Enterprises Seek to Offset Their Own AI Carbon Emissions

AI’s effect on carbon emissions can be underscored by the commitments toward carbon neutrality (or better) previously set by Microsoft, Google, and Amazon.

Net zero by 2040 (Amazon). 24/7 carbon-free by 2030 (Google). Microsoft was boldest, pledging to be carbon-negative by 2030—removing more carbon from the atmosphere than it emitted.

But AI’s rapid advancement and the resulting increase in AI carbon emissions have dramatically altered these trajectories, clouding these goals with doubt and raising a crucial question mark: is AI a source for good in the climate movement?

Enterprise organizations are willing to put their carbon reduction goals on hold, therefore increasing AI carbon emissions, for the simple reason that they must in order to compete with the recent waves of AI-driven tech. Driving the rapid investment in AI tech is its immense potential for increased efficiency and revenue; according to Bank of America research, AI is poised to drive margin expansion for 23 of 25 industry groups.

Google's recent disclosure paints a stark picture: between 2019 and 2023, the company's greenhouse gas emissions surged by 48%, with a 13% increase in 2023 alone. This spike in AI carbon emissions can primarily be attributed to the escalating electricity consumption of data centers supporting AI operations—a single AI query now consumes nearly ten times the energy of a traditional Google search. Microsoft reported a 29% increase in emissions since 2020, acknowledging that AI carbon emissions are jeopardizing its "moonshot" target of becoming carbon-negative by 2030. Amazon, while yet to release its 2023 climate report, anticipates comparable hurdles as it deepens its investment in AI.

Corporate environmental strategies have changed tone considerably since carbon. Google stepped back from claims of carbon neutrality and no longer purchases carbon credits to offset annual emissions. While still aiming for net-zero emissions by 2030, the company now describes this goal as "very ambitious." There's a growing recognition that the method of achieving net zero may be more crucial than the speed at which it's accomplished.

Despite these challenges, AI also presents unique opportunities, with many remaining optimistic that it will mitigate its AI carbon emissions and drive broader sustainability efforts. As Microsoft CEO Satya Nadella said in 2023, “Although this new era promises great opportunity, it demands even greater responsibility from companies like ours.”

Environmental Cost: Understanding AI Carbon Emissions

In what specific ways are AI carbon emissions adding to anthropogenic climate change? The rapid growth of AI technology has led to several concerning environmental issues:

• Carbon Footprint of Large AI Models: Training large AI models requires immense computational power, resulting in substantial AI carbon emissions. For instance, training a single large language model can produce carbon emissions equivalent to the lifetime emissions of five cars. As models grow in size and complexity, this carbon footprint continues to expand. The energy-intensive nature of training these models, often requiring weeks or months of continuous computation, contributes significantly to the tech industry's growing carbon footprint.

• High Energy Consumption of AI Operations: Data centers housing AI infrastructure consume vast amounts of electricity, often relying on power grids still dependent on fossil fuels, which contributes significantly to AI carbon emissions. The continuous nature of many AI applications, such as recommendation systems or real-time data processing, means this energy consumption is ongoing and cumulative.

• Water Usage for Cooling Infrastructure: Intense computational activity necessitates extensive cooling systems. Data centers use millions of gallons of water annually for cooling purposes. In regions already facing water scarcity, this high water usage can strain local resources and ecosystems. For example, some estimates suggest that by 2027, AI-related water usage could reach billions of cubic meters annually.

• Electronic Waste from AI Hardware: The rapid advancement of AI technology leads to frequent hardware upgrades. Specialized AI hardware, such as graphics processing units (GPUs) and tensor processing units (TPUs), often have shorter lifespans than traditional computing equipment. The production and disposal of this hardware involve rare earth elements and toxic materials, posing additional environmental and health risks if not properly managed.

Despite the current state of unsustainable AI carbon emissions, organizations have the opportunity to offset AI carbon emissions through technological innovation, energy optimization, more efficient consumer products, and other strategies. Below, we’ll cover some of the strategies and technologies companies are using to balance the scale.

AI-driven Climate Solutions

Despite the growing recognition of AI's potential in addressing climate change, there's a significant gap in leadership vision. A 2022 report from Boston Consulting Group revealed that only 43% of leaders overseeing climate and AI topics have a clear strategy for using AI to combat climate change. This disconnect also highlights a crucial opportunity for enterprise organizations to take charge of developing and implementing solutions to AI carbon emissions. 

The following examples demonstrate how forward-thinking companies are already harnessing AI's power to tackle various environmental challenges, optimize energy usage, and more. These initiatives showcase the versatility of AI, provide inspiration to other organizations, and offer the optimistic hypothesis that AI-driven climate solutions can actually offset their own carbon emissions.

‍

‍

Energy Efficiency and Optimization

Google DeepMind is a prime example of AI's potential to address climate challenges. In data centers, DeepMind's AI-controlled cooling systems have achieved up to a 40% reduction in cooling costs, significantly enhancing energy efficiency. This approach not only sets a precedent for eco-friendly technology infrastructure but also showcases how AI can optimize existing systems for greater sustainability.

