Microsoft is trying out cross-laminated timber (CLT) material for building a new datacenter in Northern Virginia, to help reduce its carbon footprint. Instead of using the usual steel and concrete most commercial buildings are made of, the company decided to incorporate CLT in its latest data centers to help reduce greenhouse gas emissions. According to Microsoft, it estimates using wood for parts of its data centers will cut down its carbon footprint by 35% versus steel construction. Greenhouse emissions are reduced by an even greater 65% when CLT is substituted for precast concrete.
The company won’t build its Northern Virginia data centers completely from wood, though. Instead, it will replace the concrete used on the floors and ceilings with CLT and then apply a thinner layer of concrete to make the wood last longer and protect it from the elements. CLT is more expensive than steel and concrete. But since it’s also a lighter material, using it for major parts of the building means that the structure would need less steel for support, thus saving on materials and labor. “We have to be system thinkers across the entire value chain of these materials that go into our datacenters and the equipment that supplies them,” says Jim Hanna, team lead for Microsoft’s datacenter engineering team.
The AI boom’s massive power and infrastructure requirements are setting back the climate goals of several companies, so Microsoft’s technique of offsetting its emissions with the use of alternative construction techniques could help it get back on track. According to its Environmental Sustainability Report, its emissions jumped by 29.1% from 2020, with the company blaming construction as the primary driver for this undesirable upswing. Unfortunately, Microsoft is not alone in this, as Google has also fallen behind its climate goals, with the search giant’s greenhouse gas emissions increasing by 48% from 2019.
Aside from trying out new construction techniques, AI bandwagon companies are also investing huge amounts in nuclear power. Microsoft has already signed a deal to restart the Three Mile Island reactor, while Google and Oracle are investing in modular nuclear reactors. These sources are expected to come online in the early the 2030s — just in time, too, as future AI models that will consume up to 5 gigawatts of power during training are expected to come online then. Without these cleaner alternative sources, AI models could overburden the current electrical grid, potentially making it impossible for companies to train efficiently. So, the only way that these companies can realistically compete for AI supremacy is to ensure that they will have portable and easily deployable clean electricity located near their data centers.
