TensorX raises €8M for European AI inference on Nvidia Blackwell
▼ Summary
– TensorX raised €8 million to purchase Nvidia B300 GPUs and expand a GDPR-compliant AI inference platform for regulated European industries like banks, hospitals, and law firms.
– The platform runs on dedicated Nvidia hardware in European data centres, ensuring customer data never leaves Europe and is not retained or used for training.
– TensorX was founded to address the need for data sovereignty, as US-hosted services are subject to US law, which can disrupt access for European firms.
– Demand is driven by research showing 62% of European organizations seek sovereign AI solutions, with projections of $144.6 billion in European AI spending by 2028.
– The company’s “sovereign” offering covers data and jurisdiction, not hardware, as it relies on Nvidia chips sourced through Dell, reflecting the current market reality.
Irish startup TensorX has secured €8 million to purchase Nvidia Blackwell GPUs, including the latest B300 chips, as it expands a GDPR-compliant AI inference platform designed for regulated industries like banking, healthcare, and legal services that cannot risk sending sensitive data abroad.
The funding is earmarked for hardware rather than hiring. TensorX operates private AI inference on dedicated Nvidia infrastructure housed entirely within Europe, currently from data centers in Dublin and Helsinki. This allows customers to deploy open-weight models while guaranteeing that their data never leaves the continent. The company emphasizes that no data is retained, reused, or fed back into model training. The platform supports over 33 open-weight models and features an OpenAI-compatible API, enabling developers to migrate with minimal code adjustments.
The concept emerged from a persistent frustration. Founder Shane Morton, with a background in fintech, repeatedly heard from businesses eager to adopt AI but requiring certainty that their data would remain under European jurisdiction. For a bank, hospital, or law firm, data residency is not a preference but a regulatory mandate. And an American-hosted service, regardless of where its servers physically sit, struggles to satisfy this due to the reach of US law over US-owned providers.
Recent events have sharpened this argument. TensorX cites the “Anthropic Fable 5 situation” as a stark example of how quickly trust assumptions in AI infrastructure can shift. This refers to the US government’s order, earlier in June, that Anthropic suspend access to its most capable model on national-security grounds. For European firms watching a top model be pulled by a foreign government overnight, the abstract risk of relying on someone else’s stack suddenly felt concrete.
The demand signals TensorX points to are not its own. It references Accenture research showing that 62% of European organizations are seeking sovereign AI solutions, a figure that rises among Irish and German firms and is even higher in banking. It also cites a Gartner forecast predicting that over three-quarters of European and Middle Eastern enterprises will move workloads into geopolitically lower-risk arrangements by 2030. The company further notes projections that European AI spending will reach roughly $144.6 billion by 2028. The thesis is clear: the appetite exists, but the supply of compliant compute has not caught up.
Whether a “sovereign” label truly settles the question is itself debated. TNW has argued that the rented-GPU model can reinforce the illusion of European AI sovereignty when the underlying chips, designs, and supply chains still run through American and Asian vendors. TensorX is a member of the Nvidia Inception programme and sources its hardware through Dell, meaning its sovereign offering is built on a decidedly non-European silicon stack. This is less a contradiction than the prevailing shape of the market: Nvidia’s sprawling equity bets and its hold on high-end accelerators mean almost every European sovereignty play, TensorX included, runs on American silicon at the bottom of the stack. The sovereignty on offer is over data and jurisdiction, not over the hardware itself.
On the commercial side, TensorX claims it is already generating revenue, with customers across finance, healthcare, and legal services. Named clients include APEX, TradeLocker, and Cor Prime, and developer demand flows in through the aggregator OpenRouter. These figures are company-stated and unaudited. The financing comes from Darius Cubed Ventures, which has committed €6.5 million toward Nvidia hardware through Dell, with €1.5 million delivered and a further €5 million on order. The company says it is in advanced talks for additional financing.
The expansion plan stretches beyond the two existing sites, with GPU capacity planned across Ireland, the UK, Germany, France, and the Nordics, ahead of the EU AI Act’s tightening compliance rules for regulated sectors. Demand, TensorX says, is already outpacing supply.
(Source: The Next Web)
