Over the past two years, almost every conversation I have had with European customers about AI has eventually arrived at the same question. Not “how fast is it?” — although performance matters enormously — but “where does this come from, and who stands behind it?”
That shift tells me something important. As organisations move AI out of experimentation and into business-critical production, they are no longer evaluating infrastructure on specifications alone. They want a verifiable answer about origin, manufacturing, and the supply chain behind the systems they depend on. Today I am proud that in July, we introduced a server family built to give exactly that answer: the new PRIMERGY GX AI server family.
Thirty years in the making
The GX family did not appear overnight. It is the latest chapter in a story that began more than 30 years ago, when we introduced the first PRIMERGY servers in 1994. The same engineering lineage produced the K Computer in 2011 and, in 2020, Fugaku — the world’s number one supercomputer. All of it was built at Fujitsu’s Kasashima factory in Japan, the same factory that now manufactures GX.
I find that heritage matters more than ever in the AI era, because it changes what “trust” means. When our teams talk about Made in Japan, we are not describing a label added at the end of a production line. We are describing three decades of manufacturing discipline, quality assurance, and supply-chain transparency, extended into a new class of GPU-native infrastructure.
This is a deliberate design principle, not a marketing reaction to headlines. Sovereign by design. Not a reaction to geopolitics.
PRIMERGY GX2550 M8s
One family, the right platform for every AI workload
We launched GX with two complementary systems, designed for different points on the AI journey.
The PRIMERGY GX2550 M8s brings flexible PCIe GPU acceleration for enterprise AI, inference, visualisation, rendering, media processing, simulation, and HPC. It is built to scale from focused inference to full training environments so organisations can grow their AI ambitions without replacing the infrastructure underneath them.
The PRIMERGY GX2580 M8s is our most powerful AI server, built around the NVIDIA HGX B300 8-GPU architecture for LLM training, generative AI, AI superclusters, and scientific simulation at scale. It is designed for organisations pushing the frontier of what AI can do.
Two platforms, one family: the right system whether you are starting with enterprise inference or building supercluster-scale infrastructure.
More than GPU hardware
Raw compute is only part of the story. What makes GX a foundation rather than a box is everything around it: integration into the proven PRIMERGY ecosystem, lifecycle management through Infrastructure Manager (ISM), the trusted service expertise of our European teams, and consumption-based flexibility.
For me, this is where sovereignty becomes practical. It is not an abstract principle; it is the ability to build and operate AI infrastructure with control over where your systems come from, how they are supported, and how you scale them over time, with less dependency on hyperscalers.
PRIMERGY GX2580 M8s
Build AI on your own terms
Europe is at a decisive moment in how it builds its AI future. The organisations I speak with do not want to choose between performance and trust. With PRIMERGY GX, they no longer have to.
This is a foundation for building AI on your own terms, backed by 30 years of engineering heritage and the people who stand behind it.
Made in Japan. Built for what’s next.
