Science |
August 5, 2026 |
Will AI success spoil Rock Habitat?
Oh, how we love that big data!
August 5, 2026
Interesting news spotted on a Facebook feed. It appears China is experimenting with placing large data centers on the seabed. The subject of environmental effects immediately came up.
I thought I'd ask Claude for a quantitative analysis that's beyond my simple skills and even simpler data set. My prompt and its response follow.
The general theme is that in and of itself this approach would not make much of a dent in ocean temperatures, but a followup question brought forth several legitimate concerns regarding impacts on marine life stemming from secondary effects. (Yes, systems folk, I know that "secondary" is a bad term, that all effects should be considered equally, and shame on me for not being more euphemistic.)
It might be interesting if others were to submit the same prompt to other LLMs and especially interesting to compare responses from US and Chinese systems, looking for vocabulary indicative of political influence.
Prompt to Claude: Act as an oceanographer. Consider the heat signature of the world's data centers and multiply that by ten. Considering the volume of Earth's oceans, what is the probable range in temperature change that might ensue if all of the world's current data centers X ten were submerged onto continental shelves. Ignore the impact of heat emanating from cabling to shore. Confirm that the estimate is for ocean temperature only and not for global general temperature.
Claude's reply:
Assumptions: Global data centers draw ~500 TWh/yr (IEA 2025 estimate); virtually all of that electricity ends up as waste heat. Ten times that is 5,000 TWh/yr = 1.8×10¹⁹ J/yr. Ocean volume ≈ 1.335×10⁹ km³, mass ≈ 1.37×10²¹ kg, specific heat ≈ 3,850–4,000 J/kg·K.
If that heat instantly mixed through the entire ocean, the temperature rise is roughly 3×10⁻⁶ °C per year — about 0.0003°C over a century. Undetectable against natural noise.
If instead the heat stayed trapped in shelf waters (not the whole ocean — the scenario you're describing) using ~27 million km² of continental shelf at ~130 m average depth (mass ≈ 3.6×10¹⁸ kg), the rise is roughly 0.001°C per year, assuming zero heat loss to the atmosphere or exchange with the open ocean. In reality, continental shelves are among the most vigorously mixed and ventilated parts of the ocean (tidal currents, wind mixing, exchange with open water), so heat wouldn't actually accumulate there — it would disperse on timescales of days to weeks, keeping any local anomaly far below that ceiling except very close to the source.
Context check: that 1.8×10¹⁹ J/yr is about 0.15% of the ~1.2×10²² J (12 ZJ) the oceans already absorb annually from anthropogenic climate forcing. So even the worst-case, no-mixing shelf estimate is a rounding error next to existing ocean warming trends.
Bottom line: probable range is roughly 10⁻⁶ °C/yr (well-mixed, whole-ocean case) to ~10⁻³ °C/yr (worst-case, heat trapped locally on shelves with no dissipation) — several orders of magnitude below anything oceanographically measurable or ecologically significant, and dwarfed by existing ocean heat uptake from greenhouse forcing.
Every calculation above is ocean-only. The temperature changes (10⁻⁶ °C/yr whole-ocean, up to ~10⁻³ °C/yr worst-case shelf-trapped) are the result of dividing the heat energy by the mass and specific heat of seawater alone. I did not fold in atmosphere, land, or ice mass, and I didn't compute a global mean surface temperature (the metric used for "global warming" figures, which blends ocean, land, and air and uses a much smaller effective heat capacity per unit area near the surface). So this isn't a global-temperature estimate — it's specifically how much the ocean's own heat content would rise if it absorbed 10x the world's data center waste heat with no losses to atmosphere or space.
What concerns might one have about the effect of that change on marine life?
The basin-wide average I calculated is almost meaningless for marine life — organisms don't experience an ocean-wide mean, they experience whatever's happening in the water column right around them. The real concern is the near-field thermal plume at each facility, which could be many degrees above ambient even if the global or shelf-wide average shift is imperceptible. This is the same physical situation as cooling-water discharge from coastal power plants, which is well studied and gives a useful reference point: outfall temperatures are often 5–15°C above ambient at the point of discharge, dropping off with distance and mixing.
A few specific concerns follow from that. Continental shelves host nursery grounds, benthic communities, and fisheries that are disproportionately important relative to their area, so siting heat sources there concentrates risk in ecologically valuable habitat rather than the open ocean. Sessile and low-mobility organisms — corals, shellfish beds, seagrass, benthic invertebrates — can't relocate away from a chronic local plume the way fish can, so they bear continuous thermal stress rather than a brief pulse. Warm water holds less dissolved oxygen, so a persistent plume can create localized hypoxic stress on top of the temperature effect itself, compounding metabolic strain. Chronic, always-on point-source heat (unlike a seasonal heatwave) also removes the recovery windows organisms rely on to reset after thermal stress.
There's also a documented behavioral effect worth flagging: thermal discharges can act as artificial refugia, attracting species that wouldn't normally persist there (tropical or subtropical fish overwintering near warm outfalls is a well-known pattern) or enabling non-native/invasive species to gain a foothold. And any additional warming stacks on top of an already-warming ocean — species and reefs near their thermal tolerance limits from climate change have less buffer before local heat stress tips into bleaching or die-off events.
Finally, heat wouldn't be the only stressor from physically siting hardware on the shelf: biofouling-control chemicals or biocides in cooling systems, construction turbidity, physical habitat loss under the footprint, and low-frequency noise are all plausible co-stressors that would compound with the thermal plume rather than act independently.
Graphics omitted.
