A crypto miner became Europe’s most valuable AI startup in two years. Now it is spending $812 million to prove the model scales.


A crypto miner became Europe’s most valuable AI startup in two years. Now it is spending $812 million to prove the model scales.

TL;DR

Nscale, a crypto miner turned $14.6 billion neocloud, is investing €695 million in Portugal to supply 66,000 Nvidia Rubin GPUs to Microsoft’s Start Campus site in Sines. The deal extends Europe’s fastest AI infrastructure buildout, funded by private capital following GPU demand economics rather than by sovereign compute programmes.

Two years ago, Nscale was a cryptocurrency mining operation. On Tuesday, the company announced €695 million ($812 million) in new infrastructure investment in Portugal, an expansion of its partnership with Microsoft that will deliver more than 66,000 Nvidia Rubin GPUs to a site with 1.2 gigawatts of permitted capacity. Nscale is now valued at $14.6 billion after a $2 billion Series C in March, backed by Nvidia, Lenovo, Dell, Citadel, Jane Street, and Nokia. It operates data centres across the UK, Norway, Portugal, Iceland, and the United States, and its CEO has said the company plans to go public in 2026. The trajectory from crypto miner to Europe’s most valuable AI infrastructure company took twenty-four months. The question it raises is whether Europe’s answer to the compute crisis will come not from governments or hyperscalers but from a new class of company that barely existed three years ago.

The deal

The Portugal investment is split between 230 million in shared infrastructure and €465 million for a second 200-megawatt building at the Start Campus data centre site in Sines, a deep-water port town on Portugal’s Atlantic coast. Start Campus is permitted for 1.2 gigawatts of total capacity, making it one of the largest data centre developments in Southern Europe. Microsoft announced a multiyear deal in October 2025 to lease capacity at the site and has committed $10 billion to the build-out as it works to overcome computing capacity shortages that have constrained Azure’s AI growth. Nscale’s role is to supply the GPU infrastructure that Microsoft will use to serve its cloud customers, a partnership model in which the neocloud provides the hardware layer and the hyperscaler provides the customer relationship.

The 66,000 Nvidia Rubin GPUs scheduled for delivery from late 2027 represent a substantial allocation of next-generation silicon. Rubin, Nvidia’s successor to the Blackwell architecture, entered production in 2026, and demand for the chips already exceeds supply. Nscale’s ability to secure a 66,000-unit commitment reflects both its relationship with Nvidia, which invested £500 million in the company during its Series C, and the scale of Microsoft’s demand for compute capacity in Europe. Nscale’s partnership with BT and Nvidia in the UK is already delivering 14 megawatts of sovereign AI data centre capacity, and the Portugal expansion extends the same model southward to a jurisdiction with cheaper energy and fewer grid constraints than Britain.

The neocloud model

Nscale operates as what the industry calls a neocloud: a cloud provider that focuses exclusively on leasing GPU compute to AI developers, without the hundreds of ancillary services that hyperscalers like AWS, Azure, and Google Cloud offer. The model is simple in theory: acquire GPUs at scale, house them in data centres with sufficient power and cooling, and rent access to AI companies that cannot secure hardware allocations directly from Nvidia. In practice, the model requires enormous capital, long-term offtake agreements with creditworthy customers, and the ability to build data centre capacity faster than the hyperscalers themselves. CoreWeave, the category leader in the United States, went public in 2025 and has secured more than $50 billion in contracts with OpenAI, Meta, and Anthropic. Nscale is pursuing the same model with a geographic thesis: Europe needs its own compute layer, and the company that builds it first will own a structural position in the continent’s AI economy.

The crypto-to-AI transition is not incidental to the model. It is the model. Nscale spun out of Arkon Energy, a cryptocurrency mining infrastructure provider, in early 2024. Crypto mining companies already owned the two assets that AI infrastructure requires: access to large-scale electrical capacity and experience operating GPU-dense computing environments in power-constrained facilities. When AI demand made GPU compute vastly more profitable per megawatt than mining Bitcoin, the economics of the pivot were obvious. Morgan Stanley reclassified the entire crypto mining sector as an energy infrastructure play for the AI economy in February 2026. Nscale simply executed the transition faster and at larger scale than any competitor, raising $3.1 billion across three rounds in less than eighteen months.

