Microsoft has connected datacenters in Wisconsin and Atlanta into what it calls its first AI superfactory. The company disclosed the linkage on 15 April 2025 through its official news channel, Microsoft Source. The announcement names two endpoints and a label. It offers little else.

The term "superfactory" has no established definition in the industry. Microsoft applies it here to a networked set of datacenters that train and run AI models at scale. By Priya Raman's read, the word does most of the work in this announcement, and the engineering does the rest of the explaining, or fails to.

What does the announcement actually contain?

The public statement contains a geographic claim and a first: datacenters in Mount Pleasant, Wisconsin, and the Atlanta, Georgia, area now operate as one connected system. Microsoft describes this as its first such facility. The company has not published figures on:

• power capacity in megawatts at either site
• capital expenditure for the connection or the sites
• interconnect bandwidth between Wisconsin and Atlanta
• GPU counts or accelerator types deployed
• a completion date or commissioning timeline

That is a long list of absences. Each one matters to buyers, utilities and competitors tracking Microsoft's buildout against rivals such as Google and Amazon.

Where do these datacenters sit in Microsoft's wider plan?

Microsoft opened its Mount Pleasant campus in Racine County, Wisconsin, part of a multi-billion-dollar investment in the state announced in 2024. The Atlanta-region sites form part of the company's Georgia datacenter footprint. The new claim is that these facilities now function jointly rather than as separate regional capacity. Microsoft has previously stated ambitions to expand AI infrastructure across dozens of sites, and this linkage appears to extend that strategy across state lines for the first time.

The comparison worth making: a single datacenter is a factory; a connected pair across 1,200 kilometers is a supply chain. Microsoft has not yet shown the logistics that make the second thing work, only the name for it.

Why connect distant sites at all?

Distributed training, the practice of splitting an AI model's computation across multiple locations, requires high-bandwidth, low-latency links. Vendors pursue it for three reasons: power is cheaper and more available in different regions; local content rules favor geographic spread; and failure at one site need not stop training at another. Microsoft's announcement implies all three benefits but confirms none with data.

The company also gains political cover. Wisconsin and Georgia both competed for these facilities with tax incentives. Linking them lets Microsoft tell two state governments that their investments serve one national asset. Whether the linkage reduces or concentrates operational risk is not addressed in the statement.

What should readers watch next?

Watch for numbers, because the story currently runs on geography. Microsoft will eventually file power agreements, county permits and utility interconnection requests, and those documents will carry megawatt figures the press release omits. Earnings calls may add capital expenditure detail for the fiscal year. Third parties, including fiber route operators between the Midwest and the Southeast, may confirm interconnect capacity.

Until then, the claim rests on two place names and one word. The skepticism here is not that Microsoft has failed to build something. The company operates one of the largest datacenter estates in the world. The skepticism is that "first AI superfactory" is a verifiable claim only when Microsoft publishes the specifications that distinguish a superfactory from two datacenters on the same network diagram.

Die Signal will track permits, filings and utility data as they appear.