Google、Nvidia 与 Anthropic 联手 Emerald AI，为更多数据中心争取电网容量

电网软件独角兽 Emerald AI 已与 Google 和 Nvidia 组建联盟，将利用这家创业公司的技术，为更多数据中心争取电网接入空间。

AI Energy Management Alliance（AEMA）希望让需求响应——即数据中心暂时降低用电量的一种机制——成为数据中心开发不可或缺的一部分。AEMA 表示，通过暂停非关键任务并转移部分计算负载，可使额外 100 吉瓦的数据中心接入电网。

除三家创始成员外，Anthropic 以及 AES、Constellation、National Grid 和 NRG Energy 等多家公用事业公司也加入了该联盟。

需求响应是一个已有数十年历史的概念，公用事业公司长期以来一直在使用它。其有效的原因在于，电网是按照峰值负载进行设计的。在一年中的大部分时间里，实际需求远低于电网可支持的最大容量。

传统上，需求响应主要面向工厂等大型用电户。在用电需求高企时，这些用户会同意减少用电，通常是暂停生产或切换至备用发电机。作为交换，公用事业公司会为这项服务付费，且报酬往往相当可观。

如今，数据中心同样可以利用这些项目，通常的做法是启用备用发电机。

Emerald AI 一直在提出一种替代高污染柴油发电机的方案。该公司的软件会协调公用事业公司与数据中心之间的调度请求；数据中心可以暂停非关键任务，或将负载转移至电网仍有余量的其他数据中心。

这家创业公司并不是唯一从事此类技术研发的企业。Google 一直在开发自己的工具，而 Enel X 则让数据中心能够调用不间断电源，以削减用电需求高峰。根据高盛去年发布的一项研究，如果数据中心每次将最大电网用电量限制在 90%、持续数小时，便可能释放 76 吉瓦的容量。

不过，Emerald AI 确实具备一项优势。其软件可将公用事业公司直接连接至数据中心，这应能使数据中心像电池一样快速响应调度请求。该创业公司最近还在由 Energize Capital 和 DCVC 领投的 A 轮融资中筹集了 1.5 亿美元，获得了以更大规模部署该技术所需的资金。

考虑到 AI 计算负载的峰谷特性，以及数据中心快速升降负载的能力，成立类似 AEMA 的组织早该提上日程。该联盟还承诺帮助科技公司和公用事业公司寻找新的数据中心选址，而这项工作对任何一方而言都并不容易。

尽管前景可观，更复杂的需求响应机制目前仍无法解决电网面临的所有问题。Emerald AI 首席科学家 Ayse Coskun 向 TechCrunch 表示，该公司的技术有望减轻行业对新增发电资源的需求，但无法将这种需求完全消除。

Grid software unicorn Emerald AI has formed a coalition with Google and Nvidia that will use the startup’s technology to secure space on the grid for more data centers.

The AI Energy Management Alliance (AEMA) hopes to make demand response — a paradigm in which data centers temporarily reduce their electricity use — an integral part of data center development. Pausing noncritical tasks and shifting some compute loads, AEMA says, could allow an additional 100 gigawatts of data centers to connect to the grid.

The founding trio is joined by Anthropic, and a range of utilities including AES, Constellation, National Grid, and NRG Energy.

Demand response is an old concept that has been used by utilities for decades. It works because the grid is designed with peak loads in mind. For much of the year, demand is well below the maximum the grid can support.

Traditionally, demand response covered large electricity users like factories, which in times of high demand would agree to reduce their usage, typically by pausing production or switching to backup generators. In exchange, the utility would pay them for the service, often handsomely.

Today, data centers can also take advantage of those programs, usually by turning on backup generators.

Emerald AI has been proposing an alternative to dirty diesel generators. The company’s software coordinates requests from utilities with data centers, which can pause noncritical tasks or shift loads to other data centers where the grid has some headroom.

The startup isn’t the only one working on such technology. Google has been developing its own tools, and Enel X lets data centers tap their uninterruptible power supplies to round off peaks in power demand. Limiting max grid usage to 90% for a few hours at a time could let data centers free up 76 gigawatts of capacity, according to a study by Goldman Sachs published last year.

Emerald AI does have an advantage, though. Its software connects utilities directly to data centers, which should allow data centers to respond quickly to requests, much like batteries. The startup also recently raised $150 million in a Series A led by Energize Capital and DCVC, giving it the sort of funding needed to roll out the technology on a wider scale.

Given the peaky nature of AI compute loads, and data centers’ ability to quickly ramp up and down, the creation of something like AEMA is overdue. The coalition also promises to help tech companies and utilities find new sites for data centers, an effort that hasn’t been easy for either party.

Despite its promise, more sophisticated demand response won’t solve all of the grid’s problems today. Ayse Coskun, Emerald AI’s chief scientist, told TechCrunch that her company’s technology promises to blunt the industry’s need for new generating sources, but it won’t eliminate it entirely.