TensorX Is Betting Europe’s AI Race Comes Down to Who Owns the GPUs

Dublin’s TensorX has committed €8 million to Nvidia Blackwell GPUs, with roughly €100 million more financed and coming. The wager underneath it: Europe’s AI race turns on who owns the chips. TensorX rents that hardware back out as private, zero-retention AI inference to banks, hospitals and law firms barred from sending data outside EU borders.

A financing line worth up to $1 billion, arranged with a Solana-based crypto lender, sits behind that plan. It backs an asset class nobody has depreciated through a full ownership cycle yet, which makes it an unusual bet layered on top of an already unproven one.

A Fintech Founder’s Data Problem Becomes a GPU Company

Shane Morton built and sold financial trading software before buying ICT Services, a Dublin data-centre infrastructure company with two decades of hardware and supply-chain experience. Running a portfolio of fintech firms afterward, he kept hearing the same complaint from operators who wanted to adopt AI.

They needed proof their data would never leave European jurisdiction first.

“We realised there was no viable way for many enterprises to adopt AI without a sovereign, zero-data-retention solution,” said Tim Grant, TensorX’s executive chairman. Grant and Morton built TensorX to fuse ICT’s hardware relationships with software that keeps every prompt and response inside EU infrastructure. The company went public with the plan through an €8 million seed round led by Darius Cubed Ventures, earmarked for Nvidia Blackwell hardware.

Inside TensorX’s Sovereign Inference Stack

The mechanics are straightforward. TensorX buys physical GPUs, tunes them for the AI models enterprises already want to run, then sells access as private inference with nothing logged or reused for training.

“We buy the GPUs, optimise them for today’s leading AI models, and sell that capability to customers who need complete control over their data,” Grant said.

The platform now supports more than 33 models through an OpenAI-compatible API, so engineering teams can switch away from a US cloud provider without rewriting their code. Revenue already comes from three distinct groups:

• Regulated enterprises in finance, healthcare and legal services, where GDPR and the EU AI Act increasingly dictate where AI workloads can run
• AI marketplaces, including OpenRouter, which route developer demand onto sovereign GPU capacity
• Software vendors building their own AI products on TensorX infrastructure, among them APEX:E3, TradeLocker and Cor Prime

Each group is paying for the same underlying promise: a prompt that never crosses a border.

GPUs, Money and Power Choke the Entire Industry

Grant frames the whole sector around three chokepoints: getting hold of GPUs, financing them, and finding enough electricity to run them.

Blackwell scarcity isn’t unique to TensorX. Demand for AI compute is straining supply across the continent, with 35 new AI supercomputers now under development continent-wide pulling on the same allocation queues. The squeeze isn’t limited to GPUs either. SK Hynix’s own chief executive has warned a memory shortage will outlast its record IPO, a sign the crunch runs across the wider AI hardware supply chain.

Power is the more local constraint. Ireland’s data centres alone consumed 22% of the country’s metered electricity in 2024, and Grant says nobody in the industry can forecast capacity even a year or two out. “We don’t yet know how supply chains will evolve or how much new capacity will come online,” he said. “There is significant effort across the industry to build more sustainable infrastructure, but it’s still early.”

A Crypto Lender Bankrolls the Balance Sheet

“One of our biggest advantages is that we have our own capital. That allows us to move quickly,” Grant said. Own capital only stretches so far, though, and TensorX has looked outside traditional lenders for the rest.

The company partnered with Solstice, a blockchain-based infrastructure firm that finances real-world assets, on a facility capped at up to $1 billion in onchain financing for GPU and data-centre purchases across the EU. Solstice supplies that capital and is launching aiUSX, a yield-bearing asset that lets companies holding cash earmarked for AI lend it into the same infrastructure.

It reflects a financing gap most of the industry shares. Traditional infrastructure lenders exist, Grant said, but “AI hardware is still so new that lenders don’t yet understand depreciation cycles or long-term asset values.” Companies like Anthropic can raise enormous rounds and deal with Nvidia directly. “The rest of us have to navigate traditional supply chains,” he said, which is why TensorX went looking for a new kind of lender rather than a bigger cheque from an old one. The pattern isn’t isolated. A single data-centre raise recently drove a €3.1 billion funding week across Europe’s AI infrastructure sector.

Germany Leads a Demand Signal TensorX Didn’t Expect

TensorX’s strongest market so far isn’t Ireland. It’s Germany.