While DeepMind’s optimizations for reducing AI carbon emissions aren’t available to outside organizations, several third-party companies now offer offer AI-powered cooling optimization solutions, including Vigilent and BrainBox AI.

DeepMind is leveraging AI to address broader climate issues through various initiatives. Their precipitation nowcasting system and advanced machine learning-based atmospheric models enhance weather and climate forecasting. In the transportation sector, DeepMind's "Eco-friendly routing" and "Green Light" project optimize driving routes and traffic light systems to reduce emissions.

DeepMind's climate efforts also focus on adaptation and disaster response, with initiatives like Flood Forecasting and wildfire prediction models. Through "Project Contrails" and their "Startups for Sustainable Development" program, they're tackling aviation's climate impact and mentoring impact startups.

Renewable Energy and Carbon Offsetting

To reduce AI carbon emissions, Salesforce is pioneering the development of smaller, more sustainable AI models while prioritizing energy-efficient hardware for training and deploying AI systems. The company utilizes graphics processing units (GPUs) that offer better performance and higher efficiency, as well as specialized hardware like Tensor Processing Units (TPUs) optimized for AI workloads. This approach minimizes the environmental footprint of Salesforce's AI operations while maintaining high-performance standards.

In addition to hardware optimization, Salesforce has strategically chosen to train its AI models in data centers with significantly lower carbon emissions. By selecting facilities powered by electricity that emits 68.8% less carbon than the global average, Salesforce has achieved savings of approximately 105 tons of carbon dioxide equivalents (tCO2e) compared to using data centers with average global carbon intensity.

AI-supported Biodiversity Conservation

Microsoft's AI for Earth program exemplifies how AI can support biodiversity conservation efforts. With a $50 million, 5-year commitment, the program has supported over 700 grants across more than 100 countries. Projects span a wide range, from precision agriculture to analyzing mosquito blood samples for ecosystem data. The program's Planetary Computer initiative aims to democratize access to environmental and geospatial datasets (hosted on Azure cloud); the program also creates open-source tools to accelerate research and conservation efforts worldwide.

Deforestation Monitoring

Descartes Labs is harnessing AI to monitor deforestation on a global scale. By combining multi-sensor satellite imagery analysis with machine learning, they can detect forest changes in near real-time, even in frequently cloudy areas. This technology enables rapid response to new deforestation events and provides valuable insights for industries ranging from consumer goods to mining.

AI-powered Outage Prediction

In the energy sector, Duke Energy partnered with Accenture to develop an AI-powered Outage Prediction Model. This platform analyzes weather forecasts, historical outage information, and geographical data to predict potential power outages with greater accuracy. The system has helped Duke Energy reduce outage duration, improve resource allocation efficiency, and enhance customer communication during severe weather events.

Climate-Conscious AI-enhanced Consumer Products

AI offers an opportunity for organizations to create innovative, climate-conscious consumer solutions as well.

Boston-based startup Cala Systems is developing an AI-powered heat pump water heater. By predicting household hot water demand and optimizing energy usage based on factors like weather forecasts and energy pricing, this system aims to eliminate cold showers while maximizing energy efficiency. With water heating accounting for about 20% of typical American household energy usage, such innovations could significantly impact residential energy consumption.

Regulatory Advocacy and Industry Standards

For AI to reach the lofty goal of offsetting its own carbon emissions, sustainable AI practices ane regulations on AI carbon emissions may need to be a legal mandate rather than a mere suggestion.

Salesforce is taking a leading role in this advocacy work, actively lobbying for new regulations that would require companies to disclose their AI-related emissions and establish efficiency standards for AI systems. Salesforce has introduced Sustainable AI Policy Principles to guide AI regulation and is supporting specific legislative efforts like the Artificial Intelligence Environmental Impacts Act of 2024 and the Transformational AI to Modernize the Economy (TAME) legislation.

To complement its advocacy efforts, Salesforce is investing $2 million to support organizations focused on using AI to accelerate decarbonization and has launched the Salesforce Accelerator – AI for Impact program to empower nonprofits in leveraging AI for climate action.

Salesforce emphasizes the importance of transparency to create awareness and drive positive change in the industry, urging companies using general purpose AI models to publicly share data on energy efficiency and carbon footprint. The company also proposes that the environmental impact of AI systems be considered as a risk factor when classifying high-risk models. Salesforce’s initiatives underline the need for both regulatory action and industry-wide commitment to addressing environmental challenges created by AI.

Future Outlook

As AI continues to evolve, so too will the strategies for mitigating AI carbon emissions. Emerging research in eco-friendly AI focuses on developing more energy-efficient algorithms and hardware. The potential for AI to drive further innovations in climate solutions is vast, from optimizing renewable energy systems to enhancing climate modeling and prediction capabilities.

However, realizing this potential requires a concerted effort from tech companies, policymakers, and researchers, especially as AI adoption moves at a much faster clip than anticipated even five years ago. As we navigate the complex relationship between AI and the environment, it's crucial to prioritize responsible development practices that consider the full lifecycle impact of AI systems and AI carbon emissions.

‍By continuing to invest in energy-efficient AI technologies, leveraging AI for climate solutions, and advocating for responsible AI policies, we can work towards a future where AI cancels out its own environmental footprint and becomes a net positive force for global sustainability.