DT and Nvidia collectively funded the Industrial AI Cloud, Europe’s most formidable sovereign AI facility
In early February 2026, Deutsche Telekom’s CEO Tim Höttges stood inside a gutted financial institution vault in Munich’s Tucherpark district and declared that Europe can do AI. Behind him sat 10,000 Nvidia Blackwell GPUs, wired along with 75 kilometers of fiber optic cable, cooled by river water from the close by Eisbach, and already operating workloads for Siemens, Perplexity, and a handful of robotics startups. The ability had taken six months to construct. It price one billion euros. And it carried a really particular label: Sovereign.
The Industrial AI Cloud is essentially the most formidable sovereign AI facility Europe has constructed. DT and Nvidia collectively funded it. T-Techniques operates it. SAP supplies the enterprise software program. Siemens plugs in its simulation instruments for digital twins. A European analysis consortium known as SOOFI is utilizing it to coach a 100-billion-parameter open-source language mannequin solely on European soil. Over a 3rd of its capability was spoken for at launch, with industrial prospects paying industrial charges. No authorities subsidy required.
All of which raises a query that European policymakers, enterprise patrons, and know-how strategists ought to care about. Is it really sovereign? And what would we even imply by that?
What sovereign really means
The phrase has been stretched to the purpose of uselessness. Hyperscaler advertising and marketing groups use it. Nationwide governments use it. EU commissioners use it. Every means one thing totally different. The outcome has been a discourse the place sovereignty is simple to assert and unattainable to confirm.
That modified in November 2025. On the Summit on European Digital Sovereignty in Berlin, all EU member states signed what grew to become referred to as the Berlin Declaration. It contained the primary formal political definition. Digital sovereignty is the power of the EU and its Member States to behave autonomously and to freely select their very own options, whereas reaping the advantages of collaboration with world companions when potential.
Two issues matter right here. First, the definition facilities on the power to decide on, not the duty to go it alone. Sovereignty just isn’t autarky. The Declaration explicitly preserves area for world partnerships. Second, the accompanying Franco-German commitments demanded sovereignty throughout all the worth chain. Not simply on the knowledge layer, the place GDPR already supplies robust protections, however down by the infrastructure, compute, and silicon layers the place Europe’s dependencies are deepest. Throughout all the worth chain is a really excessive bar.
The European Fee’s Cloud Sovereignty Framework, issued in September 2025, added operational specifics. Knowledge residency necessities. Operational management by EU-based entities. Jurisdictional insulation from extraterritorial laws, notably the U.S. CLOUD Act. Software program portability to stop vendor lock-in.
Taken collectively, these coverage devices counsel a five-layer check for any facility claiming sovereignty.
Knowledge sovereignty. Knowledge stays inside the jurisdiction, ruled by its legal guidelines, processed by its residents, and insulated from international authorized entry.
Operational sovereignty. The ability is run by entities underneath the jurisdiction’s authorized management.
Platform sovereignty. Software program and cloud providers come from entities underneath the jurisdiction’s authorized and operational management.
Compute sovereignty. Processing {hardware} may be sourced, maintained, upgraded, and if obligatory changed by provide chains the jurisdiction can affect or management.
Strategic sovereignty. The jurisdiction retains the power to make autonomous selections concerning the facility’s future with out requiring permission from international entities.
No facility will rating completely on all 5. The Berlin Declaration acknowledges as a lot by framing sovereignty as a capability quite than an absolute state. However the framework sharpens the query. The place does a facility ship real autonomy? The place does it create dependencies? And are these dependencies managed and diversified, or concentrated and single-threaded?
That final distinction is the one which issues most. Good sovereignty just isn’t the usual. Managed dependency is. The actual query is whether or not operators management their important selections, can affect their key relationships, or just rely on exterior events for outcomes they can’t form.
Working the check
DT’s facility opened on February 4, 2026. Almost 10,000 Nvidia Blackwell GPUs. Half an exaFLOP of compute. One billion euros invested. Over a 3rd of capability utilized at launch by paying prospects together with Agile Robots, PhysicsX, Siemens, EY, and Perplexity.
That industrial traction is itself a sovereignty sign. A facility that survives on authorities life assist just isn’t really unbiased. DT proved that European industrial prospects pays for sovereign AI compute at industrial charges. Financial viability is a prerequisite for sustained autonomy.
Right here is how the ability performs towards every layer.
Knowledge sovereignty: Sturdy.
