Changing what humanity eats avoids far more greenhouse
gas than shutting down all artificial intelligence — on emissions, by a margin of roughly 130 to 1. On
water, the world's golf courses out-consume all of artificial intelligence by five to thirteen times. Every figure
below is drawn from peer-reviewed research and listed in full.
The Index
Greenhouse gases avoided each year, in billion metric tons of CO₂e, if…
… all humanity ate a vegan diet6.6[1]
… all humanity ate a vegetarian diet4.5[2]
… all artificial intelligence were permanently shut down0.05[3]
CO₂e avoided over fifty years (1976–2026), in billion metric tons, by…
… half a century of global veganism~330[4]
… half a century of global vegetarianism~225[5]
… half a century with all AI banned<0.5[6]
CO₂ drawn back out of the air by rewilding freed land, in billion metric tons, if we…
… ended animal agriculture and let pasture revert to wild330–550[7]
… dismantled all computing and rewilded the data centers≈ 0[8]
Blue water consumed over fifty years, in billion cubic meters, by…
… the world's golf courses~200[9]
… the entire existence of artificial intelligence15–40[10]
Annual CO₂ emissions avoided — if all humans changed
All humans were Vegan
6.6 Gt
All humans were Vegetarian
~4.5 Gt
All AI shut down
0.05 Gt
Shutting down all artificial intelligence avoids about 1/130th of the carbon that a global
vegan diet would.
CO₂ emissions avoided over fifty years (1976–2026)
Fifty years Vegan
~330 Gt
Fifty years Vegetarian
~225 Gt
Fifty years no AI
<0.5 Gt
Banning AI for half a century avoids less than two weeks of present-day global food
emissions.
CO₂ drawn down by rewilding the freed land
End animal agriculture
330–550 Gt
End all computing
≈ 0
Livestock occupies roughly 3.1 billion hectares; data centers occupy a rounding error.
Dismantling computing frees essentially no land for regrowth.
Blue water consumed over fifty years
World's golf courses
~200 bn m³
All artificial intelligence
~15–40 bn m³
Measured the same way on both sides — consumptive blue water only. Golf consumes five
to thirteen times more than all of artificial intelligence.
AI's carbon problem is when, not
how much: per-query emissions keep falling even as the total keeps
climbing — the buildout outruns every efficiency gain.
Google's absolute
emissions rose 51% over 2019 while its energy per prompt fell 33-fold. Today AI is a small wedge of global carbon,
but the fastest-growing industrial electricity load on Earth.
— Google Environmental Report 2025.
AI's water problem is where, not
how much: about two-thirds of new US data centers sit in water-stressed
regions.
An aquifer is local;
the aggregate total is not. A modest global footprint can still drain a single watershed.
— Bloomberg 2025 (WRI / DC Byte data).
Sources
[1] Poore, J., & Nemecek, T. (2018). Reducing food's environmental impacts
through producers and consumers. Science, 360(6392), 987–992.
doi:10.1126/science.aaq0216.
A plant-based shift cuts food's greenhouse gases by about 49%; food ≈ 13.6 Gt CO₂e/yr, so 13.6 ×
0.49 ≈ 6.6 Gt/yr.
[2] Scarborough, P., et al. (2023). Vegans, vegetarians, fish-eaters and
meat-eaters in the UK show discrepant environmental impacts. Nature Food, 4(7), 565–574.
doi:10.1038/s43016-023-00795-w. Retained dairy and eggs leave vegetarians at roughly
65–75% of the vegan saving.
[3] de Vries-Gao, A. (2025). Patterns (Cell Press).
Full
text. The first peer-reviewed isolation of AI's specific share: 32.6–79.7 Mt CO₂ in 2025
(0.033–0.080 Gt). For context, all data centers ≈ 182 Mt
(IEA, Energy and AI,
2025).
[4] 6.6 Gt/yr × 50 ≈ 330 Gt. An illustrative ceiling: historical food
emissions were lower than today's.
[5] 4.5 Gt/yr × 50 ≈ 225 Gt, with the same ceiling caveat as [4].
[6] AI's footprint was negligible before about 2012; integrating the actual curve
gives a cumulative total under 0.5 Gt — less than two weeks of present-day global food emissions.
[7] Hayek, M. N., et al. (2020). The carbon opportunity cost of animal-sourced
food production on land. Nature Sustainability, 3(11), 939–942.
Full text.
Releasing roughly 3.1 billion hectares of grazing and feed cropland could sequester 332–547 GtCO₂ as
vegetation matures over decades.
[8] The global land footprint of data centers is negligible; dismantling them
frees essentially no land for ecological regrowth.
[9]USGA water survey; Shaddox, T. W., et al. (2022). Crop, Forage &
Turfgrass Management, 8(1), e20145.
doi:10.1002/cft2.20145. US golf
≈ 2.07 billion m³/yr over ~16,000 courses, scaled to ~38,000 courses worldwide ≈ ~4 billion
m³/yr, or ~200 billion m³ over fifty years (range 175–250).
[10] de Vries-Gao, A. (2025). Patterns — AI's water footprint reaches
312.5–764.6 billion liters (0.31–0.76 billion m³) in 2025; × 50 ≈ 15–40 billion
m³. By a broader withdrawal boundary (most water returned), 4.2–6.6 billion m³/yr by 2027
(Li, Yang, Islam & Ren, 2025, CACM).
All works and media on Glia.ca by David Jhave
Johnston
are licensed under CC
BY-NC-SA 4.0.
Jhave is currently an AI-narrative researcher at the UiB Centre for Digital
Narrative (2023–27) with the Extending Digital Narrative project.
This work was partially supported by the Research Council of Norway through its Centres of Excellence scheme,
project number 332643 (Center for Digital Narrative), and its SAMKUL project scheme, project number 335129
(Extending Digital Narrative).
