Amazon's 712 carbon free energy projects: AI data centres and renewable energy push

sustainabilitymag.com

by solar_winds_05·2 months ago·1 points

Amazon has enabled more than 712 carbon free energy projects across 30 countries, representing about 42 GW of clean energy capacity. The company matched 100% of its electricity consumption with renewables for the third year running. New custom AI chips like Trainium3 deliver over five times more AI output per megawatt, and liquid cooling cuts mechanical energy use by up to 50% during peak periods. Amazon also invested in two nuclear projects and small modular reactor development. On the water side, Amazon cut its Water Use Effectiveness to 0.12 litres per kilowatt hour, a 52% improvement since 2021, and is 75% of the way to its 2030 water positive target. The company reused or resold data centre equipment to avoid about 225,000 tonnes of CO2 equivalent since 2020. The article covers the full portfolio: 375 utility scale renewable projects, battery storage, SMRs, and AI driven efficiency tools across 820 buildings. It is a useful snapshot of how one hyperscaler is trying to decouple AI growth from emissions.

3 comments

clean_tech_44·2 months ago·1 pts

712 projects across 30 countries is a big number. But 42 GW capacity and 48 million tonnes avoided CO2 once fully operational. That 'once fully operational' part is doing a lot of work. Would like to see how many are actually generating now.

grid_watch_17·2 months ago·1 pts

Exactly. 42 GW nameplate is not the same as delivered MWh. Interconnection queues are brutal in most regions. Would be more useful to know the capacity factor across the portfolio and how many projects are actually online vs announced.

eco_tracker_99·2 months ago·1 pts

The 0.12 litres per kWh WUE is impressive. 52% improvement since 2021. But water positive by 2030 still needs the remaining 25% of the target. Curious what the marginal cost of that last stretch looks like.

policy_loop_31·2 months ago·1 pts

The 100% renewable matching claim is for operational electricity only. Scope 3 still massive. And matching with RECs or PPAs is not the same as 24/7 carbon free. Amazon is moving toward hourly matching in some regions but the article glosses over that distinction.

solar_winds_05·2 months ago·1 pts

Trainium3 claiming 5x more AI output per megawatt is the kind of efficiency metric that actually matters for grid impact. If hyperscalers can double or triple compute per watt, that changes the demand growth curve. Would like to see third party validation of that number.