DT operates the ability underneath German knowledge safety regulation, GDPR, and NIS2. Knowledge by no means leaves German soil. T-Techniques, a German subsidiary, manages every thing with EU-based workers.
This issues as a result of the U.S. CLOUD Act provides American regulation enforcement the authorized authority to demand knowledge held by U.S.-controlled corporations no matter the place that knowledge bodily sits. By working by T-Techniques quite than a U.S. hyperscaler, DT removes this vector solely. No American entity has administrative or authorized entry to buyer knowledge or workloads.
Knowledge sovereignty is essentially the most mature layer of Europe’s digital independence, constructed on years of GDPR enforcement and regulatory iteration. It’s also the place the hole between European and American approaches is widest. DT’s facility delivers the actual factor.
Operational sovereignty: Sturdy.
T-Techniques manages infrastructure. Polarise, a European companion, constructed the information heart in a repurposed former Hypovereinsbank facility. Cooling runs on Eisbach river water. Energy comes solely from renewables. Waste warmth will provide the encompassing Tucherpark district. The complete operational chain, from facility administration to incident response, runs by European entities underneath European jurisdiction.
Platform sovereignty: Sturdy, with a coupling threat.
That is the place DT made its most deliberate architectural decisions. The Deutschland Stack combines T-Techniques infrastructure and T Cloud with SAP’s Enterprise Expertise Platform. Siemens contributes its simulation portfolio for digital twins. SOOFI will practice a European language mannequin solely inside the facility.
The platform layer is nearly solely European. Prospects develop on SAP BTP, run Siemens simulation instruments, and work together with T Cloud, all supplied by entities underneath German and EU authorized management. This can be a real full-stack European different to AWS, Azure, or Google Cloud.
However each workload on this platform in the end executes by Nvidia’s CUDA-X libraries and associated frameworks. The platform is European. The runtime beneath it isn’t. This coupling means platform sovereignty just isn’t absolutely unbiased of the compute layer beneath it. If Nvidia materially modified its CUDA licensing phrases or restricted entry to key libraries, the platform layer would really feel the results regardless that no platform-level determination had modified.
DT delivers robust platform sovereignty at present. Its sturdiness is determined by the soundness of what sits beneath it.
Compute sovereignty: The important hole.
Each GPU within the facility is an Nvidia Blackwell chip. Designed in Santa Clara. Manufactured by TSMC in Taiwan. Ruled by Nvidia’s proprietary CUDA software program ecosystem. Topic to U.S. Export Administration Laws.
This creates a triple dependency:
Chip design. Nvidia is the only provider. Europe has no indigenous GPU able to industrial AI workloads at this scale. SiPearl’s Rhea1, developed underneath the European Processor Initiative, is an Arm-based CPU simply reaching the sampling stage, designed for supercomputing quite than GPU-class AI. The EU-funded DARE mission has 240 million euros for RISC-V chiplets however won’t produce industrial {hardware} for years.
Chip manufacturing. Even SiPearl’s European-designed processor is fabricated by TSMC in Taiwan. If Europe designed its personal GPU tomorrow, it might nonetheless rely on a non-European foundry to construct it. The European Chips Act’s purpose of 20 % world semiconductor manufacturing by 2030 stays removed from actuality.
Software program ecosystem. CUDA is a proprietary parallel computing platform with over 20 years of developer tooling and optimized libraries. The diploma of lock-in varies by workload. Pure AI coaching and inference have gotten extra moveable by PyTorch, JAX, and AMD’s maturing ROCm platform. However industrial simulation workloads, the type Siemens and PhysicsX run on this facility, are deeply built-in with CUDA and would take years of engineering emigrate.
The chance that the US would really limit GPU exports to Germany is low. Germany is a NATO ally, deeply embedded in Western protection provide chains, and allied pushback killed the January 2025 AI Diffusion Rule that may have imposed chip quotas on companion nations. However sovereignty is about structural functionality, not present chance. The authorized and administrative equipment to limit GPU provide to any nation now exists inside the U.S. export management framework. Europe’s AI infrastructure operates inside a system the place a international authorities possesses the power to constrain provide, even when exercising it towards a detailed ally stays unlikely.
DT doesn’t ship compute sovereignty. The hole is structural. DT can not shut it by higher execution or smarter procurement. Solely long-term European industrial coverage can deal with it.
Strategic sovereignty: Constrained by compute.
DT can autonomously determine to develop capability, onboard prospects, develop new providers, or shift industrial technique. These selections are absolutely inside European management.
