The environmental account of industrial digitalisation, with the numbers brought up to date

Digitalisation has an environmental cost, and an industrial digitalisation programme that leaves it out of the books is incomplete. This article sets out that cost first, then the case for digital technology as a tool for environmental improvement.

This is part 8 of 9 of the “OT/IT convergence” series; it continues with part 9, Purdue, ISA-95 and IEC 62443. It leans harder on published statistics than the rest of the series, and those statistics move fast, so every figure has been checked against the current primary source.

Electronic waste

The source document quotes the UN figures for 2019: a record 53.6 million tonnes, up 21% in five years, a projection of 74 million tonnes by 2030, and only 17.4% formally collected and recycled. Gold, silver, copper and platinum worth some USD 57 billion were mostly thrown away or burned rather than recovered.

Those figures are two editions old, and things have got worse. The Global E-waste Monitor 2024 reports 62 million tonnes generated in 2022, or 7.8 kg per person, and raises the 2030 projection to 82 million tonnes. The documented formal collection and recycling rate is 22.3%. That is better than 17.4%, but the UN’s own framing is the one to remember: e-waste generation is growing almost five times faster than documented recycling. The metals in that waste are worth an estimated USD 91 billion, of which about USD 28 billion is actually recovered.

The regional picture is starker than the tonnage. Documented collection and recycling rates reach 42.8% in Europe and 41.4% in Oceania, but 30% in the Americas, 11.8% in Asia and 0.7% in Africa. Asia generates almost half of the world’s e-waste. Europe generates the most per head, 17.6 kg per person.

(A correction to the source text: it gives 53.6 Mt for both 2019 and 2020 in different paragraphs. The figure is the 2019 total, published in the Global E-waste Monitor 2020.)

Greenhouse gases

The document says digital technologies emit 1.5 billion tonnes of greenhouse gases a year, 4% of the global total, growing at 8% a year. The figure is widely quoted and genuinely disputed. Estimates of the digital sector’s share of global emissions range from roughly 1.5% to 4%, depending on what is counted, so it is safer to quote a range than a single number.

The French figure needs a firmer correction. The document quotes 3.7% of French greenhouse gas emissions, about 15 million tonnes of CO2 a year. That is not the national reference. The reference series is the ADEME–Arcep assessment: its January 2022 edition put digital at 2.5% of France’s carbon footprint (16.9 MtCO2e, 2020 data), and its January 2025 update puts it at 4.4% of the carbon footprint in 2022, or 29.5 MtCO2e.

The jump is worth understanding, because it changes how you read the number. It is mostly methodological: the update counts the foreign data centres that serve French users, which the earlier edition left out. The split is now 50% terminals (televisions, computers, smartphones), 46% data centres and 4% networks.

That split matters to anyone specifying an industrial digitalisation programme. Terminals and data centres carry nearly all of the footprint; networks are close to a rounding error. How many screens you deploy and where your computing runs drive the environmental result. Network topology barely registers.

Rare minerals

Making electronic equipment means extracting rare minerals, with damaged ecosystems and polluted water as part of the price. This part of the account has no offsetting argument. It is a straight cost, and the only levers on it are buying less equipment and keeping it longer.

Energy, and the oversizing problem

The document quotes the IEA’s 2019 estimate of data centres at around 1% of global electricity use. That has moved a long way. The IEA now puts data centres at about 1.5% of world electricity in 2024, or 415 TWh, and expects that to more than double to around 945 TWh by 2030, slightly more than Japan uses today. AI is the main driver.

The more useful part of the document is not the total, though. It is the oversizing, and that argument has aged well.

Work by APC (now part of Schneider Electric) compared how server rooms are planned with what they actually draw. In the edition the document relies on:

• the planned start-up load was typically 30% of the planned final load;
• the actual final load turned out to be about 30% of installed capacity.

In other words, the typical installation ended up about three times larger than it needed to be, and at start-up the gap was closer to ten times. Later revisions of the same white paper give smaller ratios, around five times at start-up and one and a half times at final load, which is better but still a lot of idle equipment. Set that against a service life of about ten years and you get a large amount of hardware manufactured, powered, cooled and finally scrapped without ever doing useful work.

The oversizing finding is the most actionable item in this article. It needs no new technology and no new policy, only an honest estimate of the real load and the discipline not to add headroom on top of headroom that was already generous.

Two references help with that estimate: the Best Practice Guidelines of the EU Code of Conduct on Data Centre Energy Efficiency, updated every year by the European Commission’s Joint Research Centre, and APC White Paper 37, Avoiding Costs from Oversizing Data Center and Network Room Infrastructure.

On an industrial site, the same habit shows up in the server room next to the control room: the UPS and the air conditioning sized at two or three times the need “to be safe”, then running for a decade at partial load, where they are least efficient.

The other side: digital as an environmental tool

For all the criticism of its footprint, and with the energy crisis still fresh in people’s minds, digital technology can genuinely help industry pollute less.

In its December 2020 conclusions on digitalisation for the benefit of the environment, the Council of the EU recognised the potential of digital technology to help reach the goals of the European Green Deal, and the Green Deal has produced a series of initiatives built on it.

On a plant, the mechanisms are concrete. Sensors and connected meters show energy use in real time, line by line, instead of once a month on the utility bill. Better process control trims the overdosing of raw materials and the off-spec batches that end up as waste. Data analysis points to where the compressed air leaks and which machines idle at full power during breaks. Digital records cut paper, and remote support saves a fair number of flights and van trips. Traceability helps recycling and reuse. And the simple ability to see information as it happens is a precondition for running a process close to its optimum.

How to hold both halves at once

The two halves of this article do not contradict each other, but the balance does not take care of itself. A digitalisation programme reduces net environmental impact only when the savings it enables exceed the footprint it creates. That is arithmetic, not branding.

The practical rules are short. Size the infrastructure to the real load. Extend equipment life rather than refreshing on a calendar. Process at the edge when shipping data to a distant data centre adds nothing to the analysis. And measure the promised energy savings after deployment: a programme that projected a saving and never checked it has not made one.

Next, and last in this series: the regulatory and standards framework (Purdue, ISA-95, IEC 62443) that makes a secure transformation possible.

References

• UNITAR / ITU, The Global E-waste Monitor 2024: ewastemonitor.info.
• ADEME / Arcep, update of the environmental impact of digital technology in France (January 2025, 2022 data), summary by the French government’s responsible digital mission: numerique.gouv.fr.
• IEA, Energy and AI, April 2025: iea.org.
• N. Rasmussen, APC White Paper 37, Avoiding Costs from Oversizing Data Center and Network Room Infrastructure (APC / Schneider Electric): schneider-electric.com.

Every statistic has been checked against the current primary source at the date of publication.