arxiv: 2603.20897 · v3 · submitted 2026-03-21 · 💻 cs.CY · cs.AI· cs.AR

Recognition: 2 theorem links

· Lean Theorem

The data heat island effect: quantifying the impact of AI data centers in a warming world

Andrea Marinoni , Erik Cambria , Weisi Lin , Mauro Dalla Mura , Jocelyn Chanussot , Edoardo Ragusa , Chi Yan Tso , Yihao Zhu

show 1 more author

Benjamin Horton

Authors on Pith no claims yet

Pith reviewed 2026-05-15 06:50 UTC · model grok-4.3

classification 💻 cs.CY cs.AIcs.AR

keywords AI data centersdata heat island effectland surface temperatureremote sensingmicroclimateheat dissipationenvironmental impactsustainable computing

0 comments

The pith

AI data centers raise surrounding land surface temperatures by 2°C on average after operations begin.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper uses decades of remote sensing measurements to track temperature changes around AI data centers worldwide. It reports a consistent average increase of 2°C in land surface temperature once these facilities start running, which creates localized microclimate zones. The authors name this pattern the data heat island effect. They further calculate that more than 340 million people live in areas exposed to the added warmth. If the pattern holds, the effect becomes a measurable factor in how communities experience the growth of AI infrastructure.

Core claim

The central claim is that AI data centers produce a data heat island effect: global remote sensing data show land surface temperatures rising by 2°C on average after operations commence, forming distinct local warming zones whose reach extends to more than 340 million people and carries implications for regional welfare.

What carries the argument

The data heat island effect, the localized temperature elevation driven by heat dissipation from AI data center operations.

If this is right

More than 340 million people could experience measurable local warming from nearby AI facilities.
The data heat island effect adds a new environmental cost to AI scaling that must be weighed in sustainability planning.
Regional welfare and community health metrics will incorporate data center heat outputs in future assessments.
Continued expansion of hyperscale AI infrastructure will enlarge the geographic footprint of these microclimate zones.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Site selection for new data centers could incorporate buffer zones or cooling designs to limit spillover to residential areas.
National energy and land-use policies may need explicit accounting for data center heat when modeling urban heat budgets.
The same remote sensing approach could be applied to other high-power computing clusters to test whether the 2°C figure is specific to AI-scale loads.

Load-bearing premise

The recorded temperature increases result directly from data center heat output rather than concurrent urban development, regional climate shifts, or remote sensing errors.

What would settle it

Temperature records from matched pairs of data center sites and nearby non-data-center industrial zones over identical time windows would show whether the 2°C rise occurs only at data centers.

read the original abstract

The strong and continuous increase of AI-based services leads to the steady proliferation of AI data centres worldwide with the unavoidable escalation of their power consumption. It is unknown how this energy demand for computational purposes will impact the surrounding environment. Here, we focus our attention on the heat dissipation of AI hyperscalers. Taking advantage of land surface temperature measurements acquired by remote sensing platforms over the last decades, we are able to obtain a robust assessment of the temperature increase recorded in the areas surrounding AI data centres globally. We estimate that the land surface temperature increases by 2{\deg}C on average after the start of operations of an AI data centre, inducing local microclimate zones, which we call the data heat island effect. We assess the impact on the communities, quantifying that more than 340 million people could be affected by this temperature increase. Our results show that the data heat island effect could have a remarkable influence on communities and regional welfare in the future, hence becoming part of the conversation around environmentally sustainable AI worldwide.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper flags a 2°C land surface temperature rise around AI data centers but the before-after design leaves the causal link to heat dissipation open to urban growth and trend confounders.

