On the southern shore of Lake Ontario, where upstate New York meets a steady industrial wind, a data center campus has been expanding quietly for the better part of two years. The facility at Lake Mariner, near Buffalo, is not particularly famous outside technical circles. But its most recent expansion, a 42MW addition that brings total capacity to 60 megawatts, says something significant about how the global AI infrastructure race is actually unfolding.

[For more news, click here]

The company behind it is Core42, a G42 company headquartered in Abu Dhabi and specializing in what the industry calls sovereign cloud and AI infrastructure. In plain terms, Core42 builds the physical foundations on which frontier AI runs: the racks of accelerators, the cooling systems, the networking fabric, and the software layers that allow organizations to train massive models and run them at scale. The company's expansion at Lake Mariner is the latest move in a buildout that now spans the United States, Europe, and the Middle East, and it reflects a broader shift in where AI compute power is being concentrated.

The Architecture Behind the Numbers

The Lake Mariner site already had a proven record before this expansion. Its Maximus cluster, built on AMD Instinct MI300 accelerators, earned a Top-20 ranking on the global TOP500 supercomputing list, a benchmark that measures raw performance across the world's most powerful machines. The new expansion integrates additional infrastructure from both AMD and NVIDIA, creating what Core42 describes as a heterogeneous design. In practice, that means the facility can route different types of workloads to different accelerator platforms depending on which delivers the best price-performance ratio for the task.

This approach is more consequential than it might sound. Most discussions of AI infrastructure reduce the conversation to a single variable: how many GPUs does a facility have. But the actual challenge of running frontier AI at production scale involves matching workloads intelligently to hardware. A large training run has different computational requirements than a real-time inference task. A heterogeneous architecture, one that maintains both AMD and NVIDIA infrastructure under a unified operating model, gives operators more flexibility to optimize across those different demands. Core42 has made this architectural flexibility a central feature of its differentiation strategy.

"We are scaling our U.S. infrastructure in line with long-term deployment programs. 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," said Talal M. Al Kaissi, Chief Executive Officer, Core42

The language Al Kaissi uses is deliberate. Hyperscale, AI-native, and large enterprise are three distinct customer categories with very different infrastructure demands. Hyperscalers need raw density and reliability at enormous scale. AI-native companies, the generation of businesses whose products are built on top of foundation models, need speed and flexibility. Large enterprises need security, compliance, and the ability to operate across jurisdictions. Core42 is positioning Lake Mariner, and its broader network, to serve all three simultaneously.

A US Footprint That Runs Deeper Than Most Realize

Lake Mariner is not an isolated deployment. Core42's US presence now includes sites in Dallas, Texas; Sunnyvale and Stockton, California; and Minneapolis, Minnesota. The Minneapolis infrastructure includes the Condor Galaxy supercomputers, developed in collaboration with Cerebras, a company known for its wafer-scale processor architecture that takes a fundamentally different approach to AI compute than conventional GPU-based systems.

Together, these deployments form a national-scale compute network that spans both coasts and the Midwest, covering multiple climate zones and power grids. The geographic distribution is not accidental. Redundancy, latency optimization, and jurisdictional compliance all factor into where serious AI infrastructure gets built. A company running workloads for both American enterprises and international clients needs to be able to route traffic intelligently across a footprint that is genuinely distributed rather than nominally so.

The European and Middle Eastern Layers

The US expansion sits inside a larger international buildout. In 2025, Core42 established its European headquarters in Dublin, with AI compute deployments extended across Italy and France. The Middle East, where G42 operates as a flagship technology group backed by Abu Dhabi's sovereign wealth infrastructure, provides the organizational and financial foundation from which Core42 operates.

This structure matters because it shapes how Core42 approaches questions of sovereignty and data residency that are becoming increasingly central to enterprise AI adoption. Organizations in regulated industries, governments, and any customer dealing with data that cannot leave a particular jurisdiction need compute infrastructure that is physically and legally contained within specific boundaries. Core42's distributed network, spanning the US, Europe, and the Middle East, is explicitly designed to support what the industry is calling sovereign AI: the ability to operate frontier AI workloads under a consistent model while maintaining local control over where data actually resides.

The AI Cloud Platform Connects It All

The physical infrastructure is only part of the story. In October 2025, Core42 introduced its AI Cloud platform, which functions as the access layer connecting all of the company's distributed infrastructure into a single operating model. Customers can provision compute across jurisdictions, choosing the location and hardware configuration that best suits a given workload, without needing to manage the underlying infrastructure complexity themselves.

The platform supports the full AI lifecycle, from large-scale training and fine-tuning through to real-time inference, and is designed to take advantage of the heterogeneous hardware architecture across Core42's sites. The practical implication is that a customer can run a training job at Lake Mariner on AMD accelerators, fine-tune a model on NVIDIA hardware in a European deployment, and serve inference from a Middle Eastern facility, all under a single commercial and operational framework.

With ten operational sites currently active and additional deployments planned for 2026, the platform becomes more valuable as the network grows. Each new site adds compute capacity, but it also adds flexibility, redundancy, and jurisdictional coverage. The network effects of a distributed AI infrastructure are cumulative in ways that a single large data center, however powerful, cannot replicate.

Why Upstate New York Makes Sense as an AI Hub

The choice of Lake Mariner as a North American anchor site reflects practical considerations that rarely make it into infrastructure announcements. The facility benefits from access to renewable energy from the nearby wind resources and proximity to established transmission infrastructure, both significant factors when a site is drawing tens of megawatts continuously. Buffalo's position in the broader northeast corridor also offers low-latency connectivity to major financial and enterprise markets along the eastern seaboard.

The site's history as an industrial location, common to much of upstate New York's former manufacturing belt, provides large available footprints, existing utility infrastructure, and a regulatory environment that has, in recent years, worked to attract data center investment as a form of economic redevelopment. For a company looking to expand rapidly, those are meaningful advantages relative to building on greenfield sites in already-saturated markets.

What This Expansion Signals About AI Infrastructure More Broadly

Core42's Lake Mariner expansion is, taken in isolation, a straightforward capacity announcement. Forty-two megawatts of new AI infrastructure at an existing site with a strong performance track record. But the broader pattern it represents is worth examining carefully. The companies building the most durable positions in the AI infrastructure market are not those with the largest single site. They are the ones assembling geographically distributed, architecturally flexible networks that can serve customers across multiple regulatory environments under a unified platform.

Core42 has moved methodically toward that model, building out the US footprint incrementally while establishing the European and Middle Eastern presence simultaneously. The AI Cloud platform launched in late 2025 represents the unification layer that makes those separate deployments into a coherent commercial offering rather than a collection of regional data centers. The Lake Mariner expansion adds raw capacity to a network that is, in structural terms, already further along than many of its peers.

For enterprises and hyperscalers evaluating where to run frontier AI workloads in 2026, that architecture may matter as much as the megawatt count. The question is not simply how much compute is available. It is whether the infrastructure can support the full lifecycle of AI development, across the jurisdictions and hardware configurations the workload actually requires. On that measure, Core42's buildout at Lake Mariner, and across its global network, is a considered answer to a question the industry is still working out how to ask.

Related Articles:

How Pattern Intelligence Is Turning Ecommerce Operations Into a Real-Time AI System

Why Trust Has Become the Most Valuable Currency in MENA's Digital Payment Revolution

How the 2026 FIFA World Cup Became the Most Revealing Test of AI in the Workplace