THE ENERGY COST OF A WEBSITE

Every website runs on a physical machine somewhere. That machine draws power 24 hours a day. It sits in a building that needs cooling, networking, lighting, and security. That building was filled with hardware that was mined, refined, assembled, and shipped from the other side of the world.

This page puts numbers on all of that. Not to argue that rehosted replaces conventional hosting — it obviously doesn't. But to make visible the energy cost that most people never think about, and to show what happens when you go the other way: less power, reused hardware, static files, solar panels.

THE BIG PICTURE: DATA CENTRES IN 2024

Global data centres consumed an estimated 415 TWh of electricity in 2024 — roughly 1.5% of all global electricity. That's more than the entire country of South Africa uses in a year.1

The United States alone accounts for 180 TWh of that (44%), with China at 102 TWh (25%).2 The IEA projects global data centre consumption will double to 945 TWh by 2030, driven largely by AI workloads.3

To put 415 TWh in perspective: if global data centres were a country, they'd be the 14th largest electricity consumer on Earth, between France and Brazil.

  [GLOBAL DATA CENTRE POWER — 2024]

  Total:         415 TWh/year
  Share of grid: ~1.5% of global electricity
  Growth rate:   12% per year (last 5 years)
  2030 forecast: 945 TWh (doubling)
  AI share:      5-15% now, projected 35-50% by 2030
      
1 IEA, "Energy and AI" (2025) — iea.org/reports/energy-and-ai
2 Pew Research (2025) — pewresearch.org
3 IEA, "Electricity 2025" — iea.org/reports/electricity-2025
WHAT DOES ONE WEBSITE ACTUALLY COST?

Let's trace the full energy cost of hosting a single modern website on conventional infrastructure. We'll use a typical shared hosting scenario: one site on a standard 1U rack server in a commercial data centre.

The server

A typical 1U rack-mounted server draws 300–350W continuously.4 High-performance models can exceed 2,000W. Over a year at 325W average, that's 2,847 kWh — just for the server itself.

The overhead (PUE)

Servers don't run in isolation. They need cooling, power distribution, UPS systems, networking, lighting, and security. The industry measures this with PUE (Power Usage Effectiveness) — a multiplier on the raw IT power draw.

OperatorPUE (2024)Meaning
Google (best)1.077% overhead
Google (fleet avg)1.099% overhead
AWS (global avg)1.1515% overhead
Microsoft (global avg)1.1616% overhead
Industry average1.5656% overhead
4 RackSolutions, "Server Rack Power Consumption" — racksolutions.com
PUE data: Google Data Centers — datacenters.google; The New Stack — thenewstack.io; Uptime Institute 2024 survey via Statista — statista.com

Google's hyper-efficient Oregon facility is the exception. The industry average PUE of 1.56 means that for every watt your server uses, another 0.56W goes to keeping the building running. That pushes our 325W server to an effective 507W at the wall, or 4,441 kWh/year.

Your share of it

On shared hosting, a single website typically gets 1 CPU core and 256–512MB RAM out of a 16-core, 64GB server.5 Hosting providers routinely oversell by 3–4x because 95% of sites use a fraction of their allocation. But the server runs at full power regardless. Your "share" of the energy is roughly 1/16th to 1/32nd of the total — but the server doesn't use less electricity because your site is idle.

For a dedicated or VPS setup (increasingly common), you're paying for — and powering — the whole server or a guaranteed slice of it.

5 WebHostMost, "Hosting Resource Allocation" — webhostmost.com

Where the power goes

  [DATA CENTRE POWER BREAKDOWN]

  Servers / IT equipment  ████████████████░░░░  ~40%
  Cooling systems         ██████████░░░░░░░░░░  ~30%  (7% hyperscale, 30%+ enterprise)
  Power distribution      ████████░░░░░░░░░░░░  ~20%
  Networking              ██░░░░░░░░░░░░░░░░░░   ~5%
  Storage                 ██░░░░░░░░░░░░░░░░░░   ~5%
      
Stanford, "Data Center Energy Use" — stanford.edu; Socomec — socomec.us
THE HIDDEN COST: MANUFACTURING THE HARDWARE

Before a server ever powers on, it has already produced a significant carbon footprint. Mining the metals, refining the silicon, manufacturing the chips, assembling the boards, and shipping the finished product — all of this has a measurable CO2 cost called embodied carbon.