read the letter

The paper's key finding is an estimated 2°C average increase in land surface temperature after AI data centers start operating, which they term the data heat island effect, with potential impacts on more than 340 million people. It applies remote sensing techniques for land surface temperature to the locations of these facilities worldwide. This gives a first global look at local temperature changes tied to their operational start dates. The calculation of affected populations adds a practical dimension by showing how many communities might experience the microclimate shift. The main limitation is in the causal interpretation. The design relies on comparing temperatures before and after commissioning, but it does not appear to include explicit steps to account for simultaneous urban development or broader warming trends. Without matched control areas or adjustments for these factors, the observed rise could reflect general city growth rather than heat from the data centers alone. The abstract calls the assessment robust, yet the lack of methodological specifics on controls leaves the central result open to alternative explanations. Readers working on the environmental costs of AI or urban heat management would find this relevant for highlighting a local effect that goes beyond total energy consumption. It could spark discussion on site selection and cooling strategies. The work deserves peer review. The question is current, the data sources are accessible, and referees could help clarify the robustness of the temperature attribution.

Referee Report

2 major / 1 minor

Summary. The manuscript uses multi-decadal remote-sensing land-surface temperature (LST) observations to estimate that AI data centers produce an average 2°C temperature increase in surrounding areas after operations begin. The authors label this the 'data heat island effect,' quantify exposure for more than 340 million people, and argue that the phenomenon should enter discussions of environmentally sustainable AI.

Significance. A well-identified global estimate of local thermal impacts from hyperscale computing would be policy-relevant for siting decisions and environmental impact assessments. The remote-sensing approach and population-exposure calculation are potentially scalable strengths, but only if the before-after comparison isolates facility heat dissipation from concurrent land-use change and regional climate trends.

major comments (2)

[Abstract] Abstract: the central claim of a 2°C average LST increase after commissioning is presented without any description of the identification strategy, statistical model, control for regional warming trends, or adjustment for concurrent urban development around the sites. This is load-bearing for the causal attribution to data-center heat dissipation.
[Abstract] Abstract and Results: no mention is made of difference-in-differences specifications, synthetic-control weights, land-cover-matched control sites, or error bars. Without these elements the reported temperature jump cannot be distinguished from ordinary urban-heat-island growth driven by non-data-center development.

minor comments (1)

[Abstract] The term 'data heat island effect' is introduced without a formal definition or comparison to the conventional urban heat island literature; a brief literature contrast would clarify novelty.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important opportunities to strengthen the presentation of our identification strategy and robustness checks for the estimated data heat island effect. We address each major comment below and will incorporate revisions accordingly.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim of a 2°C average LST increase after commissioning is presented without any description of the identification strategy, statistical model, control for regional warming trends, or adjustment for concurrent urban development around the sites. This is load-bearing for the causal attribution to data-center heat dissipation.

Authors: We agree that the abstract should briefly outline the identification approach. The manuscript estimates the 2°C average increase via a before-after comparison of multi-decadal remote-sensing LST observations at each site, using pre-commissioning periods to establish local baselines and subtracting regional warming trends through site-specific detrending. In the revised abstract we will add a concise clause describing this temporal design and the use of long-term LST records to isolate post-commissioning deviations. The full statistical model and any regression specifications are already detailed in the Methods section; we will ensure the abstract cross-references them. revision: yes
Referee: [Abstract] Abstract and Results: no mention is made of difference-in-differences specifications, synthetic-control weights, land-cover-matched control sites, or error bars. Without these elements the reported temperature jump cannot be distinguished from ordinary urban-heat-island growth driven by non-data-center development.

Authors: The referee is correct that the current draft relies primarily on site-level before-after LST changes without explicitly reporting difference-in-differences or synthetic-control results. To address potential confounding from concurrent urban development, we will add a new subsection in Results that implements difference-in-differences using land-cover-matched control sites (selected via satellite imagery for comparable pre-period land use) and reports standard errors for the 2°C estimate. Where data permit, we will also include synthetic-control robustness checks for a subset of large facilities. These additions will be summarized in the revised abstract and fully documented in Methods. revision: yes

