Is Dark Mode Actually Saving the Planet?
The truth about digital bloat, phantom data centers, and why your cloud storage is a building full of servers overheating in the desert
There is a small, satisfying ritual that a lot of us perform without thinking much about it. You open your phone settings, toggle on dark mode, and for a brief, quiet moment, you feel like you have actually done something. Like you have opted out of something bad and into something slightly better. It costs nothing, by the way. It looks sleek. And somewhere in the back of your mind, you have absorbed the idea that dark pixels use less energy, so the planet is marginally better off because of your preference.
I am not here to take that away from you entirely, but rather to tell you that the story is significantly more complicated, and that the part of your digital life that is actually destroying the environment has nothing to do with your screen brightness.
Your “cloud” is actually a building
Let’s start with the thing almost no one pictures when they hear the world “cloud.” The cloud is not a diffuse, weightless phenomenon. It is not floating somewhere above your city, passive and efficient. The cloud is a warehouse. Often many warehouses, stacked with servers, running at a full capacity around the clock, generating enormous amounts of heat, and requiring industrial-scale cooling systems to keep from melting. These facilities are called data centers, and they are among the most energy-intensive structures humans have ever built.
There are data centers in the desert outside of Phoenix. There are data centers in the hills of Virginia. There are data centers in Ireland, Singapore, and rural North Carolina, and other places, chosen partly for their cheaper land and power costs, and partly because the local governments were willing to offer tax incentives in exchange for the promise of infrastructure investment. You’d think they are glamorous, but in actuality, they don’t even show up in the marketing materials for the apps they power. They just run, continuously, consuming electricity at scale that rivals mid-sized cities.
According to the International Energy Agency’s 2025 Energy and AI report, data centers globally consumed roughly 200 to 250 terawatt-hours of electricity in recent years. This figure has continued to climb as synthetic media app workloads, video streaming, and cloud storage demand accelerate. To put that in terms that are slightly more legible, that is roughly the annual electricity consumption of some entire countries and it is not even close to a rounding error. It is a structural feature of the digital economy that the digital economy prefers not to discuss.
So when you tap “back up photos to iCloud” without a second thought, you are not doing something weightless. You are writing to a physical server in a physical building that is consuming physical energy. The photos from 2019 that we occasionally look at from time to time is sitting in a rack somewhere, costing a fraction of a cent to store, multiplied by billions of similar decisions made by billions of similar people, adding up to something that matters.
The dark mode situation, spoken honestly
Back to the toggle.
Dark mode does save battery life but only on OLED and AMOLED screens, where dark pixels are achieved by turning off individual LEDs rather than filtering a backlight. On an older LCD screen, dark mode saves essentially nothing because the backlight stays on regardless. Even on OLED, the savings are highly dependent on brightness. Purdue University researchers who built one of the first accurate OLED power profiling tools found that at typical indoor brightness levels of 30-50%, switching to dark mode saves only 3-9% of screen power on average. At maximum outdoor brightness, those savings climb to 39-47% percent depending on screen brightness and usage, which is a range wide enough to tell you that context matters enormously. That is a narrow scenario: screen power, or what the girls like to call “screen time,” is a small slice of your device’s total energy consumption to begin with.
The more important point is this: screen power consumption is a tiny fraction of the energy story of your device, which is itself a tiny fraction of the energy story of the services your device connects to. You could run your phone screen at maximum brightness 24 hours a day and it would still be negligible compared to the server infrastructure required to deliver the feed on your favorite video-doomscrolling app.
Dark mode is not nothing. But it is also not the variable that matters most. It is a local optimization on a system that has much larger ineffective upstream.
The actual problem: software as physical infrastructure
Here is the part that does not get nearly enough attention, and the part I find genuinely fascinating as someone who works in engineering.
Software has weight.
No, not literal weight, obviously. But software that is written inefficiently, that makes more computational requests than it needs to, that stores more data than it needs to, that runs more processes simultaneously than the task requires, consumes more energy to execute. And at scale, that difference is physically measurable in the output of power plants.
This is the discipline of green software engineering, and it is an emerging field that sits at the intersection of performance optimization and environmental accountability. The core argument is simple: a leaner, faster application is almost always a greener one, because efficiency in computation maps directly to efficiency in energy use.
Consider what happens when a developer writes a database query that retrieves ten thousand records when the application only needs fifty. That query runs on a server somewhere. The server uses electricity to execute it. The electricity was generated somewhere, possibly from a gas-fired plant, possibly from a coal plant. The developer will never see that connection. Instead, they are looking at a screen in an office, thinking about load times. But the connection still exists.
Green software engineering asks developers to hold that connection in mind. It involves practices like: reducing the number of API calls an application makes, optimizing algorithms so they complete tasks with fewer computational steps, designing systems that scale down during low-traffic periods rather than running at full capacity by default, and choosing data storage architectures that mining redundancy without sacrificing reliability.
