Core42 triples Buffalo capacity as its AI infrastructure buildout spans three continents
Core42 has expanded its Buffalo AI cluster from 18MW to 60MW, adding 42MW of compute capacity at its Lake Mariner site in New York and more than tripling the production capacity at one of its largest US locations. The size of the jump matters less than where it is happening, because Core42, which sells sovereign cloud and AI infrastructure, is putting more compute inside the US at a point when governments increasingly care about which country holds a company's AI capacity rather than simply how much of it exists.
Core42 announced the expansion on 2 June, and as a G42 company it belongs to a group that drew close US scrutiny over its past ties to Chinese technology suppliers, the same scrutiny that preceded Microsoft's $1.5 billion investment in G42 in 2024. That history is why a 42MW buildout inside the US carries weight beyond the capacity itself, signalling which side of a dividing technology order the group is choosing to operate within at a point when Washington treats advanced compute as a national security matter rather than ordinary trade. The question follows the company to every site it opens, from Dublin to Abu Dhabi, because customers buying sovereign cloud are buying the jurisdiction as much as they are buying the hardware that happens to sit inside it.
"We are scaling our U.S. infrastructure in line with long-term deployment programs," said Talal M. Al Kaissi, Chief Executive Officer, Core42. "Increasing our U.S. capacity at Lake Mariner strengthens our ability to serve hyperscale, AI-native and large enterprise workloads, and further extends the build-out of our AI infrastructure globally."
The hardware choices fit that environment, since the expansion adds NVIDIA and AMD capacity to the site's existing AMD base, continuing a design that runs workloads across more than one type of accelerator rather than tying the site to a single vendor. With AI chips in short supply and that supply itself uneven across vendors and regions, the ability to run different silicon has become a commercial advantage in its own right, and it is one Core42 is repeating at sites spread across three continents.
Lake Mariner had a track record before this expansion, because its AMD Instinct MI300-based Maximus cluster reached the top 20 of the global TOP500 supercomputing list, a result Core42 uses as evidence that it can run high-end production infrastructure rather than simply lease out space and power to others.
The Buffalo site sits within a larger US presence that also includes deployments in Dallas, Texas, in Sunnyvale and Stockton, California, and in Minneapolis, Minnesota, the last of these home to the Condor Galaxy supercomputers built in collaboration with Cerebras. Taken together, the sites are meant to handle both frontier training, which is hardware-intensive and runs over long periods, and the fast inference that live AI services depend on, two very different jobs that increasingly call for very different chips.
The US growth is only one part of a buildout that is international by design, given that Core42 set up its European headquarters in Dublin in 2025 and added compute across Italy and France, leaving it with infrastructure that spans the US, Europe and the Middle East. The AI Cloud platform, launched in October 2025, ties those locations together by letting customers run compute in different countries under a single operating model that covers training, fine-tuning and inference, and the company now operates 10 sites worldwide with more due through 2026. For a sovereign cloud provider the spread is itself the product, because the pitch is capacity inside whichever borders a given customer needs, and for the enterprises and AI-native firms Core42 is targeting that decision about location now sits alongside price and performance rather than somewhere behind them.