Circularity Check

0 steps flagged

No significant circularity in observational LST estimation

full rationale

The paper's central result is an empirical average of 2°C LST rise derived from remote-sensing time series by comparing pre- and post-commissioning periods around data centers. No equations, parameters, or derivations are shown that reduce the output to the input by construction. The estimate is presented as a direct data-derived quantity without self-definitional loops, fitted inputs renamed as predictions, or load-bearing self-citations. The analysis remains self-contained against external benchmarks as a straightforward before-after observational comparison.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities beyond the newly coined term 'data heat island effect'; the central claim rests on the unstated assumption that remote-sensing LST differences isolate data-center heat.

axioms (1)

domain assumption Satellite land surface temperature measurements accurately capture localized heat dissipation from data centers without significant confounding from other land-use changes.
Invoked implicitly when attributing the 2°C rise solely to data-center operations.

invented entities (1)

data heat island effect no independent evidence
purpose: Label for the observed local temperature increase around AI data centers.
New descriptive term introduced in the abstract; no independent physical evidence supplied.

pith-pipeline@v0.9.0 · 5509 in / 1301 out tokens · 30247 ms · 2026-05-15T06:50:35.973738+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We estimate that the land surface temperature increases by 2°C on average after the start of operations of an AI data centre... Δ0i(k)=T0i−1k∑j=1kT0i−j
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

the data heat island effect... consistent across the data centres worldwide

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages

[1]

Science of The Total Environment935, 173359 (2024)

Filonchyk, M.,et al.: Greenhouse gases emissions and global climate change: Examining the influence of co2, ch4, and n2o. Science of The Total Environment935, 173359 (2024)

work page 2024
[2]

Available at https://www.unep.org/resources/ emissions-gap-report-2025 (2025) 10

UNEP: Emissions Gap Report 2025: Off Target. Available at https://www.unep.org/resources/ emissions-gap-report-2025 (2025) 10

work page 2025
[3]

Science of The Total Environment584-585, 1040–1055 (2017)

Gunawardena, K.R., M.J.Wells, Kershawa, T.: Utilising green and bluespace to mitigate urban heat island intensity. Science of The Total Environment584-585, 1040–1055 (2017)

work page 2017
[4]

Urban Forestry and Urban Greening62, 127128 (2021)

Tan, J.,et al.: The urban heat island mitigation potential of vegetation depends on local surface type and shade. Urban Forestry and Urban Greening62, 127128 (2021)

work page 2021
[5]

npj Urban Sustainability5(11) (2025)

Yuan, Y., et al.: Surface urban heat island effects intensify more rapidly in lower income countries. npj Urban Sustainability5(11) (2025)

work page 2025
[6]

Available at https://stories.ecmwf.int/ urban-heat-islands-and-heat-mortality (2025)

ECMWF: Urban heat islands and heat mortality. Available at https://stories.ecmwf.int/ urban-heat-islands-and-heat-mortality (2025)

work page 2025
[7]

Available at https://www.un.org/development/desa/pd/ world-urbanization-prospects-2025 (2025)

UN: World Urbanization Prospects 2025. Available at https://www.un.org/development/desa/pd/ world-urbanization-prospects-2025 (2025)

work page 2025
[8]

EPA: What Are Heat Islands? Available at https://www.epa.gov/heatislands/what-are-heat-islands (2025)

work page 2025
[9]

Urban Forestry and Urban Greening37, 33–41 (2019)

Nastrana, M., Kobala, M., Elerb, K.: Urban heat islands in relation to green land use in european cities. Urban Forestry and Urban Greening37, 33–41 (2019)

work page 2019
[10]

Cognitive Computation9, 646–658 (2017)

Cambria, E., Chattopadhyay, A., Linn, E., Mandal, B., White, B.: Storages are not forever. Cognitive Computation9, 646–658 (2017)

work page 2017
[11]

Preprint at https://arxiv.org/abs/2409.11416 (2024)

Li, Y., Mughees, M., Chen, Y., Li, Y.R.: The Unseen AI Disruptions for Power Grids: LLM-Induced Transients. Preprint at https://arxiv.org/abs/2409.11416 (2024)

work page arXiv 2024
[12]

IEEE Power Electronics Magazine12(1), 37–43 (2025)

