About the social and environmental impact of the ITC sector

Written in April 2025, edited in August 2025 for public release, many thanks to Dan and Veronica for the proofreading and feedback.

As a member of an organisation working towards collective justice I have been asked to explain what would be the “gain” of stopping using Google Workspace (although this applies to most tech giants) in favour of alternative services – at least more privacy preserving, ideally using open-source / libre software and run by not-for-profit, like-minded organisations.

A social and democratic cost

Were they ever on our side?

For me it is a question of what kind of organisation we want to be. Using Google Workspace the way we are using it means giving away our members personal data but also those of most of the people we are interacting with to companies that have been known to collaborate with law enforcement and intelligence agencies in a country where, for the second time in a few years, has been overtaken by an openly fascist administration. An administration that would heavily discriminate against a majority of the groups and people we are trying to attract, if they don’t deem their existence illegal.

For me, that alone only should be a good enough reason to chose to change provider. But let’s dive a little bit more into the subject.

First, requiring form respondents to have a Google account (required for file upload with Google Form), while trying to be attractive to climate justice activist who might be sensible to digital security (XR France guide on digital hygiene (in French), XR Netherlands guide on digital tools), is for me not a sensible position. This is especially true in a period where a number of climate activists are facing trial or being prosecuted, for example in France where the state repression against climate activist is only getting worse (in French). But the increase of digital-base surveillance is also true for other countries that have been known to try or succeed in reducing privacy and increasing mass surveillance. For example in the UK where the government demanded that Apple introduced a backdoor to its new encrypted iCloud services (the company responded by disabling the feature for UK users), in Sweden where the government asked the encrypted messaging service Signal access to the messages by introducing a backdoor (Signal responded similarly to Apple and would withdraw from the country if the law would come to pass), in France again where the recent Narcotrafic Law Project plans to extend global digital surveillance in the country (in French).

Tech giants have been harmful to democracy for years, and while they tried to maintain a figure of progressive Silicon Valley companies, this is no longer true – although it is difficult to believe they ever were. From the role of Facebook in the Cambridge Analytica case, turning the 2016 election in the US, and the possible involvement in multiple elections before that, to the role of TikTok in the 2025 election in Romania. In addition to the front page scandals, there are also the day to day failures to welcome marginalised voices and provide the required moderation tools. Threads (Meta), in 2024, failed to moderate appropriately hate speech while shadow-banning LGBT activists. The Vagina Museum, a London museum “dedicated to vaginas, vulvas and the gynae anatomy” is shadow-banned on Instagram, banned on Facebook and TikTok, and automatically flagged as “adult content” on BlueSky as they explain in a thread on Mastodon.
The common goal of those companies is, in the end, to make as much money as they can: 7 out of 10 of the companies with the biggest by market capitalisation are tech companies, among them, Apple, Microsoft, Alphabet (Google), Amazon and Meta (Facebook, Instagram, WhatsApp). Companies like Meta have already proven that they were willing to put their profit before their users safety and well being, by having captive platforms that thrive on spreading hate, anger and fear, as those emotional responses are the best to keep the users engaged, along with mechanisms feeding us dopamine, which resemble those from slot machines and casinos.
Today, those companies have no problem bowing to a fascist administration, if that means keeping their current power, as the events of the past few months have shown. This has lead to big tech stopping diversity, equity, and inclusion (DEI) programs,updating their moderation rules for more lax ones, stopping their fact checking programs or simply complying with the latest presidential whims. What we are witnessing is a tech bros oligarchy: a broligarchy.

The post-modern mechanical Turk

The social impact of tech giant goes beyond their democratic impact, and the harm toward marginalised communities is real and direct. The fast growing OpenAI, most recently known for stealing the famous hand-drawn work of Miyazaki to train its model and entertain the global Internet community, at an astronomic ecological cost, employs African workers at the extremely low hourly wage of 2$ to do the handiwork behind its famous model ChatGPT. A 2023 case revealed that the training of the moderation model of ChatGPT, which prevents the millions of ChatGPT users to see the horrors of a statistical generation of Internet scraped content, was done by a contractor employing Kenyan workers whose job was to manually label content, text and images, from the darkest corners of the Internet with little to no psychological support. Such cases of outsourcing to developing countries is common for the Silicon Valley giants. This contractor was revealed to also employ workers in India and Uganda for clients such as Google, Meta or Microsoft.
This is the flip side of AI: an industry who casually employs underpaid, unprotected workers, mostly from the global South, to train and correct their models while congratulating themselves for the technological “progress” that we are witnessing.

