Analysis of the Material and Environmental Implications of Global Artificial Intelligence Infrastructure Expansion
全球人工智慧基礎設施擴張之物質與環境影響分析
Introduction
A report by the United Nations University Institute for Water, Environment and Health (UNU-INWEH) examines the ecological costs of artificial intelligence, emphasizing that its impact extends beyond carbon emissions to include significant water and land consumption.
聯合國大學水、環境與健康研究所 (UNU-INWEH) 的一份報告研究了人工智慧的生態成本，強調其影響不僅限於碳排放，還包括大量的水資源與土地消耗。
Main Body
The material requirements of AI are characterized by a substantial physical footprint. In 2025, data centers consumed approximately 448 terawatt-hours (TWh) of electricity; projections indicate this may escalate to 945 TWh by 2030. The UNU-INWEH report posits that AI-specific electricity demand could reach 378 TWh annually by 2030, a volume sufficient to sustain the residential requirements of sub-Saharan Africa for over two years. Furthermore, the water footprint for 2025 is estimated at 9.3 trillion liters, which would theoretically satisfy the global population's drinking water requirements for eighteen months.
AI 的物質需求具有顯著的物理足跡。2025 年，數據中心消耗了約 448 兆瓦時 (TWh) 的電力；預測到 2030 年可能會增加至 945 TWh。UNU-INWEH 報告認為，AI 專用的電力需求到 2030 年每年可能達到 378 TWh，此數量足以供應撒哈拉以南非洲地區超過兩年的居民用電需求。此外，2025 年的水足跡預計為 9.3 兆公升，理論上可滿足全球人口 18 個月的飲用水需求。
Institutional analysis reveals a critical tension between carbon reduction and other ecological metrics. The report notes that a transition to low-carbon energy sources does not inherently ensure low water or land usage. Brazil serves as a primary example, where a hydropower-dominant grid yields a carbon footprint 77% below the global average, yet water and land footprints are nearly triple the global mean. Additionally, the shift from model training to continuous inference has intensified resource consumption, with inference now accounting for 80% to 90% of total energy use in many models. The proliferation of high-resolution AI video generation is identified as a particularly resource-intensive frontier, with energy requirements increasing exponentially relative to resolution and frame count.
機構分析顯示，碳減排與其他生態指標之間存在關鍵緊張關係。報告指出，轉向低碳能源並不必然確保低水耗或低土地使用率。巴西是一個主要例子，其以水力發電為主導的電網使碳足跡比全球平均低 77%，但水足跡與土地足跡幾乎是全球平均值的三倍。此外，從模型訓練轉向持續推論（inference）加劇了資源消耗，推論目前在許多模型中佔總能耗的 80% 至 90%。高解析度 AI 影片生成的普及被視為一個極其耗資源的前沿領域，其能源需求隨解析度與幀數的增加而呈指數級增長。
Geopolitical and social disparities are further exacerbated by the concentration of AI infrastructure. Currently, 90% of specialized cloud infrastructure is situated within the United States and China, leaving over 150 nations without sovereign compute capacity. This asymmetry suggests a systemic divide where the benefits of AI are concentrated in a few states while the material costs—including the extraction of critical minerals and the generation of an estimated 2.5 million metric tons of annual e-waste by 2030—are disproportionately borne by the Global South. Research from the University of Cape Town corroborates these findings, citing 'sovereignty erosion' and urban displacement as significant political and social pressures resulting from the dominance of technology 'hyperscalers.'
地緣政治與社會差距因 AI 基礎設施的集中而進一步加劇。目前，90% 的專業雲端基礎設施位於美國與中國，導致 150 多個國家缺乏主權運算能力。這種不對稱表明了一種系統性分歧，即 AI 的利益集中在少數國家，而物質成本——包括關鍵礦產的開採以及到 2030 年預計每年產生的 250 萬公噸電子垃圾——則由全球南方國家不成比例地承擔。開普敦大學的研究證實了這些發現，指出由技術「超大規模業者」（hyperscalers）主導而導致的「主權侵蝕」與城市人口遷移是顯著的政治與社會壓力。
Conclusion
The current trajectory of AI expansion necessitates a transition toward lifecycle governance and integrated environmental planning to mitigate systemic ecological risks.
目前 AI 擴張的軌跡需要轉向生命週期治理與整合環境規劃，以緩解系統性生態風險。
Vocabulary Learning
The Architecture of 'Nominalization' as a Vehicle for Academic Authority
To bridge the gap from B2 to C2, a student must move beyond describing actions and start describing concepts. The provided text is a masterclass in Nominalization—the linguistic process of turning verbs (actions) or adjectives (qualities) into nouns. This shift transforms a narrative into an analytical discourse.
⚡ The Mechanism: From Dynamic to Static
Compare these two conceptualizations of the same idea:
- B2 (Verb-centric): AI is expanding globally, and this exacerbates geopolitical disparities. (Focuses on the process/action).
- C2 (Noun-centric): The proliferation of AI infrastructure... exacerbates geopolitical and social disparities. (Focuses on the phenomenon as an object of study).
In the text, notice how verbs are replaced by high-register nouns to create a sense of "objective distance":
- Instead of saying "The way the environment is impacted," the author uses "ecological costs."
- Instead of "The way nations lose their power," the author uses "sovereignty erosion."
- Instead of "How resources are used," the author uses "resource consumption."
🎓 Strategic Nuance: The "Abstract Subject"
At the C2 level, you must master the Abstract Subject. By nominalizing a phrase, the author creates a subject that can be analyzed logically rather than chronologically.
Take the phrase: "This asymmetry suggests a systemic divide..."
"Asymmetry" here is not just a word; it is a conceptual anchor. It encapsulates the entire preceding paragraph regarding US/China dominance into a single noun, allowing the writer to apply a new predicate (suggests a systemic divide) to it. This is the hallmark of sophisticated academic English: condensing complex data into a single, manipulatable noun.
🛠️ Linguistic Precision: Collocational Weight
Notice the "weight" of the adjectives paired with these nominalizations. C2 writing does not use generic modifiers. The text employs high-density collocations:
- Sovereign compute capacity
- Integrated environmental planning
- Lifecycle governance
- Disproportionately borne
The Takeaway: To achieve C2 mastery, stop asking "What is happening?" (Verb) and start asking "What is the name of this phenomenon?" (Noun). Shift your focus from the agent of the action to the concept produced by the action.