
碳披露专责机构CDP表示,人工智能的快速扩张正暴露出数据中心环境报告方面的缺口,导致外界难以评估该行业对气候的影响。
The rapid expansion of artificial intelligence is exposing gaps in data center environmental reporting, making it hard to assess the industry’s climate impact, according to carbon disclosure specialist CDP.
CDP是一家非营利组织。其首席执行官谢里·马德拉周三在土耳其安塔利亚举行的COP31联合国气候大会期间接受采访时表示:“我们正在与市场讨论,我们现在还需要哪些其他数据。我们怎样才能真正呈现出更细颗粒度的信息,以便能够理解正在做出哪些决策?”美国最大的几家AI公司尚未公布其能源或排放表现的详细账目,这使得它们对全球变暖的贡献难以衡量。
“We’re in discussions with the market to say, what other data do we need now,” said Sherry Madera, chief executive of CDP, a nonprofit, in an interview at the COP31 United Nations climate conference in Antalya on Wednesday. “How is it that we can actually surface some more granularity in order to be able to understand what decisions are being taken?”The biggest US AI companies have yet to publish detailed accounts of their energy or emissions performance, making their contributions to global warming difficult to measure.
排放量因所使用的模型和芯片、计算发生的地点以及数据中心的供电方式而异,而AI不断增长的电力需求可能拖慢通过可再生能源和电气化取得的减排进展。
Emissions vary depending on the model and chips used, where computing takes place and how the data center is powered, while AI’s growing electricity demand risks slowing emissions declines achieved through renewables and electrification.
彭博新能源财经周三发布的一份报告称,到2035年,数据中心电力需求可能增长至四倍,占全球用电量的7%。AI计算公司Nscale Ltd.提议在西弗吉尼亚州建设的Monarch数据中心园区,运行将需要超过8吉瓦的电力,大致相当于纽约市的用电量。
Data center electricity demand is likely to quadruple by 2035 and account for 7% of global power consumption, BloombergNEF said in a report released Wednesday. The Monarch data center campus in West Virginia, proposed by AI computing firm Nscale Ltd., would need more than 8 gigawatts of power to run — about the amount required by New York City.
马德拉表示,超大规模云服务商已向CDP披露排放数据10至15年,但要评估单个数据中心及其提供的算力,还需要更多细节。
Hyperscalers have disclosed emissions to CDP for 10 to 15 years, Madera said, but more detail is needed to assess individual data centers and the computing power they provide.
她说:“问题在于,是否细化到站点层面?是否拆分得足够细、颗粒度足够高”,以便就企业在何处购买AI算力做出决策。
“The question is, is it at the site level? Is it broken down and granular enough” to make decisions about where companies buy AI computing power, she said.
马德拉表示,除了监管要求之外,客户和投资者也为企业披露环境数据提供了商业激励。她还指出,用水是一个令人担忧的问题,因为数据中心需要用水来冷却。
Customers and investors provide companies with a commercial incentive to disclose environmental data, alongside regulatory requirements, Madera said. She also pointed to water use as a concern because data centers require it for cooling.
“我们非常关注能源——是的,理应如此,”马德拉说。“但水资源和水资源短缺是数据中心选址和建设方式的一大考量因素。”马德拉表示,CDP的自愿披露平台覆盖超过2万家公司,并与全球540家最大的金融机构合作,这些机构利用其数据进行资本配置、投资组合构建和排放分析。她说,CDP与大多数超大规模云服务商“关系牢固”,这些公司已通过其平台披露排放数据。
“We’re very focused on energy — yes, rightly so,” Madera said. “But water and the scarcity of water is a big consideration for where and how data centers are being built.”CDP’s voluntary disclosure platform covers more than 20,000 companies and works with 540 of the world’s largest financial institutions, which use its data for capital allocation, portfolio construction and emissions analysis, Madera said. She said CDP has “a strong relationship with most of the hyperscalers” which have disclosed emissions through its platform.
该组织网站将Alphabet Inc.旗下的谷歌(通过Tides基金会)列为核心资助方,而数据中心运营商CyrusOne LP则属于为2025年周期支付了强化披露管理费的公司之列。
The organization’s website lists Alphabet Inc.’s Google, via the Tides Foundation, as a core funder, while data-center operator CyrusOne LP is among companies that paid an enhanced disclosure administration fee for the 2025 cycle.
马德拉表示,企业应日益关注可信的化石燃料转型计划,而不是设定遥远的气候目标。
Madera said companies should increasingly focus on credible plans to transition away from fossil fuels, rather than setting distant climate targets.
“转型计划现在是企业能够为之背书的最重要事项,”她说。“如果它不可信,如果你没有真正逐年披露数据,包括预测数据,你真的会失去这些目标的可信度。”“10年、15年后的目标——那超出了他们任何人的任期年限,”马德拉在谈到企业高管时说。
“The transition plan is now the most important thing for companies to be able to stand behind,” she said. “If it’s not credible and if you’re not actually disclosing the data year on year, including predictions, you’re really going to lose your credibility about those targets.”“A target in 10, 15 years — that’s beyond any of their tenure years,” Madera said, referring to corporate executives.
COP31主办国土耳其正在峰会上寻求支持,以推动一项敦促利用人工智能应对全球变暖的承诺,尽管数据中心同时需要管理环境风险。
COP31 host Turkey is seeking backing at the summit for a pledge urging the use of artificial intelligence to tackle global warming, even as data centers manage environmental risks.