Chen, M.,et al.: Power for ai and ai for power: The infinite entanglement between artificial intelligence and power electronics systems. IEEE Power Electronics Magazine12(1), 37–43 (2025)

work page 2025
[13]

Luccioni, A.S., Jernite, Y., Strubell, E.: Power hungry processing: Watts driving the cost of ai deployment? In: ACM (ed.) Proceedings of Conference on Fairness, Accountability, and Transparency (2024)

work page 2024
[14]

Power stabilization for ai training datacenters,

Choukse, E., et al.: Power Stabilization for AI Training Datacenters. Preprint at https://arxiv.org/abs/ 2508.14318 (2025)

work page arXiv 2025
[15]

Available at https://www.mckinsey.com/industries/private-capital/our-insights/ how-data-centers-and-the-energy-sector-can-sate-ais-hunger-for-power (2024)

McKinsey: How data centers and the energy sector can sate AI’s hunger for power. Available at https://www.mckinsey.com/industries/private-capital/our-insights/ how-data-centers-and-the-energy-sector-can-sate-ais-hunger-for-power (2024)

work page 2024
[16]

Available at https://www.mckinsey.com/industries/private-capital/our-insights/ scaling-bigger-faster-cheaper-data-centers-with-smarter-designs (2025)

McKinsey: Scaling bigger, faster, cheaper data centers with smarter designs. Available at https://www.mckinsey.com/industries/private-capital/our-insights/ scaling-bigger-faster-cheaper-data-centers-with-smarter-designs (2025)

work page 2025
[17]

Nature Clim Change 3(7), 627–630 (2013)

Masanet, E., Shehabi, A., Koomey, J.G.: Characteristics of low-carbon data centres. Nature Clim Change 3(7), 627–630 (2013)

work page 2013
[18]

Sustainable Computing: Informatics and Systems38, 100857 (2023)

Desislavov, R., Mart´ ınez-Plumed, F., Hern´ andez-Orallo, J.: Trends in ai inference energy consumption: Beyond the performance-vs-parameter laws of deep learning. Sustainable Computing: Informatics and Systems38, 100857 (2023)

work page 2023
[19]

Available at https://www.iea.org/reports/energy-and-ai (2025)

IEA: Energy and AI. Available at https://www.iea.org/reports/energy-and-ai (2025)

work page 2025
[20]

Science367(6481), 984–986 (2020)

Masanet, E.,et al.: Recalibrating global data center energy-use estimates. Science367(6481), 984–986 (2020)

work page 2020
[21]

Available at https: //www.bloomberg.com/graphics/2024-ai-power-home-appliances/ (2024) 11

Nicoletti, L., Malik, N., Tartar, A.: AI needs so much power, it’s making yours worse. Available at https: //www.bloomberg.com/graphics/2024-ai-power-home-appliances/ (2024) 11

work page 2024
[22]

Available at https: //www.jll.com/en-us/newsroom/global-data-center-demand-surges-despite-supply-and-power-constraints (2025)

Steele, K.: Global data center demand surges despite supply and power constraints. Available at https: //www.jll.com/en-us/newsroom/global-data-center-demand-surges-despite-supply-and-power-constraints (2025)

work page 2025
[23]

In: Proceedings of Machine Learning and Systems (2022)

Wu, C.-J.,et al.: Sustainable ai: Environmental implications, challenges and opportunities. In: Proceedings of Machine Learning and Systems (2022)

work page 2022
[24]

Available at https://nepc.raeng.org.uk/policy-work/ engineering-responsible-ai-foundations-for-environmentally-sustainable-ai/ (2025)

Engineering, R.A.: Engineering responsible AI: foundations for environ- mentally sustainable AI. Available at https://nepc.raeng.org.uk/policy-work/ engineering-responsible-ai-foundations-for-environmentally-sustainable-ai/ (2025)

work page 2025
[25]

Remote sensing6(5), 3822–3840 (2014)