The inverse of this looks like software designed not for efficiency but instead designed for maximum engagement, and that has a measurable physical cost. Short-form video platforms are the most intrusive case. According to a 2024 analysis by Greenly, a carbon accounting firm, video-heavy platforms consumes significantly more energy per minute of use than text or image-based ones. The gap is a direct consequence of the platform’s architecture: continuous video streaming, autoplay that never pauses to ask if you want to continue, and an algorithm engineered to maximize time spent. The cumulative result across the category is an emissions footprint that rivals the annual output of mid-sized countries.
The platforms most responsible for this design pattern have also been the least transparent about its cost. The Greenhouse Gas Protocol, the dominant emissions reporting framework, does not require social media platforms to disclose their full environmental impact. That means the companies that most aggressively redesigned how a generation consumes content have largely have been able to opt out of accounting for what that redesign costs the atmosphere. Carbon neutrality pledges exist. Meaningful, verifiable progress toward them is harder to find.
None of this is glamourous. It is simply the kind of engineering work that rarely gets celebrated because it produces invisible results, like the absence of wasted energy rather than the presence of a shiny new feature. But the Green Software Foundation, an industry consortium whose Software Carbon Intensity specification became an ISO standard in 2024, has documented real-world cases where efficiency improvements produced measurable emissions reductions. One implementation across 60 applications achieved an average 15 percent reduction in CO2 output. The gains do not require new hardware, new infrastructure, or new supply chains. They require better decisions at the code level.
Software efficiency improvements could represent one of the most scalable levers available for reducing the tech sector’s environmental footprint.
Digital hoarding is an environmental behavior
I also want to stay practical for a moment, because I think it matters that this does not feel “abstract.”
Most of us have a digital accumulation problem, mainly because the systems we use are designed to make storage feel free and infinite. Gmail has trained us to archive everything. Google Photos offers unlimited backups. Slack retains message history indefinitely on paid plans. Streaming services keep our watchlist and viewing histories going back years. The default is always to keep, to store, and to retain, and honestly that is because storage is cheap, and the cost of that cheapness is externalized somewhere we cannot see.
The average person’s email inbox contains thousands of unread newsletters, promotional messages, and automated notifications that are never opened and never deleted. Those emails exist on servers. Those servers use electricity. This is a small thing, but small things are what scale looks like from the inside.
Some concrete things that are genuinely not nothing:
Unsubscribing from emails lists you do not read reduces the number of automated messages being generated, stored, and delivered to your account. Deleting photos and videos you will never use (especially cloud-baked ones) reduces active storage. Streaming video at resolutions appropriate to your screen size (you do not need 4K on a laptop) reduces the amount of data required from content delivery networks. Losing browser tabs you are not using reduces the active compute load on your machine and, marginally, on the servers those tabs are pinging.
None of this will solve the problem, yes. The structural solutions are more upstream. The structural solutions live in policy, in how data centers are powered, in what standards we hold cloud providers to, in whether green software engineering becomes a professional expectation rather than a personal preference of conscientious developers. BUT the individual behaviors are worth naming because they correct the misperception that the digital world exists outside the physical one.
The design problem
I want to close with something that I think gets lost in a lot of sustainability discourse.
The reason most people do not think about the environmental cost of their digital behavior is not ignorance or apathy. It is simply the design. The interfaces we use are deliberately built to make accumulation feel effortless and disposal feel laborious. Signing up for a newsletter takes one click. Unsubscribing requires finding the link, waiting for the confirmation page, sometimes confirming a second time. Uploading a photo to a cloud storage happens automatically in the background. Deleting it requires you to find it, delete it, then go find the “recently deleted” folder and delete it again. This is intentional. Systems architects and solution engineers design this asymmetry because it serves the direct-consumer-business interests of platforms that monetize data storage and engagement. And it means that the idea that individual digital hygiene (“personal responsibility“ framing) is the solution, is doing a lot of work to cover for a structural design.
The more honest conversation is about what we should expect from the companies building these systems. We should expect them to default to energy-efficient infrastructure. We should expect them to default to energy-efficient infrastructure. We should expect them to pursue renewable energy procurement aggressively and transparently. We should expect software engineers to be trained in efficiency as a professional standard, not as an optional extra. And we should expect the companies that market themselves on their environmental commitments to be held accountable for the energy consumption of the products they ship.
Dark mode is fine. It is a reasonable preference. But the planet’s future is not going to be decided by your screen settings. Actually, it’s going to be decided by whether the systems powering your screen are built by people who treat energy as a real cost with real consequences, rather than an abstraction to be optimized away in the next fiscal quarter.
That’s more of a design question. And just know, the design questions have the real answers.
Sources & References
International Energy Agency — Energy and AI (April 2025). Global data center electricity consumption and growth projections.
Purdue University — Y. Charlie Hu & Pranab Dash, OLED Power Modeling Study (2021). Dark mode energy savings research; also published in ACM MobiSys conference proceedings.
Pew Research Center — What We Know About Energy Use at U.S. Data Centers Amid the AI Bloom (October 2025). Virginia data center concentration and household cost impact figures.
Fortune — Doomscrolling Has Reportedly Made TikTok’s Annual Carbon Footprint Almost the Same as the Entire Country of Greece (December 2024). Reporting on Greenly’s The Hidden Environmental Cost of Social Media analysis.
 | A guest post by
|