• AI growth strains hyperscalers’ net-zero targets.
• Data centre surge fuels carbon and nuclear bets.

As hyperscalers roll out AI infrastructure, the energy demands tied to it are scaling too. Nvidia’s 120 kW-per-rack systems are only just gaining traction, yet plans are already underway to push toward 600 kW designs.

The surge is putting pressure on the climate commitments made by tech behemoths like Microsoft, Google, and Amazon. Each has laid out long-term sustainability goals: Microsoft aims to be carbon negative by 2030, Google is pursuing net-zero emissions across its value chain in the same timeframe, and Amazon aims for 2040. Despite these timelines, emissions across the industry continue to rise.

Microsoft’s greenhouse gas emissions grew nearly 30 per cent between 2020 and 2023. Google and Amazon posted similar increases – up 48 and 34.5 per cent, respectively, from 2019. And while it’s difficult to isolate the specific impact of AI-related infrastructure, the scale of GPU deployments suggests that AI is playing a significant role.

According to the International Energy Agency (IEA), global data centre electricity use is on track to more than double by 2030 – surpassing the total consumption of Japan. The US, which already accounts for 45 per cent of global data centre electricity use, could see its figure rise sharply. By the end of the decade, American data centres may consume more electricity than the entire domestic manufacturing sector, including steel, cement, and chemical production.

Even so, the IEA estimates data centres will contribute less than 1.5 per cent of global energy sector emissions to 2035. That share, ranging between 300 to 500 megatonnes, may seem small in context – but it’s growing fast.

Turning to carbon capture

To stay on track with their goals, tech firms are increasingly relying on carbon removal technologies.

Microsoft recently signed an agreement with Terradot to remove 12,000 tons of CO2 between 2026 and 2029 using enhanced rock weathering (ERW). ERW spreads finely-ground minerals over large areas of land, where they react with carbon dioxide in rainwater to form stable compounds.

The Microsoft deal supports a broader research initiative: Terradot plans to carry out advanced monitoring at its commercial ERW site in Brazil – beyond what’s required for credit validation – to better understand the technique’s impact on carbon cycling and natural systems.

Brazil’s tropical climate, existing agricultural infrastructure, and proximity to quarries make it a prime location for scaling this type of carbon removal.

Microsoft’s earlier investments include a partnership with 1PointFive to remove 500,000 metric tons of CO2 via direct air capture (DAC). Amazon and Google are also exploring DAC options through projects with CarbonCapture Inc. and Holocene, respectively. However, the technology is still maturing, with high costs and efficiency challenges that limit its near-term impact.

Investments in nuclear energy

Beyond carbon removal, some hyperscalers are placing bets on nuclear power. Microsoft and Amazon are investing in nuclear-powered data centres, with Microsoft teaming with Constellation Energy to revive the Three Mile Island Unit 1 reactor. Amazon has acquired access to nuclear generation capacity through a $650 million deal with Cumulus Data.

These efforts are complemented by interest in small modular reactors (SMRs). Microsoft, Amazon, and Google have each announced partnerships to explore SMRs as a future energy source for their growing data centre footprints. Even Oracle has floated plans to use SMRs to power a large-scale AI facility. Yet SMRs are years away from being commercially viable, and projects like Microsoft’s power purchase agreement with fusion startup Helion Energy are wildly speculative. Despite prior expectations that it will start producing power in 2024, Helion has yet to generate meaningful energy.

Renewables and natural gas in the mix

Meanwhile, hyperscalers continue to make investments in solar, wind, and geothermal energy. Amazon just secured 870 megawatts in new renewable power deals across Spain, while Microsoft signed a 389-megawatt solar agreement in the US. Google is also backing geothermal projects and clean energy-powered industrial parks.

Renewable sources struggle to provide the consistent output required by energy-hungry AI infrastructure. At least partly as a result, companies are also turning to natural gas. Microsoft has supported gas-powered generation in locations like Dublin, and Meta plans to power its Louisiana campus with a 2.2-gigawatt natural gas plant. A separate 4.5-gigawatt gas project is underway in Pennsylvania.

These decisions suggest that, despite public-facing climate commitments, big tech’s primary focus remains on keeping pace with AI demands. While alternative energy and carbon capture are in play, natural gas and fossil fuels still play key parts ins the equation.