Metz, M., Rocchini, D., Neteler, M.: Surface temperatures at the continental scale: Tracking changes with remote sensing at unprecedented detail. Remote sensing6(5), 3822–3840 (2014)

work page 2014
[26]

https://www.datacentermap.com/

DataCenterMap: (2025). https://www.datacentermap.com/

work page 2025
[27]

Available at https://developers.google.com/earth-engine/datasets/catalog/ WorldPop GP 100m pop (2025)

Catalog, E.E.D.: WorldPop Global Project Population Data: Estimated Residential Population per 100x100m Grid Square. Available at https://developers.google.com/earth-engine/datasets/catalog/ WorldPop GP 100m pop (2025)

work page 2025
[28]

Available at https://www.verneglobal.com/blog/ data-center-waste-heat (2024)

Verne: Harnessing Data Center Waste Heat. Available at https://www.verneglobal.com/blog/ data-center-waste-heat (2024)

work page 2024
[29]

Rettie, F.,et al.: Historical trends reveal significant increase in hot-dry extremes in mexico’s baj´ ıo region. Environ. Monit. Assess.198(3), 228 (2026)

work page 2026
[30]

npj Clim Atmos Sci7, 218 (2024)

Tejedor, E.,et al.: Recent heatwaves as a prelude to climate extremes in the western mediterranean region. npj Clim Atmos Sci7, 218 (2024)

work page 2024
[31]

Aerosol Sci Eng (2025)

Souza, A., et al.: Analyzing maximum temperature trends and extremes in brazil: A study of climate variability and anthropogenic influences from 1960 to 2020. Aerosol Sci Eng (2025)

work page 1960
[32]

Available at https://ukgbc.org/our-work/topics/ resilience-roadmap/ (2024)

Council, U.G.B.: Climate resilience roadmap. Available at https://ukgbc.org/our-work/topics/ resilience-roadmap/ (2024)

work page 2024
[33]

https://philsci-archive.pitt.edu/27741 (2025)

Chen, M., Cambria, E., Laschi, C., Mengaldo, G.: Intelligence is physical: Energy, information, and a thermodynamically grounded theory of minds and machines. https://philsci-archive.pitt.edu/27741 (2025)

work page 2025
[34]

IEEE Transactions on Pattern Analysis and Machine Intelligence48(3), 2221–2235 (2026)

Marinoni, A., Li` o, P., Barp, A., Girolami, M.: Improving embedding of graphs with missing data by soft manifolds. IEEE Transactions on Pattern Analysis and Machine Intelligence48(3), 2221–2235 (2026)

work page 2026
[35]

In: Proceedings IEEE International Conference on Impage Processing (2025)

Nema, A., Zhu, H., Lin, W.: Holistic coreset selection for data efficient image quality assessment. In: Proceedings IEEE International Conference on Impage Processing (2025)

work page 2025
[36]

Available at https://openai.com/index/ai-and-compute/ (2025)

Amodei, D., et al.: AI and compute. Available at https://openai.com/index/ai-and-compute/ (2025)

work page 2025
[37]

Neural Computing and Applications34, 10397–10408 (2022)

Pandelea, V., Ragusa, E., Young, T., Gastaldo, P., Cambria, E.: Toward hardware-aware deep-learning- based dialogue systems. Neural Computing and Applications34, 10397–10408 (2022)

work page 2022
[38]

In: Proceedings of IEEE ICASSP (2023)

Pandelea, V., Ragusa, E., Gastaldo, P., Cambria, E.: Selecting Language Models Features Via Software- Hardware Co-Design. In: Proceedings of IEEE ICASSP (2023)

work page 2023
[39]

Preprint at https://arxiv.org/abs/2502.01671 (2025)

Schneider, I., et al.: Life-Cycle Emissions of AI Hardware: A Cradle-To-Grave Approach and Generational Trends. Preprint at https://arxiv.org/abs/2502.01671 (2025)

work page arXiv 2025
[40]

In: ACM (ed.) Proceedings of 29th ACM International Conference on Architectural Support for Programming Languages 12 and Operating Systems (2024)

