Scaleway Picks AMD for Sweden, Nvidia for Denmark

 

By TechPulse USA · Paris, 13 October 2026

Tuesday produced the strangest tech announcement of the European autumn, and it came with a French accent. Scaleway — the cloud arm of Xavier Niel's Iliad empire — walked onto the stage of its own annual AI conference in Paris and refused to pick a side in the GPU war. Instead, in the space of a few hours, it signed up with both sides: an AMD-powered AI supercomputer for Sweden, and an Nvidia-powered AI computing site for Copenhagen. Nvidia and AMD spend billions trying to destroy each other, and now one French cloud company has made itself a customer of both on the same day. Europe's sovereign-AI project just got a lot more interesting — and a lot more complicated.

The announcement, in one breath

Photo: File photo — the Summit supercomputer at Oak Ridge National Laboratory, Tennessee. Not Scaleway's machine, but the scale of a modern AI supercomputer hall. Credit: Carlos Jones / ORNL, Wikimedia Commons, CC BY 2.0.

Let's separate what's proven from what's claimed, because this story has two of everything. The proven part: Scaleway said Tuesday at its annual AI conference in Paris that it will build an AI supercomputer in Sweden — its first machine outside France, and the first AMD-based AI supercomputer for commercial use in Europe. It will be open to startups, AI companies and businesses across the continent. Hours later came the second shoe: a new AI computing site in Copenhagen running Nvidia HGX B300 systems — Nvidia's Blackwell Ultra line. We re-read the announcement twice to make sure we hadn't misread it. AMD in Sweden. Nvidia in Copenhagen. Same Tuesday, same stage.

The claimed part is the strategy, and we'll get to that. First, the silicon.

Sweden's machine: AMD's big commercial debut in Europe

The Swedish system runs on AMD's Instinct MI355X — the chip AMD built as its answer to Nvidia's Blackwell line. The specs are genuinely serious: 288GB of HBM3E memory, 8 TB/s of bandwidth, CDNA 4 architecture on TSMC's 3nm process, launched mid-2025, and up to 1400W per chip with direct liquid cooling. On paper, this is a flagship part, and Europe has seen it before — AMD chips already power European research supercomputers, including LUMI in Finland, and both Oracle and Vultr rent AMD chips to European customers. But nobody has yet sold an AMD-based AI supercomputer as a commercial product on European soil. Scaleway is first.

AMD's Stephanie Dismore, the company's SVP for EMEA, said Scaleway "is expanding the choice of AI infrastructure available to customers across Europe, through AMD Instinct GPUs and our open ROCm software ecosystem." That quote does a lot of heavy lifting, because the software is the whole argument. ROCm is open-source and runs AI programming tools that also work on Nvidia chips; CUDA, Nvidia's software layer, is proprietary and locks applications to Nvidia hardware. AMD's pitch has always been: why rent the walled garden when the open field is cheaper? For European companies allergic to lock-in — and these days, that is a lot of European companies — that pitch lands.

French industrial announcements have a genre all their own: the partner roll-call, five company names deep, read out like film credits. The Swedish build is assembled and maintained by Bull, the French firm that already collaborates with Scaleway in the AION consortium; a Bull representative said the project would "enhance Europe's AI capabilities." Servers come from Giga Computing, networking from HPE, storage from VAST Data, and the cooling liquid from Motul. It is a deliberately, almost theatrically European supply chain — and that is not an accident.

Sweden was chosen, Scaleway says, for its largely low-carbon electricity and its AI sector. The groundwork was laid in May, when Scaleway entered Sweden through a deal with telecom operator Tele2 to sell cloud and AI services there. Stefan Trampus, Tele2's B2B head, said the supercomputer "sends a clear signal" about Sweden's attractiveness for AI infrastructure. He is not wrong; Nordic grids and Nordic cold are doing more for AI's geography than any industrial policy document.

Copenhagen's machine: the Nvidia half of the double play

Then Copenhagen. The Danish site runs Nvidia HGX B300 systems — the Blackwell Ultra line, the current apex of the AI-hardware food chain — and it is pitched at the same audience: European businesses that want their AI hosted under EU law. "Access to high-performance computing for AI is becoming a strategic priority for European organizations," CEO Damien Lucas said in a statement, in the practiced cadence of a man announcing his second capital project of the morning.

Copenhagen earned its place for three reasons, all of them practical: a strong tech sector, largely low-carbon electricity, and a cool climate that cuts the energy cost of cooling the servers. Scaleway already serves Danish customers there, including the City of Copenhagen and the AI firm Corti. No one had to be introduced to anyone. This is expansion, not exploration.

And here is where the story gets genuinely clever — or genuinely incoherent, depending on your read.

Photo: File photo — server racks in a data centre. Scaleway's twin Nordic builds are one move in a much larger capital flood into AI infrastructure. Credit: Wil Weterings, public domain, via Wikimedia Commons.

