On the company’s 20th anniversary, Meta’s sustainability report makes informative reading for anyone with an interest in big companies’ environmental impact.
The Meta Sustainability Report‘s [PDF] contents also serve as a useful guide to the intricacies of large technology companies’ power requirements, their approach to mitigating the environmental impact of their operations and some of the vagaries of so-called carbon trading.
In the report, the company states its “location-based” emissions grew significantly between 2019 and 2023, from around five million tons of CO2 to approaching 15 million tons in 2023. That’s partly due to the fact that wherever Meta locates its data centres (DCs), it usually has to buy at least some electricity from the local grid. And given that Meta’s DCs are largely US-based, the company’s energy comes from, on average, 60% non-renewable sources – the same average experienced by every other company or individual in the country. Location-based emissions are therefore unavoidable unless the company using power generates its own more environmentally-friendly electricity. Meta is actively trying to produce more of its power from generation facilities it builds and owns.
Meta brings power generation in-house
Until it can operate on the output of its own, green power generating facilities, Meta’s sustainability profile is not at an ecological optimum. Companies like Meta are able to publish CO2 emissions figures lower than what their operations actually create by purchasing renewable energy certificates (RECs). RECs are sold by renewable power generating companies and each is certified proof of the production of one megawatt-hour of electricity generated from any renewable source. The certificates were designed to be a way that energy buyers would be able to finance investment in further green energy projects to be undertaken by the generators.
By purchasing RECs, Meta effectively offsets a portion of ‘dirty’ energy consumption, and with enough RECs purchased, it can claim that it operates a zero-emission operating model. Put simply, if company A uses power that creates a ton of carbon in the atmosphere, it can additionally buy the same amount of power (in the form of a certificate) that produced no carbon.
The irony is that in some circumstances, the value of RECs can be so low that the energy producer selling renewable energy and corresponding certificates gains too little revenue to invest in their infrastructure to any significant degree. The main value of RECs in these instances is to large, polluting companies that value what the certificates represent, which is a way to improve their environmental appearance, if not their environmental performance.
In more technical language, offsetguide.org states: “The most common underlying objective in purchasing voluntary market RECs would be to report the indirect emissions associated with its consumption of purchased electricity (i.e., Scope 2 emissions) as zero. The problem with this zero-emissions claim is that REC purchases do not reflect, nor do they alter, the emissions physically associated with or caused by a firm’s electricity use.” The carbon is still produced, in this case by Meta.
It is worth noting, however, that Meta is a big investor in renewable energy projects in the areas it operates. At the end of 2022, for example, it announced seven solar projects to be constructed in Georgia and Tennessee producing 720 megawatts of power in total. In Clonee, Ireland, the company has deployed hydro-treated vegetable oil as a diesel replacement in its backup generators, initially as a proof of concept but with potential rollout to transition all its generators to the fuel in every data centre it operates across the globe.
Speaking to The Verge, Rachel Kitchin from environmental organization Stand.earth said, “I would say, from reading their [sustainability] report, it seems as though Meta has broadly pursued a high-impact approach to renewable energy.”
The second criticism levelled at data centres on environmental grounds is facilities’ use of water to cool the hardware in each centre. Water withdrawal from groundwater reserves by Meta went up 137% in 2023, with extraction from areas already suffering from high baseline water stress increasing by 20%. Water stress is the ratio of water demand against renewable supply. High water stress, in this context, means Meta is using more water in areas that already consume 40% of what’s available to them. The greater the water stress an area (or country) faces, the less excess water is available.
Meta increased its total water consumption overall by buying electricity from other sources, which used a further 55,475 megalitres in 2023 (one megalitre is one million litres).
Big companies’ environmental impact
Meta aims to be entirely carbon-neutral in real terms by 2030. But like many of the world’s large technology companies, the impetus to invest in machine learning technologies means the companies’ energy consumption goes up – both the learning and inference stages of what is termed artificial intelligence are highly power-intensive compared to ‘traditional’ computing. A standard Google search costs hundredths of a single cent in terms of the power used to process the search and present its results. Queries to OpenAI, or, in Meta’s case, Meta AI, cost an order of magnitude more. Exact figures are difficult to determine, and estimates of cost range between one and seventeen cents for each query, depending on the model used.
In its sustainability report, Meta stated, “Our newest AI-optimized data centers currently under construction will feature dry-cooling technology,” which will at least take away some of the pressure on groundwater reserves in the future but will not ameliorate the environmental impact of increased use of AI caused by using more electricity. The costs of data centre operations tends to be split around 60:40 between powering processors and cooling them down.
The increased demand for smarter search results couched in conversational language (AKA AI) means large tech companies like Meta are investing significant resources in new facilities. Meta continues its data centre building programme to fulfil that demand, with the company’s capital expenditure having reached $8.5bn in Q2 2024.
Given the choice between cashing-in on public demand for all things AI and low-impact environmental operations, big tech chooses to take the money and hope future generations will offset the effects of today’s commercial activities. Although there are some cases in which machine learning model use is beneficial to the human race, they are outnumbered by offerings that create fake media or purport to be ways to help sell more product. Addressing the issue of demand for technologies that have appalling consequences in the long term is a huge challenge.
