Early Numbers Show How GenAI Can Reduce Agencies' Carbon Footprint

In May this year, WPP partnered with chipmaker Nvidia to create ads using generative artificial intelligence. Now, numbers are emerging on how environmentally beneficial AI is for campaign creation. The agency's transition to AI-powered content production could save up to 93% of its carbon footprint compared to that of traditional production, according to the agency. What's the Carbon Footprint of Going to the Cannes Lions Festival? “If you purely use AI to generate images versus having people doing it with edit suites, you're talking about 1300 times less carbon emissions," said Priti Mhatre, global head of marketing transformation at WPP’s Hogarth Worldwide. However, as more agencies adopt generative AI tools, there are critical human labor costs and even more impactful sustainability tradeoffs to consider. Lowering costs of production WPP’s calculations are based on AdGreen data that compared the carbon footprint of traditional production to that of virtual production. The positive environmental impact has led to most of its clients opting for this content engine. Alongside sustainability benefits, "the demand is driven by multiple factors, including time and cost efficiencies, the ability to generate personalized content at scale, and generative AI," said Mhatre. So far, five of WPP's clients, including automotive and consumer packaged goods (CPG) brands, are testing its AI-powered content engine for multiple use cases, like featuring a car on a bustling Tokyo street or setting it against the backdrop of the Eiffel Tower in Paris. WPP's content engine lets the agency’s creative teams integrate content from Adobe, Shutter Stock and Getty Images with generative AI to produce campaign assets like product images or videos. It sits on Nvidia’s Omniverse cloud services and is available through Microsoft’s Azure. Pet Food Has a Massive Carbon Footprint. Bug-Based Proteins Can Change That Other agencies like Momentum WorldWide and Media.Monks are feeding customer data, such as buying habits, into large language models (LLM). Based on that data, the LLMs can render images at a scale that is relevant to end consumers, according to Lewis Smithingham, svp of innovation, Media.Monks. “The actual waste that occurs is by non-targeted content and the creation of ads that are barely watched,” said Smithingham. “And as we start serving more targeted ads through generative AI, there’s a massive reduction in waste.” The devil's in the details LLMs such as ChatGPT can only be as efficient as the data they are trained on, which takes place within data centers across the globe. A traditional data center runs on electricity and requires servers, cooling, and lighting amenities. All this contributes to a total of 160,268 kgCO2e (kilograms of carbon dioxide equivalent) annually, according to data shared by the carbon management platform Greenly. This is equivalent to the greenhouse gas emissions of 35.7 gas-powered cars driven for one year. “The explosion of generative AI is leading to an explosion of data centers, electricity consumption, leading to an explosion of carbon footprint to power up these data centers,” said Alexis Normand, CEO & co-founder at Greenly. What's more, the location of the data centers, due to the carbon footprint of the local grid's power makeup, also contributes to the carbon footprint. Data centers in Arizona or Kansas run on coal and tend to consume more energy, whereas data centers situated in Oregon run on hydroelectricity and emit a comparatively lower carbon footprint. Even if they were all using 100% renewable energy, which is not the case, it's still a huge amount of energySasha Luccioni, researcher and climate lead, Hugging Face However, tech giants like Microsoft and OpenAI haven’t openly discussed or disclosed their data center locations. Microsoft’s blog does indicate the locations of a few data centers in regions such as India, Italy and the Middle East. Microsoft did not respond to requests for comment by press time. According to Sasha Luccioni, researcher and climate lead at AI company Hugging Face, Microsoft, which provides computing for OpenAI, has built a data center just for training these LLMs. Trillion-dollar company Nvidia’s Omniverse cloud centers sit among Microsoft’s physical data centers. “Even if they were all using 100% renewable energy, which is not the case, it's still a huge amount of energy,” said Luccioni. Beyond data centers, at the individual level, each query put into ChatGPT contributes to five grams of carbon dioxide—the equivalent of one email, according to Normand. When you factor in that ChatGPT receives about 25 million queries a day, according to Normand, that’s nearly 9,000 tons of CO2e per day, which is equivalent to the electricity use of roughly 1,589 homes for one year. Novel solutions However, agencies can take steps to further curtail their impact on the environment in terms of carbon emissions by training their LLMs during off-peak hours, according to Normand, as Microsoft did with Xbox, where substantial updates were scheduled for off-peak hours and not midday. Moreover, businesses utilizing Microsoft's Azure services have the option to request a transfer of their data centers from regions with high carbon intensity to those with lower intensity, according to Normand. Yet, “It's still frustrating that we don't have all the information we need to make these decisions,” said Hugging Face’s Luccioni.

AI News

Early Numbers Show How GenAI Can Reduce Agencies' Carbon Footprint

The tech still comes with human labor and more considerable environmental trade-offs

Lowering costs of production

Trishla Ostwal

About

Subscriptions

Events

Publications

Lowering costs of production

WORK SMARTER - LEARN, GROW AND BE INSPIRED.

Subscribe today!

Trishla Ostwal