbers describe a market that is still growing but no longer routing every workload through a hyperscaler. Colocation and hybrid arrangements are capturing the predictable, data-adjacent and latency-sensitive portion, which is precisely the segment Blackfuel says it is building for.

The adjacent failures matter too. On 1 October, The Register reported that a Microsoft infrastructure servicing event disrupted Azure ExpressRoute Gateway, VPN Gateway and Azure VMware service across 18 regions, including West Europe, North Europe, UK South and France Central. Microsoft later said it had mitigated the incident and paused the servicing activity associated with its onset. For an operator selling reserved inference capacity in Europe, that is useful marketing material, though the dossier does not show Blackfuel using it.

Observability vendors are chasing the same agentic workloads. InfoQ reported on 29 September that Amazon CloudWatch Omni, an AI-first observability platform, captures end-to-end traces, evaluates correctness, coherence, retrieval and tool selection, and supports frameworks including LangChain, LangGraph, CrewAI, the OpenAI SDK, Strands and Vercel AI SDK. On the same day, InfoQ reported Cloudflare's native support for the HTTP Vary header inside Cache Rules, aimed at preventing cache thrashing after an audit of more than 120 million HTTP responses found nearly 3,000 origins varying on four or more request headers.

Public sector and defense money is moving to approved clouds

One revenue channel that keeps compounding is government work. Amazon said on 24 September that AWS became the first cloud provider approved for NATO RESTRICTED workloads across all member nations, after documenting compliance against NATO's D32 directive and being evaluated by Spain's National Cryptographic Centre. AWS has 15 Regions in NATO member nations, seven in mainland Europe. NATO's Dylan Browne and AWS's David Appel both framed the approval as a multi-year effort.

Space and defense budgets are following a similar curve. Novaspace's Satellite Communications for Defense and Security, 2nd Edition, reported on 29 September that commercial defense satcom service revenues are projected to exceed $22.6 billion by 2035, with NGSO revenues reaching $21.3 billion on a 34% CAGR, while GEO revenues fall from $2.3 billion in 2025 to $1.3 billion in 2035 even as GEO capacity demand triples.

Smaller suppliers are staffing up against that demand. Agile Space Industries said on 28 September that Jason Wright will join as President in early October, as the company expects to reach roughly five times its 2022 revenue by the end of 2026 and now employs more than 165 people across Colorado, Pennsylvania, Oklahoma and Washington state.

For Blackfuel, the immediate test is narrower than any of these markets. It has one site, one hardware generation and a promise to contract demand before building capacity. Whether that discipline holds once competitors start signing multi-year reservations is the part no launch statement can answer.