The building at 1001 Third Avenue South in downtown Minneapolis used to be an office. Now it is a data centre, and its anchor tenant is not a Silicon Valley hyperscaler but Core42, the cloud and AI infrastructure subsidiary of Abu Dhabi’s G42 Group. The lease, reported by Bloomberg on Wednesday, covers 20 megawatts of capacity in a facility converted from commercial office space by Legacy Investing, a Virginia-based developer that has spent more than 70 million dollars transforming the property. The deal is a small piece of a much larger pattern: the UAE’s most ambitious AI company is building a network of American data centres while simultaneously constructing the largest AI compute facility outside the United States, and the buildings it is filling were, until recently, home to the kind of white-collar workers whose jobs AI is designed to replace.

The tenant

G42, led by CEO Peng Xiao and backed by Abu Dhabi’s national security adviser Sheikh Tahnoon bin Zayed Al Nahyan, has become the UAE’s principal vehicle for AI infrastructure investment. The company is building the Stargate UAE campus, the international flagship of the 500 billion dollar Stargate joint venture between OpenAI, SoftBank, Oracle, and Abu Dhabi sovereign investment vehicle MGX, across approximately 19 square kilometres of desert south of Abu Dhabi. The first phase, a 200 megawatt compute cluster powered by Nvidia Grace Blackwell GB300 systems, is scheduled to come online by the end of 2026, with full build-out designed to reach one gigawatt of capacity at a projected cost exceeding 30 billion dollars. The project has already attracted geopolitical threats, with Iran warning of retaliatory strikes against Gulf infrastructure including commercial data centres.

Core42’s US expansion extends well beyond Minneapolis. The company signed a ten-year lease with TeraWulf for more than 70 megawatts of data centre infrastructure at the Lake Mariner facility in upstate New York, with an option for an additional 135 megawatts. G42 has also announced a one billion dollar data centre investment in Vietnam and established Core42’s European headquarters in Dublin. The Minneapolis lease adds a relatively modest 20 megawatts to a portfolio that is scaling globally, but its significance lies less in the capacity than in the location: a converted office building in the centre of an American city whose downtown vacancy rate, like those of most mid-sized US cities, has climbed steadily since the pandemic.

The conversion

Legacy Investing acquired 1001 Third Avenue South in 2019 and has since invested heavily in converting the six-storey property from office to data centre use. The company spent more than 70 million dollars on improvements in 2025 alone, expanding the building’s capacity from approximately two megawatts to 21 megawatts. In January 2026, Cloud Capital and Bahrain-based asset manager Arcapita acquired the property for 235 million dollars through a joint venture, reflecting the premium that data centre conversions now command over traditional office valuations.

The economics of office-to-data-centre conversion have become compelling enough to reshape urban real estate markets. US utilities plan to spend 1.4 trillion dollars by 2030 to power the AI boom, and the demand for data centre capacity has outstripped the supply of purpose-built facilities. Converting existing office buildings offers developers a faster path to market than ground-up construction, provided the building’s structural load capacity, power infrastructure, and cooling systems can be upgraded to support the density of modern compute hardware. Not every office qualifies. The buildings that do, typically those with reinforced concrete construction, proximity to fibre networks, and access to sufficient electrical capacity, have become some of the most sought-after properties in commercial real estate.

The pattern

G42’s American data centre strategy reflects a broader shift in how sovereign AI ambitions are materialising in the United States. The company is not simply building compute capacity abroad. It is establishing a physical presence inside the American power grid, the American fibre network, and American commercial real estate markets. The Minneapolis lease, the TeraWulf deal in New York, and facilities in California and Texas create a distributed infrastructure footprint that positions Core42 as a provider of what the company calls sovereign cloud and AI infrastructure: compute capacity that can serve both G42’s own AI workloads and those of its customers, including the US-based companies that require data residency within American borders.

The race to secure data centre capacity has drawn startups, sovereign wealth funds, and hyperscalers into direct competition for the same constrained resources: power, land, cooling, and fibre connectivity. In Minneapolis, Legacy Investing’s conversion of an office building into a 21 megawatt facility illustrates how developers are working around those constraints by repurposing existing urban infrastructure rather than competing for greenfield sites in the exurban corridors where most large-scale data centres are built. The approach has limits. Twenty megawatts is a fraction of the capacity that a purpose-built hyperscale facility delivers, and the structural constraints of an existing building make future expansion difficult. But for a tenant like Core42, which needs distributed US presence rather than concentrated scale, the urban conversion model offers something that a greenfield campus in rural Virginia does not: proximity to enterprise customers and the network interconnection density of a major metropolitan area.

The context

The geopolitics of UAE-linked AI infrastructure in the United States remain sensitive. G42 severed its ties with Chinese technology companies in 2024 under pressure from the US government, a prerequisite for receiving access to advanced American AI chips. Microsoft invested 1.5 billion dollars in G42 as part of that realignment, and the company has since positioned itself as a bridge between American AI technology and Gulf capital. OpenAI’s decision to pause its Stargate UK project over energy costs and regulatory concerns has made the UAE’s willingness to build at scale and speed more strategically valuable, and G42’s ability to secure chip imports for the Stargate UAE campus has been treated by the US government as a test case for whether AI infrastructure can be deployed in allied nations without compromising national security.

The financial dynamics of AI infrastructure investment have produced valuations and capital commitments that would have been inconceivable five years ago. Oracle’s stock fell 50 per cent despite a Wall Street buy consensus as concentration risk from its OpenAI dependency spooked investors. Alphabet raised its full-year 2026 capex estimate to between 180 and 190 billion dollars. The International Energy Agency predicts that energy use from data centres will double by the end of 2026. Into this environment, a 20 megawatt lease in a converted Minneapolis office building registers as a footnote. But the tenant signing the lease is the company building the largest AI compute facility outside the United States, and the building it is moving into is one of hundreds of American offices being repurposed to house the infrastructure that is making the offices themselves obsolete. The conversion is not just architectural. It is economic. The same AI systems that emptied the office are now filling it with servers.