SWI Walked Away From Its NVIDIA Cloud Partner Deal. The Stock Tripled Anyway.

SWI announced in February 2026 that it was acquiring a majority stake in Polarise, one of Europe's few NVIDIA Cloud Partners — an end-to-end AI cloud operator with GPU-as-a-Service capabilities, existing sites in Oslo and Munich, and a reference architecture validated by NVIDIA. The deal valued Polarise at €500 million and came with a €1 billion commitment to build out AI factories across Europe. For an investment conglomerate called SWI Stoneweg Icona Group, it was supposed to be the shortcut from owning data center real estate to actually providing the compute services that the AI cycle rewards most.

Six months later, SWI walked away from that equity deal. Instead of owning Polarise, it became a lender — providing "high double-digit million euros" in debt financing while the two companies "remain separated and develop in their own distinct ways". SWI's new plan: build its own proprietary AI cloud platform from scratch, leveraging crypto-mining facilities it bought in Texas and Sweden, combined with its European data center pipeline.

The stock responded by nearly tripling. From its February 2026 listing price of €3.76, SWI (ticker: SWICH on Euronext Amsterdam) now trades at €10, a market capitalization of €4.8 billion. The market read the Polarise retreat not as a setback but as proof that SWI was confident enough to go it alone.

That reading may be generous. The story worth telling is simpler: SWI is spending billions chasing the highest-margin layer of the AI infrastructure stack without having demonstrated it can actually build, sell, or profit from compute services. The NVIDIA Cloud Partner credential it pursued through Polarise was never really about branding. It was about access — early GPU allocation, priority firmware support, and the validated architecture that hyperscalers and enterprises require before committing multi-year GPU spend. Giving that up to build an in-house platform is not a bold pivot. It is an untested one.

To understand what's at stake, you have to follow the sequence of decisions.

The original logic

SWI's core asset was AiOnX, a 2.3 GW European data center development platform with sites in Ireland, the UK, Denmark, Spain, and Italy. Data center real estate is valuable but it is also the lowest-margin layer of the AI infrastructure chain. The landlords make money when the buildings are full, but the economics don't scale with GPU demand the way compute services do. Polarise was meant to fix that gap. It had 15 years of operating experience, 14 data centers built, an NVIDIA Cloud Partner certification, and a modular "AI Pod" model that could retrofit existing buildings with high-density GPU infrastructure faster than greenfield construction. The acquisition would have given SWI an immediate GPU-as-a-Service business while its own data centers filled out.

The reversal

By August 2026, SWI announced the equity deal with Polarise was converted to debt. The press releases were carefully worded — "both parties mutually agreed," "financing to assist the Polarise founders to reorganize." What this actually means in data center finance is that SWI is no longer buying the company. It is writing a loan. Polarise's founders keep control, Polarise keeps looking for other strategic partners, and SWI gets a fixed-income return instead of upside from GPU services.

The same month, SWI confirmed that over 80% of its capital is now allocated to a transatlantic digital infrastructure platform with more than 4 GW of combined capacity. It intends to push that to above 90%. The US half of that platform is Genesis Digital Assets, a crypto mining operation SWI acquired in June for $500 million — 15 sites with 1.3 GW of energized grid connections in Texas and Sweden, being rebranded "SWI Digital". SWI plans to reposition these mining facilities for AI and high-performance computing.

And instead of buying a proven AI cloud operator, SWI will develop "its own proprietary AI-cloud platform designed to deliver GPU-accelerated compute to enterprises, research institutions, and AI developers".

What the stock price says — and what it doesn't

A 166% gain since listing tells you the market is buying the thesis. SWI's 2025 financials show €93.9 million in revenue and net profit of €423.7 million, but those profit figures are heavily driven by asset revaluations rather than operating income. The company's net loan-to-value ratio sits at a conservative 15.3%, meaning there is balance sheet room to fund this pivot. It announced €260 million in capital raises for 2026 and nearly doubled its revolving credit facility to almost $10 billion in August. The machine is well-oiled for spending.

What the numbers don't show yet is whether the spending works. No customer commitments have been disclosed for the in-house AI cloud platform. No conversion timeline for the crypto-to-AI sites. No retrofitting cost estimate for facilities that currently lack the liquid cooling, redundant power feeds, and network density that AI-grade data centers require. The FTX Recovery Trust has also alleged in litigation that Alameda Research invested approximately $1.15 billion into Genesis Digital Assets between 2021 and 2022 — a cloud SWI has not addressed and may or may not encumber the acquired assets.

The execution gap

There is a reason SWI originally wanted to buy Polarise rather than build. The transition from owning data center boxes to selling GPU compute is not a software upgrade. It requires:

• Securing NVIDIA GPUs, which are in constrained supply. NVIDIA Cloud Partners get priority access to the H200 and Blackwell pipeline. Non-partners compete for allocation on commercial terms alone.
• Building the software stack that lets customers provision, manage, and pay for GPU time — either from scratch or by validating against the NVIDIA reference architecture.
• Retrofitting physical facilities with liquid cooling and redundant power at scale. Mining facilities "lack AI-grade redundant power feeds, cooling infrastructure, and network density". This is not a cosmetic renovation.
• Finding enterprise or sovereign customers willing to commit to a new provider with no operating track record in GPU services. Polarise's selling point was sovereign compliance for European enterprises bound by GDPR. SWI now has to build that trust from zero.

SWI's 2025 annual report said it was "aggregating European and US platforms with end-to-end GPU-as-a-Service operators". That language assumed the Polarise acquisition would close. It did not. Now SWI is trying to be the operator itself, six months after listing, with no revenue line item for AI cloud services to point to.

This is not an indictment of the strategy. It is a timeline question.

The crypto-to-AI conversion play has been tried before. IREN, Core Scientific, Applied Digital, and Hut 8 all moved in this direction. Some succeeded. Others burned capital, delayed timelines, and disappointed investors who priced in faster execution than the engineering allowed. The common failure mode is the same: retrofitting mining rigs for AI workloads is more expensive and slower than the acquisition price implies.

SWI is also not the only European player. Global Switch became an NVIDIA DGX-Ready Data Centre partner in July 2025. Equinix, Digital Realty, and the major hyperscalers already dominate the colocation and cloud layers. The sovereign AI niche — where Polarise had a foothold — is attracting government-backed capital that does not carry SWI's shareholder-return pressure.

Where this leaves the investment case

SWI is a well-capitalized company pursuing a credible strategic direction. The transatlantic infrastructure footprint is real, the balance sheet has room to execute, and the demand for AI compute is not in doubt. What is in doubt is the sequence: can a company that listed at €2.6 billion in market cap and reported €94 million in revenue actually build a GPU services business from the ground up while retrofitting crypto mining facilities — or was buying Polarise the smarter move after all?

The stock price of €10 implies the market believes it can. That belief is worth monitoring against three observable facts over the next year: how quickly SWI secures NVIDIA GPU allocation without the Cloud Partner status it walked away from, whether the Genesis Digital Assets conversion timeline holds against the retrofitting costs that crypto-to-AI competitors have disclosed, and whether any enterprise customers actually sign multi-year GPU compute contracts with a provider that has no prior track record in the business.

Until those facts arrive, the thesis is that SWI is well-funded and ambitious. The investment judgment is whether ambition alone justifies the multiple.