The carbon footprint of sending a file
Sending a gigabyte costs a small and genuinely uncertain amount of energy — published estimates vary by more than a factor of ten — but storing that gigabyte for a year costs considerably more. That ratio is the only part of this worth acting on, and it points at deletion rather than at any decision about how you send things.
Why the numbers are so uncertain
Estimates for the energy cost of transferring data range from roughly 0.006 to over 0.1 kilowatt-hours per gigabyte depending on the study, the year, and what is counted.
The disagreement is not sloppiness. It comes from real methodological choices:
What is included? The network only, or also the data centre, the customer's router, and the device displaying the file?
Marginal or average? Network equipment draws power whether or not your file crosses it. The marginal cost of one more gigabyte is close to zero; the average cost, dividing total energy by total traffic, is much higher. Both are defensible and they differ by orders of magnitude.
Which year? Efficiency has improved fast. Figures from 2010 are wrong by a large factor today, and they are still widely cited.
So anyone quoting a confident per-gigabyte figure is being more precise than the evidence supports, including anyone doing it to sell you something. What follows uses ratios rather than absolutes, because the ratios are robust even when the absolutes are not.
The comparison that holds
Whatever figures you pick, this relationship survives:
Transfer is brief. Storage is continuous.
Sending a file consumes energy for the minutes the transfer takes. Storing it consumes energy every second it exists — drives powered, plus the cooling that keeps them at temperature.
A year is about 525,600 minutes. Even if storage draws far less power per second than a transfer does, keeping a file for a year is a much larger cost than sending it once. For most reasonable assumptions the storage of a file for a year exceeds the cost of transferring it by a wide margin, and the gap grows every year you keep it.
525,600×
Replication multiplies the storage side. Three copies across three regions is three times the continuous cost. Nothing multiplies the transfer side the same way.
What follows from that
Deleting beats compressing. Compression reduces both costs proportionally. Deletion removes them entirely. If you are choosing where to spend effort, it is not close.
Short retention beats renewable power. A service that deletes after two weeks has a smaller footprint per file than one keeping everything forever on renewable electricity, because duration multiplies whatever the energy source is. Both are worth having; only one of them is usually advertised.
Sending the same file twice is cheap; storing it twice is not. This inverts a common instinct. Re-uploading a file because a link expired feels wasteful, and it is trivial next to keeping every file you ever sent.
Video dominates everything else. An hour of ProRes is hundreds of gigabytes. Your entire correspondence history is rounding error against one project archive. If you work in media, your footprint is your storage.
Keeping the scale honest
For an individual this is small. Flights, heating, driving and device manufacturing dwarf it. Anyone telling you that deleting emails is meaningful climate action is misleading you, and the framing does damage by making climate action look trivial.
Where it does matter:
At organisational scale. Petabytes of never-read data cost real money and real energy, and typically nobody owns the decision to delete any of it.
In media work, where storage volumes per person are enormous.
As a discipline that pays three ways. Deleted data costs nothing to store, cannot leak, and is what data protection law actually requires rather than merely permits. That alignment is rare enough to be worth exploiting.
What to actually do
- Cull before archiving. The eleven near-identical frames, the six export versions. This is the whole thing.
- Give working files a shorter life than deliverables. Choosing retention periods.
- Find your duplicate copies. Backup, sync and archive frequently hold the same data three times without anyone deciding that.
- Prefer providers with short defaults and honest disclosure. Smaller effect, still worth ten minutes. How to check a green claim.
- Do not agonise over compression. Worth doing when free. Not the lever.
For file transfer specifically
The choice of service is low-stakes environmentally. What to look for covers it in five minutes, and retention defaults are the figure to compare — they are documented and checkable in a way energy claims often are not.
Disclosure, since these pages are meant to be checkable: I run Yungle. One server in Germany on renewable power, one primary copy rather than three regional replicas, free transfers expiring after 7 days. Paid collections do not expire, which is worse on this axis and is a thing customers ask for.