Server ModelEmbodied CO2Lifetime CO2Embodied %
HPE ProLiant DL380 Gen101,726 kg5,657 kg30.5%
Dell PowerEdge R740 (US)1,313 kg8,640 kg15.2%
Dell PowerEdge R740 (EU)1,313 kg~2,600 kg~50%
Tech Carbon Standard, "Server Lifecycle Example" — techcarbonstandard.org
Dell Technologies, "R740 Full LCA" — delltechnologies.com

Manufacturing a single rack server produces roughly 1,200–1,750 kg of CO2.6 In regions with cleaner electricity grids (like the EU), embodied carbon can account for up to half of the server's total lifetime emissions — because operational power is less carbon-intensive.

77% of the electronics industry's greenhouse gas emissions come from the supply chain: mining, manufacturing, assembly, and transport — not from using the devices.7

This is where reuse matters most. Every retro machine we put back into service is a server that didn't need to be manufactured. A Nintendo Wii that would otherwise sit in a drawer or end up in landfill has zero additional embodied carbon when repurposed as a web server.

The numbers back this up. Extending a server's life by just one extra year reduces its amortised carbon footprint by roughly 16%.8 Microsoft's Circular Centers aim for 90% server component reuse. Google has resold 44 million hardware components since 2015, including 7 million in 2023 alone.9 The industry is starting to realise that the greenest server is one that already exists.

6 ServerMonkey, "Environmental Benefits of Refurbished Servers" — servermonkey.com
7 Cloud Carbon Footprint, "Embodied Emissions" — cloudcarbonfootprint.org
8 Green Software Foundation — learn.greensoftware.foundation
9 IEEE Spectrum, "Server Reuse" — spectrum.ieee.org
STATIC HTML vs DYNAMIC SITES

Most of the modern web runs on dynamic platforms — WordPress alone powers over 40% of all websites. Every page view triggers PHP execution, database queries, template rendering, and HTML generation. A static HTML site skips all of that: the server reads a file from disk and sends it. Done.

The performance difference is dramatic:

  [THROUGHPUT BENCHMARK]

  Static HTML:   2,218 requests/second
  WordPress:         6 requests/second

  That's a 360x difference.
      
Gatsby, "Static Sites Fight Climate Change" — gatsbyjs.com

This matters for energy because CPU time is the primary driver of power consumption. A server handling static files barely touches the CPU — it's essentially a glorified file transfer. A WordPress site, by contrast, executes code and queries a database for every single request.

The Website Carbon Calculator estimates the average web page produces 0.5g of CO2 per page view.10 That sounds tiny, but multiply it by the billions of page views happening every day and you get a measurable slice of global emissions.

A static site served from low-power hardware, with no database, no server-side rendering, and no framework overhead, produces a fraction of that. The pages you're reading right now are plain HTML files served by nginx — no PHP, no Node.js, no database.

10 Website Carbon Calculator — websitecarbon.com; Wholegrain Digital — wholegraindigital.com
POWER CREEP: THE WEB KEEPS GETTING HEAVIER

The average web page has grown relentlessly. According to HTTP Archive, the median desktop page weighed about 500 KB in 2010. In 2024, it's 2,652 KB — a 5.3x increase.11 Mobile pages have grown even faster: 11.5x over the same period.

YearMedian DesktopMedian Mobile
2010~500 KB~200 KB
2014~1,200 KB~500 KB
2018~1,700 KB~1,500 KB
2022~2,312 KB~2,037 KB
20242,652 KB2,311 KB
11 HTTP Archive, "Page Weight Report" — httparchive.org; Web Almanac 2024 — almanac.httparchive.org

Where does the weight come from? The median 2024 desktop page loads 71 requests — 24 JavaScript files (613 KB), 18 images (1,054 KB), 8 CSS files, and 4 web fonts. The actual HTML content? Just 18 KB.