Beyond the training phase, the advent of AI and automated systems (which are now becoming more and more opaque with the increasing usage of neural networks) to solve every problems leads to more inequalities while giving the organisations using those systems a sense of neutrality, and a good excuse: if it is a computer – “an algorithm” – it surely is less biased that humans are. The issue being that biases are encoded in such systems, whether it comes from the input data, which by essence reflects an already oppressive society, or the team of developers whose lack of diversity failed to account for systemic issues and so produce harmful products. One would think that the issue is benign, and it may be true if you’re using AI to generate funny nonsensical text or write boring emails, but it can have dramatic consequences when AI chatbots are masquerading as psychologists or AI system are built to predict crime in the US (and soon murder in the UK). Generative language models, by being statistical models, or stochastic parrots, will deem as true an information if they find it a large number of time. The Russian misinformation network took advantage of that flaw and published millions of propaganda articles, resulting in AI chatbots propagating the Russian narrative. The difficulty of AI to explain its results makes misinformation all the more difficult to identify, and the ease of use makes it trivial to generate content, flooding the Web with AI generated articles and pictures.

Finally AI models require a very large training corpus. All the companies aggressively competing for the best model, scrape content from all over the web, putting regular websites under heavy load. As yet another hidden side effect of the AI age, personal websites (article in French), open-source software infrastructure or digital commons are now forced to play a rigged game of catch to try to escape malicious robots if they don’t want to see their hosting costs surging, as it is the case for Wikipedia, or simply becoming unavailable if they can’t handle the load. We already knew that AI was feeding on stolen content, but this also means that AI companies are externalising the cost of gathering the content directly to the ones they’re stealing from.

The materiality of the cloud

Behind the metaphors and the marketing campaigns that would almost convince us that digital services are nothing but virtual, hanging on thin air in the sky, there is an all too real infrastructure. From blocky warehouses, huge storage and computation hubs, to undersea information highways that run around the world, without forgetting the device in your hands or pocket, on which you may be reading this.

Of energy and water

There are multiple approaches to analysing the ecological impact of information technologies as the technologies are diverse in usage and types. A common approach is to distinguish the different layers of the infrastructure: the datacentres, the network, the end-user equipment, the connected objects. For each of those categories one would need to study the manufacturing, use, and disposal of the equipment. Without going in too much detail, end-user devices account for the biggest source of impact, estimated between 60% and 84%, depending on the indicator, of the total impact in 2019, due to the overwhelming number of devices. In 2022 the total number of smartphones was estimated to 16 billion worldwide, with a slowly increasing replacement cycle between 24 and 36 months (replacement cycle estimate in the US, replacement cycle estimate in India). On average, users have just above 8 devices each, with very large geographical disparities.

While end-user devices have a much bigger impact in the manufacturing phase, it is in the use phase that the impact of digital infrastructure is the largest. While the energy required to power the digital world accounts for two thirds of its total energy consumption, more than half of this energy required in the use phase is consumed by the infrastructure. As a substantial share of that power comes from fossil fuel energy, most green house gas emission are concentrated in the use phase (Green IT 2019 study).

Let’s focus now on AI costs. As we have seen above, the futuristic promise of an automated world and artificial intelligence is built on the invisible labour of thousands of underpaid workers, mainly in Africa. But the AI boom also comes at a cost for the environment as power and water consumption surges. The high computational demand of AI means that AI datacentres have a larger impact than other datacentres, it is estimated than a ChatGPT search uses 10 times the energy required to perform a Google search. This leads to a spike in water and electricity demand, putting more pressure to an already stressed infrastructure. As the majority of the water used by datacentres for cooling is drinkable water, this is increasingly worrying for the local population, who if not already facing water shortage might in a near future as a result of global warming, as it could be the case in South US, in states like Arizona or Texas, the northern Spain region of Aragon or in South-East England. In combination with the growing number of wild fires, this is painting a potentially grim picture of our future. The same goes for electricity, for instance in Ireland where the high concentration of datacentres in the region has been correlated to the rise of electricity prices. At this rate, the progress in both water reduction and low-carbon or renewable energy production could be wiped out by the rapidly growing demand and the hundreds of planned datacentres that could emerge from the ground in the coming years.

Advertisement remains a main source of revenue for some of the tech giants, with 97% of Meta revenue coming from advertisement 2024 and 75% for Google. While it is hard to estimate the share of advertisement in the global impact of digital technologies, it has been shown to be significant. One model estimates it to around 10% of total infrastructure related emissions. A 2016 study found that around 50% of the data download while reading news article was from ads, a number that may not have changed much as a quick check with the French news website, Le Monde, revealed a 74% increase in data transferred when not using an ad blocker. The same test with The Guardian showed a 92% increase. Ad blockers are not only beneficial for the environment but also increase security (article in French) and privacy, and reduce brain pollution. A 2015 study estimate that an individual in the US could be exposed to 4,000 to 10,000 ads daily, and that by the time they turn 13, “ad-tech firms would have gathered 72 million data points” on children.