The European thesis

OpenAI paused its Stargate UK data centre project in April, citing electricity prices four times higher than in the United States and unresolved copyright regulation. The decision illustrated the fragility of Europe’s position in the global compute race: even when a hyperscaler commits to building AI infrastructure on European soil, energy costs and regulatory uncertainty can derail the investment before a single GPU is powered on. Portugal offers a different proposition. Its electricity costs are lower than Britain’s, it has abundant renewable energy, its Atlantic location provides connectivity to submarine cable networks linking Europe to the Americas and Africa, and the Sines campus already has grid connections permitted for gigawatt-scale power delivery.

The EU has awarded A$180 million in sovereign cloud contracts as part of its effort to reduce dependence on American hyperscalers, which currently control roughly 70 per cent of European cloud infrastructure revenue. But €180 million is a fraction of what a single data centre campus costs to build. Nscale’s €695 million Portugal investment, and Microsoft’s $10 billion commitment to the same site, dwarf the EU’s public spending on sovereign compute. The European sovereign AI thesis is being funded not by governments but by private capital following the economics of GPU demand. Nscale’s Europe-first strategy, with major sites in the UK, Norway, and Portugal, is assembling the physical infrastructure that European AI sovereignty requires, funded by partnerships with the same American hyperscaler the EU is trying to reduce dependence on.

The scale

Microsoft has committed $16 billion to Australia, $30 billion to the UK, $10 billion to Portugal, and billions more across Asia and the Middle East. The company added 1 gigawatt of data centre capacity in the March quarter alone and plans to double its total capacity within two years. Azure AI revenue is growing faster than the company can build infrastructure to serve it, and Microsoft’s CEO told analysts the business is “compute constrained in the near term.” The Nscale partnership solves a specific problem: it lets Microsoft bring capacity online faster than it could by building and operating sites itself, because Nscale handles the GPU procurement, installation, and facility management while Microsoft focuses on the software and customer layers.

The 200-megawatt second building at Sines will represent a substantial addition to Portugal’s data centre footprint. For context, Denmark’s entire installed data centre capacity was 398 megawatts at the start of 2026, and Energinet paused new grid connections after demand overwhelmed the system. Portugal’s advantage is that Start Campus secured its grid permissions early, before the queue of data centre projects across Europe grew to the point where regulators intervened. The 1.2-gigawatt permit at Sines is a head start that may prove more valuable than the site’s other attributes, because the constraint on AI infrastructure expansion in Europe is increasingly not capital, not demand, and not GPU supply, but grid access.

The trajectory

Verda raised $117 million to expand its GPU cloud platform, joining a growing cohort of European neoclouds building compute infrastructure for the continent’s AI companies. But the scale difference is instructive. Verda raised $117 million. Nscale has raised $3.1 billion and is investing €695 million in a single site expansion. The neocloud market is stratifying rapidly into companies operating at hyperscale and companies that will remain niche providers for specific geographies or workloads. Nscale’s partnerships with Microsoft, its Nvidia investment, its CoreWeave-scale ambitions, and its planned IPO position it as the category leader in Europe, and the Portugal expansion is the largest single infrastructure commitment it has made outside the UK.

Whether the neocloud model survives long-term depends on whether the compute shortage persists. If GPU supply catches up with demand, and hyperscalers build enough capacity to serve every customer directly, the neocloud layer becomes unnecessary. If demand continues to outstrip supply, which every current forecast projects through at least 2028, then companies like Nscale that can deliver capacity faster than the hyperscalers can build it will retain pricing power and strategic relevance. Nscale’s trajectory from crypto miner to $14.6 billion AI infrastructure company is the fastest value creation story in European tech history. The Portugal investment is the bet that the conditions which made that trajectory possible, insatiable demand for compute, constrained GPU supply, and Europe’s urgent need for sovereign AI infrastructure, will persist long enough for the neocloud model to become permanent rather than transitional. A crypto mining company became Europe’s most valuable AI startup by recognising, before anyone else, that the AI industry’s real bottleneck is not algorithms but electricity and silicon. Twenty-four months and $3.1 billion later, it is still building.

Get the TNW newsletter

Get the most important tech news in your inbox each week.