JavaScript bloat

The CPU cost of modern web pages is dominated by JavaScript. Research by Tim Kadlec found that sites built with React require 248% more CPU time on desktop and 658% more on mobile compared to sites without frameworks.12

As of 2024, some popular sites ship staggering amounts of JavaScript:13

  [JAVASCRIPT PAYLOAD — 2024]

  Slack         55 MB   ████████████████████████████
  Jira          50 MB   █████████████████████████
  LinkedIn      31 MB   ████████████████
  Discord       21 MB   ███████████
  Gmail         20 MB   ██████████
  Notion        16 MB   ████████
  Facebook      12 MB   ██████
  Google Search  9 MB   █████
  Wikipedia    0.2 MB   ▏
      
12 Tim Kadlec, "The Cost of JavaScript Frameworks" — timkadlec.com
13 Tonsky, "JavaScript Bloat in 2024" — tonsky.me

Every byte of JavaScript must be downloaded, parsed, compiled, and executed. On a phone, this takes real time and real energy. A 55 MB JavaScript bundle doesn't just slow down the user — it consumes measurably more electricity on every device that loads it.

Meanwhile, only 70% of pages bother with text compression. Only 57% minify their JavaScript.14 The web has gotten heavier not because it needed to, but because hardware got fast enough that developers stopped noticing.

14 Web Almanac 2024, "Page Weight" — almanac.httparchive.org
THE REHOSTED COMPARISON

Now let's put our numbers side by side. What does it actually look like to host a website on rehosted versus conventional infrastructure?

Power draw

HardwarePower (W)kWh/yearvs 1U Server
1U rack server3252,847
1U + PUE (1.56)5074,441
Nintendo Wii1815828x less
Amiga 120076172x less
Raspberry Pi 45.14599x less
Wii: TPCDB — tpcdb.com; Laura Cowen — lauracowen.co.uk
Pi 4: Pi Dramble benchmarks — pidramble.com; RasPi.TV — raspi.tv
Amiga: English Amiga Board — eab.abime.net

A Raspberry Pi serving web pages draws 5.1W.15 A Nintendo Wii draws about 18W. A 1U rack server in a data centre, after accounting for cooling and power overhead, draws around 507W.

That's not a small difference. The Pi uses 99x less power. The Wii uses 28x less. Even the Wii — a games console from 2006 — is dramatically more efficient than a modern rack server at serving static HTML files, because static file serving requires almost no CPU. The limiting factor is network I/O, not compute.

15 Pi Dramble, "Power Consumption Benchmarks" — pidramble.com

Embodied carbon

HardwareManufacturing CO2Status
New 1U rack server1,200–1,750 kgNew build
Nintendo Wii (reused)0 kg (additional)Already manufactured
Amiga (reused)0 kg (additional)Already manufactured
Raspberry Pi 4~10–20 kg (est.)New, but minimal

The retro hardware has already been manufactured. Its embodied carbon was emitted years or decades ago. By putting it back to work instead of buying new servers, the additional manufacturing emissions are zero. The only new hardware in the rehosted stack is the Raspberry Pi controller, which is a fraction of the size, materials, and manufacturing energy of a rack server.

Solar viability

A single 100W solar panel in the UK produces roughly 300–500 Wh per day, depending on season and location.16

HardwareDaily draw100W panel output (UK)Panels needed
Raspberry Pi 4122 Wh400 Wh0.3 (one panel = 3x surplus)
Nintendo Wii432 Wh400 Wh1–2 panels
Amiga 1200168 Wh400 Wh0.4 (one panel = 2.4x surplus)
1U rack server + PUE12,168 Wh400 Wh~30 panels
16 Renogy UK — uk.renogy.com; EnergySage — energysage.com; EcoFlow — ecoflow.com

A single small solar panel can power the entire rehosted platform — controller, nodes, and all. A conventional rack server would need a rooftop array just for one machine, before you even account for cooling.

WORKED EXAMPLE: ONE YEAR OF HOSTING

Let's compare the total annual carbon footprint of hosting a simple website — a few static pages, modest traffic — on conventional infrastructure versus rehosted.

