Recent findings from the United Nations University Institute for Water, Environment and Health and academic researchers indicate that the physical infrastructure supporting artificial intelligence is exerting significant pressure on global energy, water, and land resources.

聯合國大學水、環境與健康研究所及學術研究人員最近的發現指出，支持人工智慧的實體基礎設施正對全球能源、水資源與土地資源造成顯著壓力。

The proliferation of generative AI has necessitated a substantial increase in high-density data center construction. By 2025, global data center electricity consumption reached approximately 448 terawatt-hours, with AI workloads comprising 20% of this total. Projections suggest that by 2030, total consumption may escalate to 945 terawatt-hours, with AI's share increasing to 40%. This energy demand is accompanied by a commensurate rise in carbon emissions, projected to reach 399 million metric tons of CO2 by 2030. Furthermore, the cooling requirements for these facilities are projected to consume 9.3 trillion liters of water by 2030, potentially straining aquifers in water-scarce regions such as Mexico and Uruguay.

生成式 AI 的普及導致高密度數據中心建設大幅增加。到 2025 年，全球數據中心電能消耗達到約 448 兆瓦時，其中 AI 工作量佔總數的 20%。預計到 2030 年，總消耗量可能升至 945 兆瓦時，AI 的佔比將增加至 40%。這種能源需求伴隨而來的碳排放亦隨之增加，預計到 2030 年將達到 3.99 億公噸二氧化碳。此外，這些設施的冷卻需求預計到 2030 年將消耗 9.3 兆公升水，可能使墨西哥與烏拉圭等缺水地區的含水層承受壓力。

Institutional analysis reveals a systemic paradox wherein the pursuit of low-carbon energy may exacerbate other environmental stressors. For instance, a transition from coal to bioenergy may reduce carbon emissions by 72% but increases water and land footprints by 30-fold and 100-fold, respectively. This suggests that sustainability cannot be measured by a single metric. Additionally, the concentration of AI capacity—with 90% situated in the United States and China—has fostered a digital divide. Lower-income nations often bear the externalities of this growth, including mineral extraction and the management of an estimated 2.5 million metric tons of annual electronic waste by 2030, while lacking sovereign computing capacity.

機構分析揭示了一個系統性悖論，即追求低碳能源可能會加劇其他環境壓力。例如，從煤炭轉向生物能源雖可減少 72% 的碳排放，但水資源與土地足跡將分別增加 30 倍與 100 倍。這表明永續性無法透過單一指標來衡量。此外，AI 算力的集中——90% 位於美國與中國——造成了數位落差。低收入國家通常承受此類增長的外部性，包括礦物開採以及到 2030 年預計每年 250 萬公噸電子垃圾的處理，且缺乏主權計算能力。

Stakeholder responses vary between corporate mitigation and governmental regulation. Google has proposed a framework to replenish more water than it consumes by 2030 and invest in local water stewardship. Conversely, several jurisdictions have implemented restrictive measures; Ireland and Singapore have paused new data center approvals due to grid instability and sustainability concerns. Academic perspectives suggest that the current trajectory is driven by a market imperative for continuous growth, which may diverge from planetary boundaries. Consequently, there are calls to reclassify data centers as critical infrastructure, mandating facility-level transparency and integrating AI expansion into national energy and water governance frameworks.

利害關係人的回應在企業緩解措施與政府監管之間有所不同。Google 提出了一個框架，目標是在 2030 年前補充比消耗更多的水，並投資於在地水資源管理。相反，數個司法管轄區採取了限制措施；愛爾蘭與新加坡因電網不穩定與永續性顧慮，暫停了新數據中心的核准。學術觀點認為，目前的發展軌跡是由市場對持續增長的必然要求所驅動，而這可能偏離行星邊界。因此，有呼籲將數據中心重新定義為關鍵基礎設施，強制要求設施層級的透明度，並將 AI 擴展整合至國家能源與水資源治理框架中。

The expansion of AI infrastructure continues to create a tension between technological scalability and environmental sustainability, necessitating a shift toward integrated governance and transparent resource accounting.

AI 基礎設施的擴展持續在技術可擴展性與環境永續性之間造成緊張關係，因此需要轉向整合治理與透明的資源核算。