Why playing both sides is smarter than it looks

The obvious take is that Scaleway couldn't choose, so it didn't. The better take is that it refused the choice as a product strategy. Scaleway is Iliad's cloud arm, positioned explicitly as the sovereign European alternative to the US hyperscalers. Its customers are not shopping for chips; they are shopping for jurisdiction. European businesses — and especially European public-sector and defence-adjacent workloads — increasingly want AI hosted under EU law, beyond the reach of instruments like the US CLOUD Act. Cryptobriefing reported back in July that Airbus had picked Scaleway for AI and defence cloud work, precisely the kind of workload that cannot touch infrastructure sitting under foreign jurisdiction. When your buyer cares about whose courts can subpoena the data, you sell legal shelter with a side of GPUs, and you stock both GPU brands the way a serious shop stocks both Coke and Pepsi.

There is also a supply argument that nobody at the conference said out loud but everyone in the room understood. Nvidia's newest silicon is the most allocated, most back-ordered hardware on Earth; betting a continent-scale sovereign-cloud pitch on a single vendor's delivery schedule would be, frankly, colossally reckless. AMD gives Scaleway negotiating leverage, an open software story, and a hedge — the three things every infrastructure buyer on the planet currently wants.

The ROCm question Scaleway can't dodge

Now the stance, because this column takes them: the AMD half of this announcement is the half that has to prove itself, and the proving will be harder than the press releases suggest. ROCm's openness is a beautiful argument in a conference hall and a much spikier one in production. Nvidia's CUDA moat isn't just software; it is a decade of tooling, libraries, Stack Overflow answers and tired DevOps engineers who know exactly which driver version works at 3 a.m. on a Sunday. An open ecosystem that "runs tools that also work on Nvidia chips" is, read uncharitably, an ecosystem whose best selling point is compatibility with the incumbent. Scaleway is betting that price, sovereignty and choice outweigh that friction for European buyers. It's a real bet — not a botched one, not a shoddy one, but a genuine commercial risk, and the company deserves credit for taking it in public rather than in a lab.

What would change our mind fast: named enterprise customers actually training frontier models on the Swedish machine, at scale, within a year. What would confirm the sceptics: the machine fills up with inference workloads and academic grants while the serious training money keeps flowing to Nvidia silicon — including, awkwardly, to Scaleway's own Copenhagen site.

What to watch next

Three things, in order. First, timelines: announcements are cheap, megawatts are not, and we will be watching for groundbreakings, power contracts and actual customer names on both sites. Second, the customer split: if European public-sector money flows to the AMD machine while the private sector flocks to Nvidia in Copenhagen, Scaleway has accidentally built the perfect natural experiment in sovereign AI procurement. Third, the money behind all of it: capital is flooding into AI infrastructure at a pace that makes even a two-country supercomputer spree look routine — from SWI's €800M raise to turn FTX's old bitcoin-mining sites into AI capacity, to Bridge's $6.6B debt bid for its Thailand expansion, to the secretive 15GW, $530B compute sprint now reshaping the whole industry's math. Scaleway's Nordic double-play is one move in a much larger game, and the board is getting crowded.

The bottom line: a French cloud company just told Europe it doesn't have to choose between AMD and Nvidia to choose Europe. That's a good line. Now it has to build two supercomputers to back it up.

Sources

Reuters (primary reporting on both announcements, 13 October 2026); CNA and the Economic Times as relays; Wikipedia for Scaleway/Iliad corporate background; Wccftech and TweakTown for AMD Instinct MI355X specifications; Cryptobriefing (July 2026) for the Airbus–Scaleway defence-cloud report. Quotes from Stephanie Dismore (AMD), Damien Lucas (Scaleway), Stefan Trampus (Tele2) and a Bull representative are as reported by Reuters and CNA.

By TechPulse USA
TechPulse USA is an independent tech news publication covering the global AI race with a US-first lens. We verify every fact we print, link our sources, and separate what is proven from what is claimed.

Scaleway's Nordic double-play: your questions answered

What exactly did Scaleway announce? At its AI-Pulse conference in Paris on Tuesday, 13 October, Scaleway made two announcements within hours: an AMD Instinct MI355X-powered AI supercomputer for Sweden — its first machine outside France and, the company says, the first commercial AMD-based AI supercomputer on European soil — and an Nvidia HGX B300 "GPU Zone" computing site in Copenhagen.

Why pick AMD for Sweden and Nvidia for Denmark? It reads less like indecision and more like a product strategy. Scaleway sells jurisdiction, not chips: European businesses — especially public-sector and defence-adjacent workloads — want AI hosted under EU law, beyond the reach of instruments like the US CLOUD Act. Stocking both GPU brands is how a serious shop covers the market, and AMD gives Scaleway negotiating leverage plus a hedge against Nvidia's back-ordered delivery schedule.

What is "sovereign AI," and why does it matter here? Sovereign AI is the idea that a country's AI workloads should run on infrastructure governed by its own laws and powered by its own grid — part of Europe's sovereign-tech push. Sweden was chosen for low-carbon electricity and its AI sector; Copenhagen for its tech sector, clean power and a cool climate that cuts cooling costs. The Swedish supply chain is deliberately European, assembled by the French assembler Bull.

When do the machines come online? Scaleway has not disclosed timelines, capacities or customer names for either site — announcements are cheap, megawatts are not. Watch for groundbreakings, power contracts and named enterprise customers. The broader context is the AMD-vs-Nvidia share battle now reshaping the whole industry's geography.