Every saved file, every replicated dataset, every backup created has a carbon cost. We store to avoid losing things, to stay compliant, to keep a trace. But this reflex of keeping everything, just in case, has quietly become one of the most polluting IT choices an organization can make. In 2026, the question is no longer whether your data has a footprint. It’s whether your architecture forces you to have one.
A weight the tech industry has long underestimated
The digital sector now accounts for 4.4% of France’s carbon footprint, representing 29.5 million tonnes of CO2 equivalent emitted in 2022, a figure on par with the total emissions of the country’s entire heavy goods transport sector (source: ADEME-Arcep, January 2025). What has shifted dramatically since earlier estimates is the share attributed to data centers: it jumped from 16% in 2020 to 46% in 2025, nearly matching the 50% attributed to end-user devices.
The explanation comes down to one number: 53% of French digital usage is hosted abroad, in data centers running on electricity mixes far more carbon-intensive than the French grid (source: ADEME-Arcep, January 2025). Electricity consumed in France is on average 10 times less carbon-intensive than that of the main data center host countries such as the United States (source: French Senate report, June 2020), which mechanically amplifies the footprint of French digital usage processed abroad. The location of the server, which nobody chooses in their browser, determines a large share of the real-world impact.
Globally, data centers absorbed 310.6 terawatt-hours in 2024, a 74% increase since 2019, now representing over 1.7% of total global electricity demand according to the International Energy Agency (source: Structure Research ESG 2025, via Allegro Informatique). In major data center hubs such as the United States, Ireland, and Singapore, this share can exceed 10% of national consumption. Cooling alone accounts for 40% of total energy consumed by these facilities (source: Greenly, 2025). According to a French Senate report, the carbon footprint of France’s digital sector is on course to increase by 60% by 2040 if current trends hold.
Default storage: an environmentally costly reflex
Enterprise data culture remains one of abundance: collect everything, keep everything, delete nothing. Retention regulations, fear of losing useful history, and the absence of data lifecycle policies combine to produce the same outcome: servers running continuously to host files nobody ever looks at.
Veritas has measured the scale: 52% of all data stored by companies worldwide is « dark data », meaning unused, obsolete data whose content IT managers are often unaware of (source: Veritas, via Le Monde Informatique). Global data storage reached 175 zettabytes in 2025, five times the 2018 figure according to IDC (source: IDC Data Age 2025, via Le Monde Informatique), of which 91 zettabytes would be dark data. Storing this unused data alone generates 6.4 million tonnes of CO2 needlessly each year: the equivalent of the annual carbon consumption of 80 countries (source: Veritas, via Global Security Mag).
This figure is all the more striking given that storage is typically framed as a cheap resource. The unit cost per gigabyte has indeed fallen, but the volume stored has exploded to such a degree that the overall bill, financial and carbon alike, has risen with it. As we outlined in our article on the hidden costs of cloud, the promise of « pay only for what you use » requires actually understanding what you consume. For most organizations, the true footprint of phantom storage has never been calculated.
Scope 3 is coming, and IT infrastructure is in it
For a long time, the digital carbon footprint of businesses was invisible in their reports. Cloud storage, outbound data flows, energy consumed by production servers: none of it appeared in ESG disclosures. That era is ending.
The CSRD directive, which came into force on January 1, 2024 for large companies already subject to the NFRD, requires the publication of an emissions report covering scopes 1, 2, and 3, in line with ESRS standards (source: economie.gouv.fr). Scope 3 is precisely the category that captures indirect emissions from the upstream and downstream value chain, including outsourced IT infrastructure. The Omnibus package adopted by the European Parliament in December 2025 reduced the scope of application from approximately 50,000 to around 10,000 companies in the European Union, with a two-year delay for waves 2 and 3 (source: Réglementation-environnement.com, 2026). But for large groups in wave 1, scope 3 reporting obligations remain.
The cascade effect extends further still. SMEs not directly subject to CSRD will need to provide structured ESG data to their customers who are (source: Claire Gérardin, 2026). Digital infrastructure will have to become visible across the entire supply chain. Yet the question is almost never asked in current carbon audits: how much data is your information system storing that nobody has accessed in years? What share of your scope 3 represents the energy consumed by servers hosting files that haven’t seen a single query in half a decade?
Processing without storing: structural sobriety, not cosmetic effort
Digital sobriety is often reduced to its most visible forms: switching off idle equipment, clearing inboxes, deactivating unused services. These actions have their place, but they operate at the margins of an architectural problem.
The real lever lies upstream, in the design of processing pipelines themselves. The dominant logic remains that of intermediate storage: data is collected, then stored, then processed, then analyzed, sometimes replicated across systems. Each step generates a footprint, and each duplication multiplies active storage surfaces. This assumption of storing first in order to process is not a technical inevitability. It is the product of a design habit.
An architecture that processes data during transit, without ever persisting it in an intermediate space, mechanically reduces the storage footprint to zero at the processing layer. This is the principle underlying iD4Connect’s DataCell: each processing unit acts on the data during its passage (cleaning, transformation, enrichment, routing) then releases it. Nothing accumulates on the middleware side. The DataGraph orchestrates these units without ever centralizing data in an intermediate hub. The result is not only a performance or cost gain: it is an architecture that is lean by design, with a structurally lower environmental footprint than a conventional pipeline.
A few figures to reframe the debate:
4.4%: digital sector’s share of France’s carbon footprint (ADEME-Arcep, January 2025)
46%: data centers’ share of digital carbon emissions, up from 16% in 2020 (ADEME-Arcep, January 2025)
53%: share of French digital usage hosted abroad, on more carbon-intensive electricity mixes (ADEME, 2025)
52%: proportion of enterprise-stored data that is unused or obsolete, meaning « dark data » (Veritas)
6.4 million tonnes of CO2: annual emissions generated solely by storing dark data worldwide (Veritas)
60%: projected increase in France’s digital sector carbon footprint by 2040 if current trends continue (French Senate report, 2020)
Three questions every CIO should ask before approving the next architecture
1. Do you know how much data your IT system stores without ever reading it back? Dark data doesn’t show up in standard dashboards. It doesn’t trigger alerts, doesn’t slow down performance, doesn’t flag any monitoring system. It simply consumes energy, silently, every second. A data lifecycle audit (how much hasn’t been accessed in more than six months, more than a year, more than three years) is the most immediate entry point for understanding a system’s real footprint.
2. Does your digital carbon footprint account for the lifecycle of your data? IT carbon assessments typically focus on physical equipment and metered electricity consumption. Egress costs, phantom storage, forgotten automated replications: these line items almost never appear in current scope 3 analyses. Yet this is precisely where a growing share of real-world footprint is generated, as we detailed in our article on the hidden costs of cloud.
3. Can your processing architectures function without storing by default? The goal isn’t to eliminate all storage: some data must be persisted. The question is whether you can distinguish data that genuinely needs to be stored from data that is stored out of habit or architectural default. A pipeline that processes in transit and stores only what has documented value is, by design, more resource-efficient than one that centralizes first and processes second.
Digital sobriety cannot be decreed in a CSR charter. It is built into architectural choices, data lifecycle policies, and the honest assessment of what an organization’s servers are actually hosting. Not storing, when processing can do without it, is not a constraint: it is a design decision. And in 2026, it is also an ecological one.
Discover how iD4Connect processes data without ever storing it →