  [CONVENTIONAL HOSTING — 1 year]

  Hardware:
    1U server, shared (your 1/16th share)
    Power draw: 325W (server) × 1.56 (PUE) = 507W total
    Your share: 507W ÷ 16 = 31.7W
    Annual energy: 31.7W × 8,760h = 278 kWh

  Manufacturing (amortised):
    1,500 kg CO₂ embodied ÷ 4-year lifespan ÷ 16 tenants
    = 23.4 kg CO₂/year (your share)

  Operational emissions:
    278 kWh × 0.233 kg CO₂/kWh (UK grid avg 2024)
    = 64.8 kg CO₂/year

  Total: ~88 kg CO₂/year
  ─────────────────────────────────────────────

  [REHOSTED — 1 year]

  Hardware:
    Raspberry Pi 4 controller: 5.1W
    Nintendo Wii node: 18W
    Total: 23.1W
    Annual energy: 23.1W × 8,760h = 202 kWh

  Manufacturing:
    Both devices already exist. Additional embodied = 0 kg CO₂

  Operational emissions (solar):
    202 kWh × 0 kg CO₂/kWh (solar, no grid)
    = 0 kg CO₂/year

  Total: ~0 kg CO₂/year
      
Even without solar, running on the UK grid, the rehosted setup would produce 47 kg CO2/year (202 kWh × 0.233) — roughly half the conventional figure. The solar panels eliminate even that.

The conventional figure of 88 kg CO2/year is actually quite conservative. It assumes shared hosting (you split costs with 15 others), a relatively efficient data centre, and UK grid carbon intensity — which is lower than the global average (0.49 kg/kWh). On a dedicated server in a country with a coal-heavy grid, the number could be 5–10x higher.

WHAT THIS DOESN'T MEAN

This comparison is not an argument that rehosted can replace conventional hosting. It obviously can't. A Nintendo Wii cannot run a database server. A Raspberry Pi cannot handle thousands of concurrent connections. Dynamic web applications, APIs, real-time services — these need real servers in real data centres.

But the vast majority of the web doesn't need any of that. Most websites are informational — they display text, images, and links. They could be static HTML. They could run on hardware that draws 5 watts. They could be powered by a solar panel the size of a laptop screen.

The question isn't "can we replace data centres?" It's: does every website need one?

Rehosted is a proof of concept. A demonstration that sustainable hosting is physically possible with hardware that already exists, power that falls from the sky, and web pages that are just files on a disk.

ALL SOURCES

Data centre energy:

IEA, "Energy and AI" (2025) — iea.org

IEA, "Electricity 2025" — iea.org

Pew Research (2025) — pewresearch.org

S&P Global (2025) — spglobal.com

Server power & PUE:

RackSolutions — racksolutions.com

Google Data Centers — datacenters.google

The New Stack — thenewstack.io

Stanford — stanford.edu

Embodied carbon:

Tech Carbon Standard — techcarbonstandard.org

Dell R740 LCA — delltechnologies.com

Cloud Carbon Footprint — cloudcarbonfootprint.org

Green Software Foundation — greensoftware.foundation

IEEE Spectrum — spectrum.ieee.org

ServerMonkey — servermonkey.com

Static vs dynamic:

Gatsby — gatsbyjs.com

Website Carbon Calculator — websitecarbon.com

Wholegrain Digital — wholegraindigital.com

Page weight & JavaScript:

HTTP Archive — httparchive.org

Web Almanac 2024 — almanac.httparchive.org

Tim Kadlec — timkadlec.com

Tonsky — tonsky.me

Hardware power draw:

TPCDB (Wii) — tpcdb.com

Laura Cowen (Wii) — lauracowen.co.uk

Pi Dramble (Pi 4) — pidramble.com

RasPi.TV (Pi 4) — raspi.tv

English Amiga Board (Amiga) — eab.abime.net

Solar output:

Renogy UK — uk.renogy.com

EnergySage — energysage.com

EcoFlow — ecoflow.com