Nvidia turmoil and Microsoft outage put tech sovereignty back in focus

If tech sovereignty wasn’t already smack-bang at the top of the political agenda, it must be now – bound up with geopolitics and foreign policy, domestic security and political stability, and questions about the future of work and society. In a matter of days, critical failures at two of the pillar firms in the supposedly stable and trusted ‘western’ AI tech stack – both American – have exposed uncomfortable truths about the ball-of-snakes corporate monopoly at the heart of the all-consuming AI industry. Nvidia is being scrutinised following allegations linking its leader – Jensen Huang, the pied-piper for the whole AI song-and-dance – to an alleged Chinese spy. Microsoft has suffered a major outage – Thursday into Friday, unresolved at press – that has knocked out Microsoft 365 services in dozens of countries. There is no suggestion that Nvidia was robbed or Microsoft was hacked.

And neither episode, on its own, makes a case to turn away from US tech. (Nvidia and Microsoft are more than a match for their European equivalents, if they exist, on security and resilience.) But both illustrate the risk of placing so much faith, and so much critical infrastructure, in the hands of such a small number of service providers. There are three strands to the sovereignty debate, vexing Europe most notably: where the AI runs, who controls it, and what it runs on. Europe, in an existential wrangle about its position in the world, and a late charge for some kind of impact in the AI race, is making progress on the first two. European telcos have placed themselves into the middle, as good corporate citizens and trusted guardians of sovereign data. They are variously building AI ‘factories’ as alternatives to the likes of Microsoft, while also working with them on separate cloud initiatives.

But the third strand is harder, verging on impossible, and eludes the telco set completely. Deutsche Telekom is the best example: it is putting another €250m into its AI factory in Munich, taking capacity to 20MW with Nvidia’s latest Blackwell systems – after the first phase was booked-out in seven months. Otherwise, its so-called Deutschland Stack is largely true to its name: made in Germany. Paderborn-headquartered Polarise is responsible for the design, build, and operation of the data-centre infrastructure; SAP provides the software platform; Siemens, among others, is handling industrial AI applications. The site is host to the European SOOFI project, geared to develop independent AI foundation models for Industry 4.0. There are other telco examples, organised differently: Orange, Fastweb, Swisscom, Telenor, and Telefónica are all building sovereign capacity. The commercial opportunity is clear.

The third-strand technological independence is much less so. They might have control of data residency, and maybe even the ‘kill switch’, but most are using the same foreign chips from Nvidia – whose leader is embroiled in a political scandal with a suspected Chinese spy. Nokia wants to put the same into European RAN systems – which are, meanwhile and separately, under pressure from EU security policy about ‘high-risk’ Chinese suppliers. Even SK Telecom’s work with domestic chipmaker Rebellions to develop sovereign AI inference infrastructure can’t avoid Nvidia: its proposed system will combine Rebellions’ own NPUs with Nvidia’s GPUs, and also use the latter’s NVLink Fusion interconnect and MGX rack architecture. The proposal, still at MoU stage, localises a part of the compute stack, but Nvidia is still central to the design. It shows how tightly wound the global tech industry is, and how conditional sovereignty can be.

There is another Nvidia dependency beyond the hardware, of course: money. As Vish explores in his excellent analysis of its post-Huang position as underwriter of the AI boom, the company’s influence extends heavily into the industry’s financing arrangements. Who funds the GPUs, who guarantees the demand, who carries the risk – if these mystical AI returns don’t materialise? For European operators, sovereignty is therefore not simply a question of where the machines sit, or even who owns them. It is also – and increasingly, given security questions around cornerstone companies, and (new) hyperscaler blackouts – about who supplies the tech, who controls the supplies, and who writes the cheques. Europe is making progress to build its own AI infrastructure. Whether it can exert much real control over the industrial and financial machinery behind it seems like a dicier question.