Data center carbon emissions are set to skyrocket by 2030, with hyperscalers producing 2.5 billion tons of carbon – and power hungry generative AI is the culprit

Data center carbon emissions are set to skyrocket over the next decade, according to research from Morgan Stanley, with generative AI prompting a massive spike in power consumption.

A study by the investment bank predicted that the data center industry at large will produce 2.5 billion tons of carbon through 2030. Notably, the research found this rapid increase is three-times higher than if generative AI hadn’t been introduced in late 2022.

This is across scope one, two, and three carbon emissions, and would constitute around 40% of the overall annual carbon emissions in the US.

Global greenhouse gas (GHG) emissions from data centers will grow from 200 million tons in 2024 to 600 million tons by 2030 as sites get bigger and more numerous.

While 60% of the stated emissions are derived from the power needed to keep data centers operational, 40% of the figure is derived from data center manufacturing emissions, such as materials and infrastructure.

The report predicted that 50% of the data center build-out will occur in the US alone and that the specific carbon emission increase by 2030 in America will equate to 2.8x the current size.

Morgan Stanley sees this carbon uptick as “driving large investment” opportunities in decarbonization solutions, particularly as hyperscalers look closely at meeting their 2030 sustainability goals and quotas.

The firm prices the 2.5 billion ton “problem” – or “opportunity” – at $325 billion, using the price forecasts of its EU commodities team for EU carbon allowances, though the report was mindful to mention that this figure is only for context as “no global carbon price exists today.”

Data center carbon emissions in the spotlight

Morgan Stanley’s predictions for an uptick in decarbonization investment come at a time when the world’s largest hyperscalers have reported repeated concerns with sustainability.

Microsoft’s 2024 Sustainability Report, for example, found that its carbon emissions had grown by just under a third (29%) on the back of enlarged data center infrastructure to meet the demand for AI tools.

This placed a serious strain on the firm’s desire to meet carbon-neutral goals by the end of the decade.

Google has reported suffering from similar problems, with its annual environmental report showing GHG emissions had increased by almost half over the last five years on the back of AI.

Only Amazon appears to be maintaining a sustainable condition, having reached its 100% renewable energy goal seven years ahead of schedule according to its 2023 sustainability report.

The emergence of generative AI and its associated power consumption has prompted a wholesale reassessment of energy efficiency across the data center industry, with stakeholders warning that this increase could scupper ESG goals at major providers.

In May, research from the Electric Power Research Institute (EPRI) warned that data center energy consumption in the US is expected to more than double by 2030 as providers look to keep pace with generative AI demand.

By the turn of the decade, almost one-tenth (9%) of US electricity generation could be consumed by data centers, with the EPRI predicting impending supply challenges, among other issues.

These rising demands have led hyperscalers such as Microsoft, AWS, and Google to consider a range of energy sources. Microsoft, for example, recently announced a deal that will see the Three Mile Island nuclear plant restarted to power data center operations.

AWS has also made inroads with regard to nuclear energy, having unveiled plans to acquire a nuclear-powered data center campus as part of a $650 million deal with a US-based electricity provider, Talen Energy.

Unveiled in March this year, the deal will see AWS acquire the Cumulus data center complex located next to the 2.5 gigawatt Susquehanna nuclear power station in Pennsylvania.