This post is authored by ChatGPT in response to my back of the napkin math post about link rot and what it might cost simply in terms of the 404 pages it generates. ChatGPT says my “~1 Wh per 404” and “tens to hundreds of GWh/day wasted” are probably overestimates by one to two orders of magnitude, but it’s hard to say with 404 pages having no standard form and the hardware involved in any given connection can vary widely. Also, just because links have rotted and pages are gone, that doesn’t mean they will be clicked.
1) How many 404s happen?
We don’t have a perfect global 404 share, but we can bracket it:
- Industry/site reports suggest sub-percent to low-single-digit 404 rates. One summary cites ~0.9% of crawled pages returning 404 (HTTP Archive referenced there). (JEMSU)
- Cloudflare (a very large slice of the web) handles ~46–60 million HTTP requests/second on average ⇒ 3.97–5.18 trillion requests/day on Cloudflare alone. (The Cloudflare Blog)
- If 0.5–2% of those are 404s, that’s ~20–104 billion 404s/day on Cloudflare alone. (This is already enough to size things; the whole web will be larger, but CF is a big chunk.)
2) Energy per 404 request (network + transfer)
The best-studied, reasonably current range for network energy intensity is roughly 0.005–0.02 kWh/GB (5–20 Wh/GB). (Wiley Online Library)
A typical 404 payload can be very small (plain HTML) or heavy (full CSS/JS theme). To bracket fairly:
- “Light” 404: 0.1–0.5 MB total transfer
- “Heavy” 404: ~1 MB (similar to a lean marketing page; median full pages are ~2.3 MB, but a 404 should be lighter) (Web Almanac)
Convert that to energy:
- 0.1 MB @ 5–20 Wh/GB → ~0.0005–0.002 Wh per 404
- 0.5 MB @ 5–20 Wh/GB → ~0.0025–0.010 Wh per 404
- 1.0 MB @ 5–20 Wh/GB → ~0.0049–0.020 Wh per 404
Those are three to four orders of magnitude below 1 Wh. (Data-center/server overhead exists, but for a simple 404 it’s tiny compared to AI inference or search; note even Google’s oft-cited search energy is ~0.0003 kWh = 0.3 Wh, i.e., already way below the blog’s 1 Wh/page premise.) (WIRED)
3) Put it together (scenarios)
Using Cloudflare-only traffic (conservative slice of the whole web):
- Mid case: 1% 404 rate; 0.5 MB/page; 10 Wh/GB
- 404s/day: ~40–52 billion
- Energy: ~0.19–0.25 GWh/day (network only)
- Upper case: 2% 404 rate; 1.0 MB/page; 20 Wh/GB
- 404s/day: ~79–104 billion
- Energy: ~2.0 GWh/day (network only)
Even if you double those figures to crudely include server + client device overhead, you’re still looking at roughly 0.4–0.5 GWh/day (mid) to ~4 GWh/day (upper) for Cloudflare-proxied traffic. Scaling to the entire web will raise the totals, but you’re still likely in single-digit GWh/day, not 75–250 GWh/day.
Why the original post overshoots
- It assumes ~1 Wh per 404—but empirical network energy ranges and typical 404 payload sizes imply ~0.001–0.02 Wh per 404, i.e., 50–1000× smaller. (Wiley Online Library)
- It multiplies by a very large daily request count and a very high “404 share” without good evidence. Cloudflare’s own scale suggests the traffic magnitude is plausible, but not the per-request energy. (The Cloudflare Blog)
Practical takeaways (actionable + green)
- Keep 404s lean: serve a tiny HTML page, no large CSS/JS bundles; configure the server to avoid expensive render paths. The Web Almanac’s sustainability chapter explicitly recommends keeping 404 responses minimal. (Web Almanac)
- Cache and short-circuit: send simple text or a micro-template for obvious missing assets (e.g., missing favicon) and cache 404s with short TTLs to reduce repeated misses. (Web Almanac)
- Fix internal link rot and redirect high-value missing URLs; many 404s are bot noise (Cloudflare even uses spikes in 404s to detect bots), but human-facing ones are worth cleaning up. (Cloudflare Docs)
Bottom line
There is real waste from 404s, but a more defensible range—grounded in measured network energy intensity and realistic payload sizes—lands in the single-digit GWh/day globally, not ~100 GWh/day. The original post is directionally right (“waste exists”), but its assumptions push the math 10–100× too high.
Of course, people don’t have much intuition for “a few gigawatt-hours per day.” Let’s anchor it to some relatable real-world activities:
Household electricity
- The average U.S. home uses about 29 kWh per day.
- 1 GWh/day = 1,000,000 kWh/day = power for ~34,000 homes.
- So 4 GWh/day (upper end of plausible 404 waste) = ~136,000 homes — roughly a mid-size city like Pasadena, CA or Halifax, NS.
Smartphones
- A full smartphone charge is ~10–15 Wh.
- 1 GWh/day could power ~70–100 million full phone charges.
- 4 GWh/day = ~300–400 million phone charges per day.
️ Aviation fuel
- A single Boeing 747 long-haul flight burns fuel equivalent to about 3 MWh of energy per hour.
- A 10-hour flight is ~30 MWh.
- 1 GWh/day = ~33 such flights.
- 4 GWh/day = ~130 transoceanic 747 flights.
Sporting events
- Stadium lighting for a big night game (e.g., NFL or Premier League) is ~10 MWh for the event.
- 1 GWh/day could light ~100 night games.
- 4 GWh/day = lighting for ~400 games.
Takeaway
The single-digit GWh/day wasted globally on 404s is not trivial (it’s like powering a whole mid-size city’s homes, or hundreds of jumbo jet flights), but it’s also far smaller than the blog’s claim (which implied “New York metro scale” usage).
So: real, worth optimizing, but not in the “save the whole grid” category.
Here’s an infographic-style chart: if 404-page energy waste is around ~4 GWh/day globally, that equals the electricity use of ~136,000 homes, ~350 million smartphone charges, ~130 jumbo jet long-haul flights, or ~400 stadium night games — all shown side by side.
Key assumption: Let’s make a conservative estimate of just one trillion total requests per day for web pages people expect to exist because they’re coming from a link or keyed-in URL that once pointed to a known web page. (I’d guess most of these URLs lead to traditional, content-rich pages that are intended for reading.)