“We’ve seen a cluster of independent enquiries from German businesses looking specifically for sovereign AI infrastructure,” Grant said. “That wasn’t something we expected, but it reflects the strong focus Germany places on regulation, compliance and data sovereignty.” France, Denmark and the Netherlands have followed, all racing to prepare for the EU AI Act’s compliance deadlines.

The demand shows up in the numbers TensorX points to:

• 62% of European organisations are now seeking sovereign AI solutions, according to Accenture research the company cites, a figure that climbs to 76% inside banking
• 75% of European enterprises are forecast by Gartner to shift AI workloads onto local providers by 2030
• 6.5% is the data-centre vacancy rate Europe is forecast to hit by the end of this year, even after capacity roughly doubled since 2019
• $6.58 billion is roughly Europe’s data-centre GPU market size projected this year

Germany alone held about a quarter of that market last year, and Frankfurt remains one of the continent’s busiest data-centre hubs, alongside London, Amsterdam, Paris and Dublin.

Can Renting Sovereignty Ever Be Sovereign?

Not entirely, say critics who separate physical location from ownership. Data staying on servers inside Dublin or Helsinki satisfies the letter of GDPR’s residency rules, but the chips themselves are still designed, priced and allocated by one foreign supplier, leaving Europe’s AI stack dependent on decisions made far outside its own borders.

The disagreement splits roughly three ways:

• TensorX’s position: dedicated EU-based GPU clusters with zero data retention satisfy GDPR and the EU AI Act, regardless of who manufactures the chip inside them.
• GPUaaS critics’ position: one widely read analysis calls the sovereignty label partly illusory, since leasing Nvidia silicon still leaves Europe reliant on a single foreign vendor’s supply and pricing decisions.
• Competitiveness researchers’ position: a review of Europe’s AI strategy warns heavy compliance costs could push investment abroad, even as sovereignty rules tighten.

Grant’s answer is that TensorX doesn’t need to resolve the philosophical question to keep signing contracts. Regulated customers care about passing an audit, not settling an academic debate.

Dublin and Helsinki Go Live First

TensorX’s immediate job is finishing what it has already ordered. Some systems are live now, with more Blackwell capacity arriving over the coming months in Dublin and Helsinki.

Germany, France, the Nordics and the UK are next, alongside a hiring push to support commercial growth. Longer term, TensorX is exploring partnerships, and possibly building its own sites, to lock down power. It wouldn’t be the first European AI infrastructure bet to stake its future on electricity rather than chips; two recent raises betting on power capacity alone already came from Nscale and Proxima Fusion. Whether €100 million in owned GPUs and a $1 billion crypto-financed line are enough to matter against hyperscale rivals is a bet Grant is prepared to let the market settle.

Frequently Asked Questions

What does zero data retention actually mean at TensorX?

It means prompts and outputs run on dedicated GPU clusters, not shared multi-tenant chips, and nothing is logged, cached or reused to train other customers’ models once a request finishes. That dedicated-hardware detail is what lets regulated firms treat TensorX’s inference as functionally equivalent to running the model on their own servers.

Where does TensorX operate today?

Live capacity runs in Dublin and Helsinki, with sites planned across Germany, France, the Nordics and the UK. The UK sits outside the EU AI Act’s jurisdiction, but TensorX included it anyway because British regulated firms face parallel data-residency expectations from their own regulators.

How does aiUSX differ from a typical crypto yield token?

Instead of generating yield from crypto trading activity, aiUSX ties returns to real loans backing GPU and data-centre purchases. Solstice is launching it with an initial cap of just $5 million, a proof-of-concept size next to the $1 billion facility it could eventually support.

Why does the EU AI Act push healthcare firms toward sovereign AI?

The Act classifies medical AI as high-risk, which triggers strict requirements around transparency, data governance and human oversight. That classification is a big part of why hospitals and healthcare providers sit among TensorX’s earliest regulated customers.

Is Nvidia the only supplier for sovereign GPU clusters in Europe?

For now, mostly yes. Nvidia holds roughly 85% of the AI GPU market, though analysts expect that share to erode as AMD and custom silicon designs scale up and give buyers like TensorX more than one supplier to negotiate with.

Disclaimer: This article is for informational purposes only and does not constitute investment, legal or financial advice. Sovereign AI infrastructure financing and crypto-linked yield products such as aiUSX carry risk, including the risk of loss of capital. Readers should consult a qualified financial professional before making investment decisions. Figures are accurate as of publication.