Sources:
- [Executive summary – Key Questions on Energy and AI – IEA](https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary)
- [AI: Five charts that put data-centre energy use into context - Carbon Brief](https://www.carbonbrief.org/ai-five-charts-that-put-data-centre-energy-use-and-emissions-into-context)
I thought I'd ask Claude for a quantitative analysis that's beyond my simple skills and even simpler data set. My prompt and its response follow.
The general theme is that in and of itself this approach would not make much of a dent in ocean temperatures, but a followup question brought forth several legitimate concerns regarding impacts on marine life stemming from secondary effects. (Yes, systems folk, I know that "secondary" is a bad term, that all effects should be considered equally, and shame on me for not being more euphemistic.)
It might be interesting if others were to submit the same prompt to other LLMs and especially interesting to compare responses from US and Chinese systems, looking for vocabulary indicative of political influence.
Prompt to Claude: Act as an oceanographer. Consider the heat signature of the world's data centers and multiply that by ten. Considering the volume of Earth's oceans, what is the probable range in temperature change that might ensue if all of the world's current data centers X ten were submerged onto continental shelves. Ignore the impact of heat emanating from cabling to shore. Confirm that the estimate is for ocean temperature only and not for global general temperature.
Claude's reply:
Assumptions: Global data centers draw ~500 TWh/yr (IEA 2025 estimate); virtually all of that electricity ends up as waste heat. Ten times that is 5,000 TWh/yr = 1.8×10¹⁹ J/yr. Ocean volume ≈ 1.335×10⁹ km³, mass ≈ 1.37×10²¹ kg, specific heat ≈ 3,850–4,000 J/kg·K.
If that heat instantly mixed through the entire ocean, the temperature rise is roughly 3×10⁻⁶ °C per year — about 0.0003°C over a century. Undetectable against natural noise.
If instead the heat stayed trapped in shelf waters (not the whole ocean — the scenario you're describing) using ~27 million km² of continental shelf at ~130 m average depth (mass ≈ 3.6×10¹⁸ kg), the rise is roughly 0.001°C per year, assuming zero heat loss to the atmosphere or exchange with the open ocean. In reality, continental shelves are among the most vigorously mixed and ventilated parts of the ocean (tidal currents, wind mixing, exchange with open water), so heat wouldn't actually accumulate there — it would disperse on timescales of days to weeks, keeping any local anomaly far below that ceiling except very close to the source.
Context check: that 1.8×10¹⁹ J/yr is about 0.15% of the ~1.2×10²² J (12 ZJ) the oceans already absorb annually from anthropogenic climate forcing. So even the worst-case, no-mixing shelf estimate is a rounding error next to existing ocean warming trends.
Bottom line: probable range is roughly 10⁻⁶ °C/yr (well-mixed, whole-ocean case) to ~10⁻³ °C/yr (worst-case, heat trapped locally on shelves with no dissipation) — several orders of magnitude below anything oceanographically measurable or ecologically significant, and dwarfed by existing ocean heat uptake from greenhouse forcing.
Every calculation above is ocean-only. The temperature changes (10⁻⁶ °C/yr whole-ocean, up to ~10⁻³ °C/yr worst-case shelf-trapped) are the result of dividing the heat energy by the mass and specific heat of seawater alone. I did not fold in atmosphere, land, or ice mass, and I didn't compute a global mean surface temperature (the metric used for "global warming" figures, which blends ocean, land, and air and uses a much smaller effective heat capacity per unit area near the surface). So this isn't a global-temperature estimate — it's specifically how much the ocean's own heat content would rise if it absorbed 10x the world's data center waste heat with no losses to atmosphere or space.
What concerns might one have about the effect of that change on marine life?
The basin-wide average I calculated is almost meaningless for marine life — organisms don't experience an ocean-wide mean, they experience whatever's happening in the water column right around them. The real concern is the near-field thermal plume at each facility, which could be many degrees above ambient even if the global or shelf-wide average shift is imperceptible. This is the same physical situation as cooling-water discharge from coastal power plants, which is well studied and gives a useful reference point: outfall temperatures are often 5–15°C above ambient at the point of discharge, dropping off with distance and mixing.
A few specific concerns follow from that. Continental shelves host nursery grounds, benthic communities, and fisheries that are disproportionately important relative to their area, so siting heat sources there concentrates risk in ecologically valuable habitat rather than the open ocean. Sessile and low-mobility organisms — corals, shellfish beds, seagrass, benthic invertebrates — can't relocate away from a chronic local plume the way fish can, so they bear continuous thermal stress rather than a brief pulse. Warm water holds less dissolved oxygen, so a persistent plume can create localized hypoxic stress on top of the temperature effect itself, compounding metabolic strain. Chronic, always-on point-source heat (unlike a seasonal heatwave) also removes the recovery windows organisms rely on to reset after thermal stress.
There's also a documented behavioral effect worth flagging: thermal discharges can act as artificial refugia, attracting species that wouldn't normally persist there (tropical or subtropical fish overwintering near warm outfalls is a well-known pattern) or enabling non-native/invasive species to gain a foothold. And any additional warming stacks on top of an already-warming ocean — species and reefs near their thermal tolerance limits from climate change have less buffer before local heat stress tips into bleaching or die-off events.
Finally, heat wouldn't be the only stressor from physically siting hardware on the shelf: biofouling-control chemicals or biocides in cooling systems, construction turbidity, physical habitat loss under the footprint, and low-frequency noise are all plausible co-stressors that would compound with the thermal plume rather than act independently.
Graphics omitted.
Sources:
- [Executive summary – Key Questions on Energy and AI – IEA](https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary)
- [AI: Five charts that put data-centre energy use into context - Carbon Brief](https://www.carbonbrief.org/ai-five-charts-that-put-data-centre-energy-use-and-emissions-into-context)
