Microsoft is finalizing a macro artificial intelligence data center in Wisconsin that, according to its plans, will be located as the most powerful in the world when it starts in early 2026. Dubbed Fairwater Located in Mount Pleasant, this campus aims to support large-scale AI model training and deployment with an industrial focus.
The project investment exceeds the 7.000 million dollars and is accompanied by an additional commitment of 4.000 million to build a second center of similar size. With this, the company is pursuing a unprecedented computing power, with performance up to ten times higher to that of current supercomputers.
Design and computing capacity
The heart of the complex will integrate Hundreds of thousands of next-generation Nvidia GPUs and other AI hardware developments, assembled in continuous clusters linked by a fiber optic mesh large enough to encircle the planet several times. The goal is for the topology to scale without bottlenecks for large AI workloads.
Each rack will function as a dedicated accelerator capable of processing up to 865.000 tokens per second at certain loads, while interconnection will combine NVLink, NVSwitch, InfiniBand and next-generation Ethernet to reduce latency and increase effective bandwidth.
Architecture aims to offer exponential scale computing, optimizing task distribution and memory usage between accelerators. This approach seeks to shorten training times and make new generation models with a greater number of parameters.
Location, impact and sustainability
Fairwater will be built on a plot of land 127 hectares more than Sqm 110.000 built in Mount Pleasant (Wisconsin), a plot with capacity to expand services and interconnections without displacing activity to other areas.
Beyond the civil works, the company plans employment and training programs for the local community, a way to balance technological innovation with the social impact in the immediate surroundings.
In operation, the campus will employ a system of closed-circuit liquid cooling that will cool more than 90% of the hardware and minimizes water usage while keeping dissipated heat at bay. This approach aims for thermal stability and sustained efficiency.
In parallel, the electrical consumption of the complex is will compensate with renewable sources, aligning the infrastructure with sustainable data centers and corporate and regulatory sustainability commitments.
The location coincides with the land where a construction project was announced in 2018. Foxconn factory that did not materialize, a context that brings local economic relevance to the new deployment.
Scale and roadmap
Microsoft has indicated that it will build other identical centers to Fairwater in the United States, while advancing international initiatives such as a hyperscale campus in Narvik (Norway) and the one who aspires to be the greatest UK supercomputer in Loughton.
All these developments will be integrated with more than 400 data centers en 70 regions that support your global cloud, so that new AI clusters benefit from connectivity, replication, and geographic availability.
The roadmap contemplates that the second twin center enter into operation in 2027 or later, in line with the investment schedule announced by the company to expand model training and deployment capacity.
From the management there is talk of a Integrated system that simplifies access to high-performance computing for AI, with the ambition to lead the next generation of infrastructure for developers and businesses.
With the financial and technical push committed, Fairwater is emerging as critical node in Microsoft's strategy for AI: a campus with Massive GPUs, high-speed interconnections, efficient cooling and energy offset, connected to a global network that aims to sustain the most advanced models of the next decade.
