Irish startup TensorX raises €8 million seed round to build a European “sovereign” AI inference platform using NVIDIA Blackwell GPUs

TensorX, an Irish startup working on “sovereign” artificial intelligence infrastructure for Europe, said on June 24 it has secured an €8 million seed funding round to launch an AI inference platform built around NVIDIA Blackwell GPUs. The company characterized the effort as a bet on European control over AI compute and deployment, positioning the project around inference, the part of AI pipelines where models run to produce outputs for applications.

According to the report, the seed round is led by Darius Cubed Ventures. The funding is intended to support the initial deployment and go-to-market work for TensorX, with the company describing plans to scale GPU usage over time rather than limit the project to a single small proof-of-concept environment.

The same report says TensorX’s planned deployments include NVIDIA Blackwell GPUs, specifically mentioning the latest B300 chips. Blackwell is the GPU architecture NVIDIA has been promoting for data center and AI workloads, and B300 refers to a later-generation variant within that platform. TensorX did not provide additional technical detail in the announcement beyond tying its launch hardware plans to those devices.

TensorX also disclosed a longer-range deployment target, saying it plans to roll out up to €100 million worth of NVIDIA Blackwell GPUs. That figure, presented as a scale-up objective, suggests the company is aiming for capacity growth as it signs new customers and expands the range of inference services it intends to offer. The report did not specify whether that €100 million target is linked to signed purchase commitments, contracted volumes, or a forecasted usage plan contingent on revenue.

The “sovereign” angle places TensorX in a wider European push to retain greater control over sensitive data and AI deployments, where procurement, hosting, and operational boundaries are often treated as policy issues as well as engineering concerns. In this framing, inference infrastructure can be just as significant as training infrastructure, because many enterprise and public-sector use cases involve running models repeatedly against organizational data once the model weights are chosen.

NVIDIA, as the supplier of the proposed GPU stack, stands to benefit indirectly if TensorX’s expansion includes additional systems based on Blackwell GPUs. For NVIDIA, deployments like these reflect the broader pattern that AI hardware demand is increasingly tied to specialized hosting and inference platforms rather than being used only in traditional supercomputing or hyperscaler settings.

Still, important details are not disclosed in the reporting. TensorX did not specify the geographic footprint of its inference operations, the initial customers it is targeting, the model types it will serve, or the deployment approach (on-premises, hosted cloud, or hybrid). The company also did not break out how much of the €8 million round is allocated to hardware procurement versus hiring, systems integration, or operating expenses, nor did it outline expected timelines for reaching the larger €100 million GPU deployment goal.

What to watch next is how TensorX converts its seed round and GPU plans into commercial traction, including any customer announcements, service-level commitments, or partnerships that clarify where the platform will run and what “sovereign” means in practice for data handling and control. Additional disclosures around deployment milestones would also help determine whether the €100 million GPU target is a near-term capacity plan or a longer-horizon objective dependent on contracted demand.