From dust, to dust

Digital devices require a large amount of different material to manufacture. A single 150 g smartphone requires around 180 kg of raw materials (L’enfer numérique: Voyage au bout d’un like (book in French)) and 12,700 litres of water, for a total of no less than 60 different metals. To create an object that will fit in the palm of your hand, metal ores are extracted from dozen of countries like Chile, China, the Democratic Republic of Congo, Bolivia, Kazakhstan, Russia and more. The workers conditions are, in many cases, dangerous if not deadly, potentially operating in artisanal mines, which still represent a significant share of the global production, up to 25% for tin production and from 17% to 20% for gold production. Metals have become such a strategic resources that armed factions are now fighting over the exploitation of mines (video in French) with mega corporations pulling the strings and filling their pockets.
In addition to its catastrophic human and social cost, the local environment is also paying a high price of the extraction of the precious ores, with heavy metal pollution in both land and water endangering the local ecosystem and population, like in Brazil for gold mining, China for graphite or cobalt mining in DRC, to just name a few.

As global demand for critical minerals used in electronic hardware and for renewable energies grows, industries are looming over possibilities of expansion underwater, now that exploratory licences for deep sea mining have been granted. Deep sea mining faces a strong opposition of environmental organisations as it could result in disruption of the marine ecosystem and biodiversity while disregarding the impact on local indigenous populations.

The technological complexity of electronic devices make them hard to dispose of. With 5 billions of smartphones thrown away in 2022, the e-waste mountain keeps growing, year after year. Of that growing mass, just around 20% is being properly recycled, and the lack of proper recycling facilities leads to illegal shipment of e-waste from developed countries to developing countries. The extraction of the precious metals has become a source of income for hundreds of thousands of people worldwide who have turned to the informal sector of electronic devices recycling, mainly in South America, Africa or South-East Asia, where the rules and legislation are less strict. The processing of e-waste can be particularly harmful as it requires burning or melting the devices to recover only the valuable materials, such as gold or copper, exposing the workers to toxic fumes. E-wastes can also end up in dumpsite where scavengers will try to find valuables parts that can be sold. Toxic elements are then carried by the rain and finish their journey in the soil or water, endangering the local population and wildlife.

From the dust ore mined in artisanal mines to the e-waste powder polluting the soil, lies the shadow of neo-colonialism. As the 500 years old mine in Bolivia, the mountain that eats men, continues to take workers, we continue to take silver. It may not be the Spanish empire mining silver in the 16th century any more, instead the empires have taken the names of global corporations, advertising noble goals of connecting people, but make no mistake, their greed for power has not changed, and we are the fuel.

Now, what?

It may be hard to not feel overwhelmed and powerless in the face of a global issue, as is the case with most social or environmental battles, but I hope that, at least within our collectives and organisations, we could do our best to mitigate the hold of the tech giants, the same way we choose to eat organic and local food. It will not change the world, but as I recently heard in a conference, with lots of complicated words (video in French), it shows that a different world is possible. Free software have flaws, a lot, but I think they are closer to an ideal of having software made for users, as digital commons. Software that takes into account the experience of many and does not try to profit from our data and behaviour, present and future.

I don’t think that we could establish a straightforward correlation between stopping using big tech services and reducing the ecological impact of IT. One could even argue that big companies can benefit from their large scale as they could be able to make a better use of their resources. That being said, this advantage is balanced by the extensive number of services that those companies are offering. Google itself has more than 200 services, whereas alternative providers may have a smaller catalogue (maybe also to their detriment) with services that have less features, thus requiring less resources. No need of thousands of graphic cards, millions of litres of water and a private nuclear plant to train your AI assistant service to help you write e-mails if you don’t have one.

On a personal level, if you are willing to do something, the first thing I would recommend is install an ad blocker on your web browser, specifically uBlock Origin, and try to use Firefox, as Google Chrome (and its derivatives) have recently restrained the effectiveness of such blockers. Try to opt-out of cookies and trackers when presented a cookie banner. Think about the business model of the services you use, and try to take care of your data, it’s more important than we may think. If you are willing to do more, ask me and I will give you some pointers.

If you want to dig a little further into the subject and like listening or watching stuff.

If there are any articles or resources linked here that you would like to consult but are being a paywall, ask me and I’ll send you a copy.