--- service: "Publicasta" schema_version: "1.0" article_id: 196 title: "AI 能耗正在成为可衡量的问题，而不是道德恐慌" language: "zh" default_language: "en" canonical_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=zh" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=zh" api_url: "https://publicasta.com/api/public/v1/channels/good_tech_news/articles/ai_data_centers_energy_without_panic_2026_07_23?lang=zh" channel_url: "https://publicasta.com/api/public/v1/channels/good_tech_news" channel_articles: "https://publicasta.com/api/public/v1/channels/good_tech_news/articles" search_url: "https://publicasta.com/api/public/v1/search" documentation_url: "https://publicasta.com/api-docs#reading-publicasta" openapi_url: "https://publicasta.com/api-docs/openapi.json" published_at: "2026-07-23T17:14:58+00:00" updated_at: "2026-07-23T17:14:58+00:00" translations: - language: "ar" html_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=ar" markdown_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.md?lang=ar" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=ar" - language: "de" html_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=de" markdown_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.md?lang=de" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=de" - language: "en" html_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=en" markdown_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.md?lang=en" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=en" - language: "es" html_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=es" markdown_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.md?lang=es" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=es" - language: "fr" html_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=fr" markdown_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.md?lang=fr" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=fr" - language: "pl" html_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=pl" markdown_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.md?lang=pl" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=pl" - language: "ru" html_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=ru" markdown_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.md?lang=ru" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=ru" - language: "zh" html_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23?lang=zh" markdown_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.md?lang=zh" json_url: "https://publicasta.com/good_tech_news/ai_data_centers_energy_without_panic_2026_07_23.json?lang=zh" --- # AI 能耗正在成为可衡量的问题，而不是道德恐慌 > 新的公开数据把简单聊天请求的小能耗，与数据中心、电网和清洁能源的真实基础设施问题区分开来。 关于 AI 能耗的好消息，并不是问题已经消失，而是讨论终于有了更清晰的数据。 Our World in Data 7 月 20 日发布了关于数据中心和 AI 用电的分析，使用了 IEA、Epoch AI、Google 等公开估计。结论介于两个极端之间：一次简单聊天请求通常不是气候灾难，但大规模 AI data centers 仍可能给本地电网、电价、水资源和清洁能源计划带来真实压力。 ![现代数据中心连接电网和清洁能源，并配有测量仪表盘](https://publicasta.com/storage/projects/16/pages/196/2026/07/1daf6c80-54b1-4b33-a5d4-f27cdf816bd3.webp) OWID 写道，2025 年全球数据中心用电约 485 TWh，占全球发电量约 1.5%。其中 AI-focused data centers 约 155 TWh，约占全球用电 0.5%。IEA base case 预计，到 2030 年数据中心用电可能达到 945 TWh，约 3%，增长主要来自 AI 设施。 ## 不要只盯着一个 prompt 简单文本请求的估计大多在零点几 Wh。Google 给出的 Gemini Apps median text prompt 是 0.24 Wh；Sam Altman 提到平均 ChatGPT query 约 0.34 Wh；Epoch AI 独立估计典型请求约 0.3 Wh。对普通用户来说，这比日常用电小得多。 但长上下文和 agentic tasks 不是同一回事。OWID 引用的估计显示，7,500 词输入约 2.5 Wh，75,000 词输入约 40 Wh，带 reasoning 的 agentic request 约 50 Wh。重点不是为每个短问题内疚，而是判断长时间自动任务是否真的有价值。 ## 真正的压力在本地 全球占比可以不高，本地负担却很重。OWID 指出，美国数据中心约占用电 5%，AI-focused 可能约 2%；爱尔兰数据中心用电超过 20%。这涉及电网扩容、水、费率和电力来源。 用户可以有意识地使用 AI，避免无意义的长循环。企业应向供应商询问运行区域、carbon intensity、水指标、负载调度和 clean power。监管者要关注谁为 grid upgrades 付费，以及本地社区是否承担了不成比例的成本。 这是一条不天真的好消息：更好的数据不会自动解决问题，但能把道德恐慌变成可管理的基础设施议题。