三分之一的核电站所有者正在与科技公司洽谈增强人工智能数据中心的能力
Third Of Nuclear-Plant Owners In Talks With Tech Firms To Power Up AI Data Centers

原始链接: https://www.zerohedge.com/commodities/third-nuke-power-plant-owners-talks-tech-firms-powering-ai-data-centers

包括亚马逊在内的主要科技公司正在与公用事业公司合作,以核能为能源建设人工智能数据中心。 这一发展是在四月初向订阅者公布的“Next AI Trade”投资趋势的出现之后出现的。 报告显示,大约三分之一的美国核电站所有者正在与科技巨头讨论,向即将建成的数据中心出售电力,以满足新兴的人工智能行业的需求。 值得注意的是,据报道,亚马逊网络服务 (AWS) 正在与美国领先的核电站所有者 Constellation Energy 敲定一项协议,为位于东海岸的数据中心获得可靠的低碳核电。 在此之前,AWS 收购了 Talen Energy 占地 1,200 英亩的数据中心园区,该园区位于宾夕法尼亚州卢泽恩县 2.5 GW 核电设施附近。 然而,这种安排可能会导致不同用户之间对电力资源的竞争加剧,并可能提高成本,从而可能阻碍减排目标。 因此,专家们越来越多地主张加强基础设施投资,特别是建设更多核电站,主要利用先进的小型模块化反应堆,以解决北美人工智能数据中心、经济电气化和工业复苏带来的不断升级的电力需求。 北美电力可靠性公司提供的数据表明,未来五年电力需求扩张可能从2.6%激增至4.7%。 资料来源:NERC 2022 长期可靠性评估; 2023 年电网战略。请注意,与传统数据中心相比,人工智能数据中心消耗的电力要多得多。 为人工智能数据中心寻求核电的科技公司之间的相互作用导致核电站关闭问题出现转变,并促使白宫启动旨在恢复美国核电生产的计划。 从宾夕法尼亚州到伊利诺伊州,广阔的土地区域显示出过剩的电力供应,使它们成为潜在的人工智能数据中心扩张的可行地点。 尽管如此,人们仍然担心这些发展对整体电网性能和最终消费者负担能力的影响。

相关文章

原文

Big tech firms like Amazon are actively speaking with utility companies to build artificial intelligence data centers powered by clean, reliable nuclear energy. This move aligns with the 'Next AI Trade' investing theme, which was introduced to pro-subs in early April. 

"The owners of roughly a third of US nuclear power plants are in talks with tech companies to provide electricity to new data centers needed to meet the demands of an artificial-intelligence boom," the Wall Street Journal says. 

In particular, WSJ sources say Amazon Web Services is securing a deal with Constellation Energy, the largest owner of US nuclear power plants, to supply clean, reliable atomic power to a data center on the East Coast. 

This is not the first time Amazon has negotiated with utilities about nuclear power.

In March, AWS purchased power provider Talen Energy's 1,200-acre data center campus, which is directly adjacent to its 2.5 gigawatt (GW) nuclear power at the Salem Township site in Luzerne County, Pennsylvania. 

Susquehanna nuclear plant in Salem Township, Penn., along with the data center recently purchased by Amazon. (Photo: Talen Energy)

WSJ pointed out that tech companies demanding nuclear power for AI data centers "would be effectively diverting existing electricity resources" and "could raise prices for other customers and hold back emission-cutting goals." 

The nuclear-tech marriage has forced many Wall Street analysts in recent months to jump on the powering up America theme. Greater load demands are projected in the coming years due to AI data centers, electrification of the economy, and reshoring efforts. 

Recent estimates from North American Electric Reliability Corporation forecast that electricity growth over the next five years will increase from 2.6% to 4.7%, driven by major utilities revising up their estimates.

Source: NERC - 2022 Long-Term Reliability Assessment (as of December-2022). Grid Strategies - The Era of Flat Power Demand is Over (as of December-2023).

The chart below shows data center power consumption by providers/enterprises in gigawatts (left) and share percent (right).

Source: McKinsey,

Notice how AI data centers demand an incredible amount of power compared to traditional data centers. 

Tech's push to acquire nuke power for AI data centers has been enough to reverse the tide of plant closures and spur the White House to begin developing a new push to revive nuclear (read: "In Historic Reversal, US To Restart A Shut Down Nuclear Power Plant For The First Time Ever"). 

Constellation's president and CEO, Joseph Dominguez, told WSJ that large swaths of the land from Pennsylvania to Illinois are oversupplied with power and could be future sites for AI data centers.

While it's too early to forecast just how much power AI data centers will need, one concern is that existing electricity resources are being diverted off-grid to data centers, which could drive up power prices for consumers. 

To solve this, more nuclear power plants will need to be constructed, and the winning plant design will likely be advanced small modular reactors.

联系我们 contact @ memedata.com