https://www.amazon.science/tag/decarbonization Decarbonization en-US Tue, 04 Aug 2026 16:32:57 GMT https://www.amazon.science/publications/amortizing-ai-training-carbon-footprint-challenges-limitations-and-a-path-forward Allocating the one-time carbon cost of training a large AI model across the inference requests it serves is an open methodological problem with no standardized solution. The choices made in boundary definition, functional unit selection, and lifetime forecasting can alter reported per-request emissions by an order of magnitude, undermining any comparison across models. We decompose this problem into three challenges: defining the emission boundary (what counts as training carbon), selecting a functional unit (how to measure inference work), and forecasting lifetime usage (how many units the model will ultimately serve). The three differ in kind yet interact, since boundary and lifetime choices compound and the functional unit constrains forecasting complexity. Boundary scope and functional unit selection are addressable through improved measurement and standardized reporting; lifetime usage, by contrast, is not deterministic and for open-source models may never be precisely measurable. We outline a starting point to tackle each challenge, arguing that imperfect but consistent measurement today is preferable to waiting for a consensus that may never arrive. Tue, 04 Aug 2026 16:32:57 GMT https://www.amazon.science/publications/amortizing-ai-training-carbon-footprint-challenges-limitations-and-a-path-forward https://www.amazon.science/blog/the-fuel-of-the-future-is-already-here-why-triso-matters Millimeter-scale particles of nuclear-reactor fuel are encased in four layers of different materials that act as a “miniature containment system”. Wed, 24 Jun 2026 19:57:09 GMT https://www.amazon.science/blog/the-fuel-of-the-future-is-already-here-why-triso-matters https://www.amazon.science/publications/power-optimization-for-sustainable-smart-speakers-echo-pop-case-study In this paper, we present a comprehensive system-level approach to advancing device sustainability through power optimization for smart home devices, with a detailed case study of Amazon's Echo Pop. Through Lifecycle Assessment (LCA), we identified that Echo Pop generates an estimated 42 kg CO2e over its product lifetime, with 24 kg CO2e (57%) attributed to use-phase emissions, highlighting the critical importance of idle power optimization for decarbonization efforts. We implemented a novel system architecture for energy efficiency that leverages CPU Suspend-to-RAM states and Wi-Fi power save modes. We minimize energy consumption during device inactivity and via coordinating various system services to maintain seamless user experience. The system intelligently transitions between power states using a low power Digital Signal Processing (DSP) core for monitoring ambient audio, while duty-cycling background connectivity tasks. We validated the effectiveness of our multi-domain power management approach across SoC, Wi-Fi, and system-level components, through comprehensive power rail analysis using high-resolution measurement methodologies in a lab setting. Results demonstrate substantial power reduction achievements, with Echo Pop achieving approximately 1.1W standby power consumption—a 49% improvement over previous generation devices that drew over 1.5W. Ground truth validation through in-field telemetry data confirms our lab projection models, with a daily average energy consumption of 28.21Wh closely matching the in-field empirical measurements of 28.07 Wh. This work establishes a validated framework for sustainable smart device design that balances environmental impact reduction with maintained functionality and user experience. Wed, 03 Jun 2026 18:45:49 GMT https://www.amazon.science/publications/power-optimization-for-sustainable-smart-speakers-echo-pop-case-study https://www.amazon.science/publications/optimized-demand-based-charging-networks-for-long-haul-trucking-in-europe Battery electric trucks (BETs) are the most promising option for fast and large-scale CO2 emission reduction in road freight transport. Yet, the limited range and longer charging times compared to diesel trucks make long-haul BET applications challenging, so a comprehensive fast charging network for BETs is required. However, little is known about optimal truck charging locations for long-haul trucking in Europe. Here we derive optimized truck charging networks consisting of publicly accessible locations across the continent. Based on European truck traffic flow estimates for 2030 and actual truck stop locations we construct a long-term charging network that minimizes the total number of required locations. Our approach introduces an origin-destination (OD) pair sampling method and includes local capacity constraints to compute an optimized stepwise network expansion along the highest demand routes in Europe. For an electrification target of 15% BETshare in long-haul and without depot charging, our results suggest that about 91% of electric long-haul truck traffic across Europe can be enabled already with a network of 1,000 locations, while 500 locations would suffice for about 50%. We furthermore show how the coverage of OD flows scales with the number of locations and the size of the stations. Ideal locations to cover many truck trips are at highway intersections and along major European road freight corridors (TEN-T core network). Tue, 03 Dec 2024 04:49:47 GMT https://www.amazon.science/publications/optimized-demand-based-charging-networks-for-long-haul-trucking-in-europe https://www.amazon.science/news-and-features/five-ways-amazon-is-preparing-for-the-energy-demands-of-the-future From investing in new carbon-free energy projects to advocating for grid modernization and collaborating with key stakeholders around the world, Amazon is working toward a cleaner energy future. Tue, 25 Jun 2024 13:45:17 GMT https://www.amazon.science/news-and-features/five-ways-amazon-is-preparing-for-the-energy-demands-of-the-future https://www.amazon.science/latest-news/re-mars-revisited-optimizing-ai-ml-workloads-for-sustainability Session focused on tips and tools that can help customers reduce the carbon footprint of artificial intelligence and machine learning workloads. Fri, 24 Feb 2023 17:50:52 GMT https://www.amazon.science/latest-news/re-mars-revisited-optimizing-ai-ml-workloads-for-sustainability https://www.amazon.science/blog/the-path-to-carbon-reductions-in-high-growth-economic-sectors Confronting climate change requires the participation of governments, companies, academics, civil-society organizations, and the public. Mon, 01 Aug 2022 14:37:12 GMT https://www.amazon.science/blog/the-path-to-carbon-reductions-in-high-growth-economic-sectors https://www.amazon.science/blog/amazons-scientific-approach-to-meeting-and-measuring-its-climate-goals How Amazon is aligning its decarbonization goals with the best available science. Wed, 30 Jun 2021 19:25:35 GMT https://www.amazon.science/blog/amazons-scientific-approach-to-meeting-and-measuring-its-climate-goals https://www.amazon.science/latest-news/8-science-related-points-from-jeff-bezoss-2019-shareholder-letter Bezos’s Shareholder Letter has become a must read, along the lines of Warren Buffet’s letter to Berkshire Hathaway shareholders, or the Bill & Melinda Gates Annual Letter. Tue, 21 Apr 2020 14:59:47 GMT https://www.amazon.science/latest-news/8-science-related-points-from-jeff-bezoss-2019-shareholder-letter