TensorX raises €8M for Nvidia Blackwell B300 GPUs to run sovereign AI inference in Europe

TensorX just raised €8 million to expand a GDPR-compliant “sovereign AI inference” platform, and the money is going where it counts: Nvidia Blackwell GPUs, including the latest B300 chips. The Irish startup’s core move is simple but high-stakes. It wants European organizations to run AI inference without shipping their data abroad, which is exactly the kind of constraint that can stall pilots, delay deployments, and sour internal timelines.

In other words, this is not an AI model training story. It is an inference story, and it is about residency and controls. TensorX’s platform is built for use cases where banks, hospitals, and law firms cannot send their data abroad, but still need AI-powered decision support, document processing, and similar inference-heavy workloads. By purchasing B300 hardware as part of the €8M round, TensorX is positioning itself to scale inference capacity while staying aligned with the compliance posture its customers require.

To understand why this is worth attention, you have to zoom out to how AI is actually adopted in regulated industries. Many organizations can try models in small ways, but scaling inference into production runs into a predictable set of friction points: where data flows, who can access it, how systems are governed, and what happens when auditors ask for explanations that are more “show your work” than “trust us.” For the kinds of customers TensorX is targeting, the question is not only “Does the AI work?” It is also “Can the architecture prove the data stayed in Europe, and can it satisfy GDPR expectations around processing?”

That GDPR angle is the engine behind the “sovereign” framing. The term usually gets thrown around in ways that are vague. Here, the source’s description is specific: TensorX is building AI inference infrastructure that keeps European data inside Europe for organizations that cannot send their data abroad. That matters because it turns AI from a pure technical project into a procurement and compliance project. If your compute vendor or hosting environment is outside your allowed jurisdiction, the model might be the best thing you have ever tested but still unusable for the mission you actually have to deliver.

Now add the hardware decision. TensorX is buying Nvidia Blackwell GPUs, explicitly including the latest B300 chips. That is a clear signal about timing: the startup is not waiting for “eventual” capacity. It is trying to lock in performance and scale now, which becomes critical in inference. Inference systems tend to have steady demand patterns once deployments start, and they often require predictable throughput for user-facing or workflow-integrated applications. In that world, the gap between having adequate GPU capacity and not having it becomes a real bottleneck, not a theoretical one.

There is also a second-order effect for the broader market: this kind of sovereign inference strategy is likely to change what boards and CFOs track. Instead of only budgeting for model development or experimentation, they increasingly need to budget for compliant infrastructure. That means GPU supply planning, cost-per-token or cost-per-inference budgeting, and the operational burden of keeping sensitive data within approved boundaries. For decision-makers, the operational reality is that compliance requirements can force you to build or buy more infrastructure than a standard “send it to the cloud and ship” approach.

And TensorX’s customer focus hints at why those CFO conversations will intensify. Banks, hospitals, and law firms are not the typical early-adopter crowd that can move slowly because they are uncomfortable or skeptical. They move carefully because regulators, clients, and internal risk frameworks demand it. That carefulness can look like “slow adoption,” but it is also how you end up with long-term deployments once the architecture fits. TensorX is trying to get on the right side of that dynamic by making the compliance story part of the product, not a later patch.

Finally, the strategic stake is bigger than one startup’s infrastructure plan. If TensorX can scale a GDPR-aligned inference platform powered by Blackwell B300 GPUs, it strengthens a template that other European AI infrastructure providers can follow: match advanced inference hardware to residency and governance needs, then sell the whole operational package to regulated buyers. For peers building, investing in, or procuring AI systems in Europe, the practical takeaway is that “sovereign inference” is no longer only a policy slogan. It is turning into a budget line item backed by real GPU procurement, including the newest Nvidia chips, at the moment where demand for compliant AI is starting to become enterprise-grade.