However DT can not autonomously substitute its GPU vendor. It can not diversify its {hardware} provide chain to a European different that doesn’t but exist. It can not improve to next-generation compute with out Nvidia’s cooperation.
DT’s personal roadmap displays this. The subsequent section, an AI Gigafactory developed with Nvidia and Brookfield Asset Administration and competing for EU funding by the 20 billion euro AI gigafactories program, deepens quite than diversifies the Nvidia relationship.
Essentially the most sincere technique to map that is by three tiers. DT controls its knowledge, operations, and platform selections. DT influences its Nvidia relationship by buying quantity and the leverage of being a flagship European buyer. However DT is determined by Nvidia for silicon design, on TSMC for fabrication, and on the U.S. authorities for export authorization. The road between what you management, what you affect, and what you rely on is the truest measure of any sovereignty declare.
What this implies
DT’s AI manufacturing facility delivers real sovereignty on the knowledge, operational, and platform layers, and structural dependency on the compute layer. This isn’t a failure distinctive to DT. It describes each AI facility in Europe and most services globally exterior the US and China.
The query just isn’t whether or not DT ought to have executed higher. Given obtainable know-how, it constructed the strongest sovereign AI facility Europe may have constructed within the time and finances obtainable. The query is what sovereignty can moderately imply in a world the place GPU provide is concentrated amongst a handful of American and nascent Chinese language suppliers.
Three issues observe:
Sovereignty is layered, not binary. Europe can obtain near-complete sovereignty on the higher layers of the stack whereas nonetheless relying on non-European compute {hardware}. For a German producer operating digital twins on DT’s facility as an alternative of AWS, the distinction is important and materials. Knowledge stays in Germany. Operations are managed by German workers. CLOUD Act threat is eradicated. The American GPU beneath doesn’t erase these protections. It qualifies them. Europe ought to declare the sovereignty it has achieved and be exact about the place dependencies stay, quite than overstating one or dismissing the opposite.
Compute sovereignty is a decade-long industrial mission, not a procurement determination. DT can not shut the hole by switching suppliers. There isn’t a European GPU to change to. The trail ahead runs by three parallel tracks. Continued funding in European chip design by SiPearl, DARE, and startups like Axelera AI, accepting that these are five-to-ten yr applications. Diversification of GPU sourcing as AMD, Intel, and rising distributors mature. And funding in open software program frameworks like SYCL, OpenCL, and Triton that scale back CUDA lock-in and make workloads moveable throughout {hardware}. Software program portability is the prerequisite for {hardware} diversification. None of those delivers compute sovereignty quickly. Collectively, over a decade, they may transfer Europe from concentrated dependency to managed diversification. That’s extra sincere and extra achievable than promising full autonomy.
The actual check is what occurs underneath stress. Sovereignty reveals itself underneath strain, not on a traditional Tuesday. What occurs when the GPU provide chain breaks, when export controls shift, when a vendor reprices, when a geopolitical disaster threatens Taiwan’s foundries? DT’s facility has real sovereignty on the layers the place European entities make selections and real vulnerability on the layer the place they don’t. The flexibility to function the information heart, serve prospects, and defend knowledge is sovereign. The flexibility to replenish, improve, and scale the compute {hardware} just isn’t.
Conclusion
Deutsche Telekom’s Industrial AI Cloud is essentially the most substantive sovereign AI facility Europe has constructed. It delivers protections on the knowledge, operational, and platform layers that no U.S. hyperscaler can structurally match. It proves industrial demand for sovereign AI compute. And it positions DT for the EU-funded gigafactory program that may form the subsequent section of European AI infrastructure.
However it additionally attracts a pointy line between sovereignty achieved and sovereignty that is still a dependency. The compute layer, constructed solely on American silicon operating American software program manufactured on a Taiwanese foundry, represents a concentrated reliance that Europe can not resolve by procurement, regulation, or operational excellence alone.
The purpose shouldn’t be good autonomy. It must be managed dependency. Reduce focus at important layers. Construct diversification paths earlier than they’re wanted. Spend money on the open software program and indigenous {hardware} that give Europe choices quite than obligations.
The Berlin Declaration outlined sovereignty as the power to behave autonomously and freely select. DT’s AI manufacturing facility reveals that Europe can act autonomously in the way it manages, operates, and governs AI infrastructure. What it can not but do is freely select who provides the silicon that makes it run.
Closing that hole is the work of the subsequent decade. Defining it truthfully is the work of at present.
