Amsterdam listed SWI Group is buying a stake in Genesis Digital Assets, the US operator of 1.3 gigawatts of permitted power connections across 15 sites, several already wired for hyperscale AI tenants.
The bottleneck in building AI infrastructure has migrated from the chip fab to the substation. In the clearest signal of that shift so far this year, Amsterdam-listed SWI Group said on 15 June 2026 it will pay $500 million for a meaningful stake in Genesis Digital Assets, the US operator of 1.3 gigawatts of grid connections spread across 15 American sites, several in Texas and described in the announcement as hyperscaler-ready (SWI Group press release on PR Newswire).
A gigawatt, the unit that anchors the deal, is a measure of power capacity rather than compute. One gigawatt is roughly what a large hyperscale data center campus draws at full tilt, and it is also roughly the size of the queue that newer entrants now wait years to obtain from regional grid operators. The legacy crypto-mining sites SWI is buying already hold the right to draw that much power, and they hold the land and substations to do it. That is the asset.
SWI's strategy is to repurpose those sites for high-performance computing and AI workloads, leaning on operational know-how from its European data center platform, AiOnX, which the company says contributes another 2.3 GW across Ireland, the United Kingdom, Denmark, Spain, and Italy. Stacked on the 1.3 GW of US capacity, the company frames the combined footprint as 3.6 GW of transatlantic AI-ready infrastructure (SWI Group press release on PR Newswire).
Three caveats attach to the framing. First, every capacity figure in the announcement is SWI-attributed, and the press release is the only source currently on the record. Independent corroboration of the 1.3 GW figure, the 15-site footprint, and the 3.6 GW aggregate has not been published. Second, the headline $500 million price tag covers a stake in Genesis Digital Assets, not a controlling acquisition. The deal is largely structured through a $1.124 billion preference share class in GDA, with SWI taking roughly 77.2 percent of the preference stack and around 38.3 percent of total equity, a structure that gives the company economic exposure to GDA's value but not majority operating control. Third, "AI-ready" is announced intent, not completed retrofit. The press release does not disclose retrofit timelines, capital expenditure per site, or any hyperscaler offtake agreement.
The economic logic of the transaction is the part that travels beyond SWI. New grid connections in the major US data center markets routinely face multi-year queues, and the cheapest way to bypass those queues is to acquire sites that already hold interconnection rights, even if they were originally built for a different use. Crypto-mining campuses in Texas and the Pacific Northwest were built for exactly that reason: cheap power, fast permitting, and industrial-scale substations. Converting them to AI workloads is not trivial, since the power-quality, cooling, and fiber requirements are different, but the gating asset, the right to draw large amounts of electricity, is already in hand.
What to watch next. SWI's claim of 3.6 GW will be tested on three things in the coming year. How much of the 1.3 GW at GDA is actually energised, and how much is permitted but not yet connected. What share of the Texas sites attract a named hyperscaler tenant on a multi-year offtake. And whether the preference-share structure leaves SWI exposed if GDA's equity value falls short of the $1.124 billion stack. Each of those questions is answerable from public filings, grid-operator interconnection reports, and tenant disclosures once the deal closes, and each will determine whether the bottleneck shift the press release implies is real or merely announced.