Microsoft Research’s podcast series “The Shape of Things to Come” has released an episode on AI and climate. Host Doug Burger, a Technical Fellow at Microsoft Research, is joined by Amy Luers, head of sustainability science at Microsoft, and Ishai Menache, a partner research manager at Microsoft Research. The episode description frames the conversation as an effort to separate data from hype. The episode is available at the Microsoft Research podcast page.
Datacenter emissions figures
The episode presents emissions data to contextualize the public debate about AI’s climate impact. As of 2024, datacenters account for less than 0.5% of all global emissions and use approximately 1.5% of global electricity, according to figures discussed in the episode. For comparison, the episode notes that energy production accounts for about 75% of all emissions; electricity generation makes up about 35% of energy emissions; and about 20% of all energy is consumed as electricity.
The International Energy Agency projects that electricity use by datacenters could double by 2035, but according to the episode that doubled share would still represent less than 1% of global emissions. The episode also distinguishes between aggregate emissions figures and local infrastructure concerns: datacenters are concentrated geographically and can strain local power grids even when their global share of emissions remains small.
AI applications in optimization
Luers describes three areas where she sees AI as potentially useful for sustainability: optimizing and predicting complex systems, accelerating discovery of climate solutions, and augmenting workforce capacity. The episode provides specific examples for the first two. Menache’s optimization research at Microsoft has achieved a 1% to 2% reduction in power fragmentation through AI-driven virtual machine allocation. In materials science, the episode cites machine learning approaches to designing cement with 20% fewer carbon emissions. AI has also been applied to identifying leaky water infrastructure, according to the episode.
Framing
The episode’s stated goal is to examine AI’s role in sustainability on the basis of available data rather than advocacy. Luers describes a shift in her own thinking over roughly the past decade from a framing focused on large compute’s potential benefits to a more detailed view of AI’s costs and capabilities in sustainability contexts. Menache’s background is in cloud economics and distributed optimization — specifically, how to utilize cloud resources more efficiently — which is the angle he brings to the AI-sustainability discussion.
The episode does not present new research findings. It is a structured conversation between two practitioners who work on these questions inside Microsoft, which is simultaneously a major builder of the datacenter infrastructure under discussion and an investor in AI sustainability applications.