Amazon, Orbital to Advance Decarbonization Technology

Amazon Web Services (AWS) has entered a three-year partnership with Orbital Materials (Orbital) to develop technologies aimed at improving data center decarbonization and efficiency. 

Orbital, a startup that uses AI to develop advanced materials and climate technologies, will collaborate with AWS to evaluate the scalability and performance of new solutions for carbon removal and increased efficiency. Jonathan Godwin, CEO, Orbital, emphasized that this partnership will accelerate the development of products related to cooling, water utilization, and carbon removal.

In 1Q24, Orbital launched its first product, a carbon removal technology based on a proprietary active material. In an interview with Reuters, Godwin explained that the material works like an atomic-level sponge, with specifically sized cavities that interact effectively with CO2 while filtering out other compounds. Since its development, the material's performance has improved tenfold thanks to Orbital's AI platform. The company plans to deploy and test this carbon removal technology by the end of 2025. 

Godwin also highlighted the potential cost savings of the new material. "It represents an estimated 10% of the hourly cost of renting a GPU chip to train powerful AI models, which is just a fraction of the cost of carbon offsets," he stated.

As part of the agreement, Orbital will make its open-source AI model, Orb, available to AWS customers. This is the first AI model for materials science to be offered on AWS platforms, accessible through Amazon SageMaker JumpStart and AWS Marketplace. According to the company's website, Orb will enable AWS customers working on advanced materials and technologies, such as semiconductors, batteries, and electronics, to access market-leading, accelerated research and development in a secure and unified cloud environment.

Orbital will also pre-train and fine-tune its advanced foundation models using Amazon SageMaker HyperPod. SageMaker is a service provided by Amazon which helps developers build, train, and deploy machine learning models. Furthermore, Orbital plans to explore the use of AWS's custom silicon, Trainium, to improve cost performance for its deep learning workloads. 

In alignment with Amazon's goal of achieving net-zero carbon emissions by 2040, the company is implementing additional measures to improve the energy efficiency of its data centers. These include developing a more efficient cooling system that reduces energy consumption without increasing water usage, optimizing rack placement to enhance power efficiency, and installing renewable diesel backup generators that can reduce greenhouse gas emissions (GEI) by up to 90% compared to fossil diesel. Moreover, Amazon is reducing the embodied carbon in its data center concrete by up to 35% and simplifying electrical and mechanical systems to ensure high availability. Together, these measures are expected to cut overall energy consumption by 46%.

AI: Environmental Challenges and Opportunities

In addition to its potential for decarbonization, AI offers opportunities to advance global adaptation to climate change and alert populations to its most devastating and deadly effects, such as floods, wildfires, and heatwaves, which result in significant loss of life and economic costs.

However, this technology is significantly more energy-intensive than traditional data center activities, according to experts. The World Economic Forum (WEF) estimates that AI systems already consume 33% more energy to complete tasks compared to traditional software designed for the same operations.

For instance, Microsoft revealed that its CO2 emissions had increased by nearly 30% since 2020, largely due to the expansion of data centers. Similarly, Google’s GHG emissions in 2023 showed a 50% increase from 2019, primarily due to the energy demands of data centers.

Balancing the challenges and opportunities of AI will be crucial to maximizing its benefits. According to the Boston Consulting Group, AI has the potential to reduce GHG emissions by 5% to 10% by 2030. Achieving this will require actions such as developing energy-efficient hardware, optimizing cooling systems, reducing overall data usage, and employing smaller, resource-efficient language models for specific tasks, as highlighted by WEF. 

Adam Elman, Head of Sustainability, Google in Europe, the Middle East, and Africa, noted that while AI is not the sole solution to the environmental crisis, it can play a vital role. "The good news is that the technology is already here, and we do not have to wait five or 10 years," he said in an interview with EFE. Additionally, Elman emphasized the importance of regulating AI at different levels, as well as providing accessible data to everyone in an understandable language to maximize the technology’s positive potential.