Public discussion of AI focuses on models, chips and software. Beneath those sits a physical layer that receives far less attention: the cabling, connectors, cooling and materials that make large computing systems work. A partnership 3M and Microsoft announced on 15 July concerns exactly that layer, alongside 3M’s internal use of AI. It is a useful reminder that the AI boom creates industrial demand that has little to do with algorithms.
A partnership beneath the software
The agreement covers AI data-centre infrastructure and 3M’s use of Microsoft technology. One part concerns expanded-beam optical connections for Azure infrastructure, which is a specific piece of engineering rather than a general statement of collaboration. The companies describe connections designed to tolerate contamination and simplify maintenance.
Why optical connections matter
Inside a large data centre, enormous amounts of data move between servers, racks and buildings. Optical connections carry that traffic, and their reliability affects the whole system. Expanded-beam connectors use lenses to transmit light across a small gap, which makes them less sensitive to dust and easier to service than physical-contact connectors. In a facility with thousands of connections, small reliability gains compound.
Maintenance is an underrated cost
A data centre is a system that must be maintained for years. Components that are easy to clean and replace reduce downtime and labour costs. The partnership’s emphasis on maintenance reflects a practical concern that operators know well and that public discussion often ignores.
The physical supply chain
AI expansion has created demand for specialised components, from advanced packaging for chips to cooling systems and power distribution. Companies that supply these components are part of the AI economy even though they do not build models. 3M’s involvement illustrates how a diversified manufacturer can find a role in a technology boom by supplying the less visible inputs.
AI inside a manufacturer
The agreement also includes AI-driven changes within 3M’s own operations. This is the other side of the story: established companies adopting AI for their internal processes. Such transformations are less covered than product announcements and often more consequential for the organisations involved, because they change how work is done rather than what is sold.
Design goals are not measured results
The announcement sets out intended advantages and does not provide an independent demonstration of benefits across every deployment. Evidence about reliability, maintenance and performance will be more useful as systems are used. That is the normal pattern for infrastructure partnerships: the claims are plausible and unverified until operational data accumulates.
What to take from it
The story is a reminder that AI investment also creates demand for specialised industrial components, and that the sector’s growth depends on supply chains that are easy to overlook. The next time a data centre is announced, it is worth remembering that someone has to manufacture what goes inside it.
The cooling problem
Data centres generate heat, and removing it is a major engineering challenge. Cooling consumes energy and, in many designs, water. As AI workloads increase the density of computing power, cooling requirements rise. Companies are exploring liquid cooling, immersion systems and alternative designs. The materials and components involved are a growing market, and they are rarely mentioned in discussions of AI that focus on models and chips.
Why hardware is a bottleneck
The AI boom depends on physical supply chains: advanced semiconductors, high-bandwidth memory, networking equipment, power distribution and cooling. Each of these can constrain how quickly capacity is built. Shortages in any one component can delay projects and raise costs. That is why partnerships between technology companies and industrial manufacturers matter more than their low profile suggests.
The investment cycle
Infrastructure investment runs on long timelines. A data centre built now will operate for a decade or more, and the components inside it will be maintained for years. Companies committing to partnerships in this space are betting on sustained demand. If the AI investment cycle slows, some of that capacity will be underused, which is a risk that both suppliers and operators carry.
The environmental footprint
Data centres consume electricity and water, and their growth has environmental consequences that are increasingly scrutinised. Efficiency improvements help, and they are partly offset by rising demand. The components covered by partnerships like this one affect efficiency at the margins: better connections and cooling reduce waste. The larger question of how much computing the sector should build, and where its power comes from, sits outside any single company’s procurement decision.
For operators, the lesson is that infrastructure partnerships deserve as much scrutiny as software ones. A connection that fails or a cooling system that underperforms affects availability, and availability is what customers pay for. The unglamorous components are where reliability is won or lost.
The partnership also illustrates how AI demand ripples through industrial supply chains that have nothing to do with algorithms. As capacity expands, the demand for materials and components grows with it, and the companies that supply them become part of the story whether or not they are described that way.
Sources
- Microsoft. 3M and Microsoft announce strategic partnership. news.microsoft.com
Image: Anthonya · CC0 · via Wikimedia Commons.