Patel, P.,et al.: Characterizing power management opportunities for llms in the cloud. In: ACM (ed.) Proceedings of 29th ACM International Conference on Architectural Support for Programming Languages 12 and Operating Systems (2024)

work page 2024
[41]

Computer Methods in Applied Mechanics and Engineering405, 115839 (2023)

Gasick, J., Qian, X.: Isogeometric neural networks: A new deep learning approach for solving parameter- ized partial differential equations. Computer Methods in Applied Mechanics and Engineering405, 115839 (2023)

work page 2023
[42]

IEEE Transactions on Pattern Analysis and Machine Intelligence47(10), 9298–9315 (2025)

Wu, K.,et al.: Efficient distortion-minimized layerwise pruning. IEEE Transactions on Pattern Analysis and Machine Intelligence47(10), 9298–9315 (2025)

work page 2025
[43]

In: Proceedings of NeurIPS (2025)

Zhang, X.: Learning grouped lattice vector quantizers for low-bit llm compression. In: Proceedings of NeurIPS (2025)

work page 2025
[44]

Applied Energy406, 127288 (2026)

d’Orgeval, A.,et al.: Generative ai impact assessment through a life cycle analysis of multiple data center typologies. Applied Energy406, 127288 (2026)

work page 2026
[45]

npj Clean Water4(1) (2021)

Mytton, D.: Data centre water consumption. npj Clean Water4(1) (2021)

work page 2021
[46]

Sensors International6, 100326 (2025)

Butt, M.A., Janaszek, B., Piramidowicz, R.: Lighting the way forward: The bright future of photonic integrated circuits. Sensors International6, 100326 (2025)

work page 2025
[47]

Pal, A.: Adiabatic Logic Circuits, pp. 303–321. Springer, New Delhi (2015)

work page 2015
[48]

Preprint at https://arxiv.org/abs/2510.11119 (2025)

Marinoni, A., Shivareddy, S., Lio’, P., Lin, W., Cambria, E., Grey, C.: Improving AI Efficiency in Data Centres by Power Dynamic Response. Preprint at https://arxiv.org/abs/2510.11119 (2025)

work page arXiv 2025
[49]

Renewable and Sustainable Energy Reviews187, 113761 (2023)

Du, Y.H.,et al.: Dynamic thermal environment management technologies for data center: A review. Renewable and Sustainable Energy Reviews187, 113761 (2023)

work page 2023
[50]

Energy216, 119223 (2021)

He, Z.Q.,et al.: Thermal management and temperature uniformity enhancement of electronic devices by micro heat sinks: A review. Energy216, 119223 (2021)

work page 2021
[51]

Journal of Building Engineering94, 110021 (2024)

Cao, K.,et al.: Comprehensive review and future prospects of multi-level fan control strategies in data centers for joint optimization of thermal management systems. Journal of Building Engineering94, 110021 (2024)

work page 2024
[52]

Nature 515(7528), 540–544 (2014)

Raman, A.P.,et al.: Passive radiative cooling below ambient air temperature under direct sunlight. Nature 515(7528), 540–544 (2014)

work page 2014
[53]

Nature Photonics16(3), 182–190 (2022)

Fan, S., Li, W.: Photonics and thermodynamics concepts in radiative cooling. Nature Photonics16(3), 182–190 (2022)

work page 2022
[54]

Science 382(6671), 691–697 (2023)

Lin, K.,et al.: Hierarchically structured passive radiative cooling ceramic with high solar reflectivity. Science 382(6671), 691–697 (2023)

work page 2023
[55]

Available at https://www.prnewswire.com/apac/news-releases/ i2cool-powers-green-event-management-at-hong-kong-coliseum-with-electricity-free-cooling-technology-302342933

i2Cool: i2Cool Powers Green Event Management at Hong Kong Coliseum with Electricity- free Cooling Technology. Available at https://www.prnewswire.com/apac/news-releases/ i2cool-powers-green-event-management-at-hong-kong-coliseum-with-electricity-free-cooling-technology-302342933. html (2025) 13

work page 2025