• Velox has started building a Stockholm data centre that is expected to bring its first 2MW of AI compute capacity online in early 2027
  • I/ONX plans to install 1,800 FuriosaAI RNGD accelerators in the first phase and more than 7,000 across later expansion phases

The fact

FuriosaAI said on 4 August that it is working with Swedish infrastructure developer Velox and AI cloud operator I/ONX HPC on a data centre in Stockholm. Velox has broken ground, and the first 2MW of AI compute capacity is expected online in early 2027. The partners plan to add another 8MW later that year and eventually expand the facility to 15MW.

I/ONX plans to install about 1,800 FuriosaAI RNGD accelerators in the first phase and more than 7,000 across the wider development. The accelerators will operate alongside GPUs on I/ONX’s Symphony SixtyFour platform. Velox will develop and operate the facility, I/ONX will integrate the systems and provide the infrastructure service, and FuriosaAI will supply the accelerators and supporting software.

FuriosaAI says each RNGD accelerator has a thermal design power of 180W. Technical validation, software integration and supply planning are under way. The companies have not disclosed customer contracts, utilisation commitments, service prices or performance results for the Stockholm facility.

The assessment

Construction and accelerator integration are separate parts of the project. Velox must complete and energise the first 2MW, while I/ONX and FuriosaAI must turn the installed hardware into a service that can run customer workloads. Expansion beyond the first phase depends on both tasks being completed. The 180W figure covers each accelerator, not the full system. Servers, processors, memory, storage, networking, power conversion and cooling will also use the facility’s capacity. The first phase will reveal how much usable compute I/ONX can provide from 2MW and whether it can operate FuriosaAI accelerators alongside GPUs on the same platform.

For BTW readers, the first 1,800 accelerators matter more than the planned fleet of more than 7,000. Until the initial capacity is online and running customer workloads, the larger accelerator count and 15MW target remain expansion plans rather than available infrastructure.

What to watch

The first 2MW is expected online in early 2027, but I/ONX has not named any customers or workloads. Installed accelerator numbers, supported models and utilisation data will clarify what the initial phase delivers. The additional 8MW will remain dependent on further construction, hardware supply and customer use.