从代币到汉堡：一场用水足迹的正面较量

Adifferentperspectiveonthedatacenterwaterdebate,forgettokens/wattortokens/dollar,it'sabouttokens/burger,howmanyIn-N-Outsisequivalenttotheworld'slargestdatacenter 以数据中⼼⽤⽔争论的不同视⻆，别再考虑每瓦代币或每美元代币，⽽是看每个汉堡的代币⸺世界上最⼤的数据显示中⼼相当于多少家In-N-Out JAN16,2026 2026年1⽉16⽇∙PAID∙已付费 Tokens and burgers are not two ideas you often see paired in the same title. If you’rewondering what they could possibly have in common, stick with us for this short piece,because today we’re putting them face to face in a duel over a hot topic - datacenter waterusage. If you want a different perspective on what a datacenter actually consumes, plus acouple of new cool metrics to use in your Friday evenings (did you hear about tokens perburger?), stay tuned. This one’s a meaty post. 代币和汉堡并不是你经常会在同⼀标题中看到的两种概念。如果你在想它们可能有什么共同点，请继续读完这篇短⽂，因为今天我们要把它们放在⼀起较量⼀个热门话题——数据中⼼的⽤⽔量。如果你想从不同⾓度了解数据中⼼实际消耗的东西，并获得⼏项可以在周五晚上⽤起来很酷的新指标（你听说过“每个汉堡的代币数”吗？），请继续关注。这是⼀篇内容丰富的⽂章。 WhattheNumbersOverlook数字忽视的⽅⾯ Datacenter water usage is under increasing scrutiny, with projects even paused or canceled.See recent discussions in Arizona. We think the debate is overstated, as the numbers are oftenmisleading and key variables are frequently ignored. By this we mean the cooling architecture(one of the main drivers of water usage), power source, location and local water scarcity orwater source, among others. In addition, datacenter water consumption is often taken at facevalue and rarely put in perspective versus other everyday industries. 数据中⼼⽤⽔正受到越来越多的审视，有些项⽬甚⾄被暂停或取消。参见亚利桑那州的近期讨论。我们认为这场争论被夸⼤了，因为相关数字常常具有误导性，且关键变量经常被忽视。我们所指的包括冷却架构（这是⽤⽔的主要驱动因素之⼀）、电⼒来源、位置以及当地⽔资源的稀缺程度或⽔源类型等。此外，数据中⼼的⽤⽔量往往被直接采⽤，很少与其他⽇常⾏业进⾏对⽐来审视其合理性。 Besides, there’s no standard for water accounting, which makes comparisons messy. Do youinclude training runs and embedded supply-chain water, or only onsite evaporation andconsumption? A lot of nuances that makes the “datacenters are choking the world” headline hard to assess. 另外，⽬前没有统⼀的⽤⽔核算标准，这使得⽐较变得混乱。你是否应当把训练运⾏和供应链中隐含的⽤⽔算在内，还是仅仅计算现场蒸发和消耗？有许多细微差别，使得“数据中⼼正在掐死世界”的头条难以评估。 To contribute a different angle to the debate, we decided to put face to face one of the world’sbiggest datacenters (want to know which place?Watch our latest Youtube video!) and one ofthe most loved elements in humanity, burgers. More specifically, we’ll calculate and comparethe overall water footprint of Elon Musk’s Colossus 2 Memphis datacenter, the Macrohardone, and an average In-N-Out store. Does Macrohard’s beat an In-N-Out store? Let the duelbegin. 为了给这⼀争论贡献⼀个不同的视⾓，我们决定将世界上最⼤的⼀个数据中⼼（想知道是哪个地⽅？观看我们最新的 YouTube 视频！）与⼈类最受喜爱的元素之⼀——汉堡——进⾏正⾯较量。更具体地说，我们将计算并⽐较埃隆·马斯克的 Colossus 2 孟菲斯数据中⼼（Macrohard 的那个）与⼀家普通 In-N-Out 门店的总体⽔⾜迹。Macrohard 的会输给 In-N-Out 吗？让决⽃开始。 Round1:Colossus2第⼀回合：Colossus2 Let’s start with Colossus 2, xAI’s datacenter that will power future generations of Grok. As wecovered inour report on Colossus 2and recent media posts, based on our satellite footage andthe cooling equipment in the facility, the datacenter is in near term progress to get to aCritical IT Capacity of 400MW. Although it is expected to expand to more than 1GW, for nowwe’ll calculate the water footprint in its current state. 让我们从 Colossus 2 开始，这是 xAI 的数据中⼼，将为未来⼏代 Grok 提供算⼒。正如我们在关于 Colossus 2 的报告和近期媒体发布中所述，基于我们的卫星影像和设施中的冷却设备，该数据中⼼在短期内正朝着 400MW 的关键 IT 容量发展。尽管预计将扩展到超过 1GW，但⽬前我们将计算其在当前状态下的⽔⾜迹。 What can we expect initially? A quick search gives thefirst hints: “Colossus 2 could consumeas much as 1 million gallons of water per day”. That sounds like a lot, but when you try to seehow thatfigure is derived, you rarely get much detail. That’s why we decided to do run somenumbers and see what we get. 最初我们可以期待什么？⼀个快速搜索给出了第⼀个线索：“Colossus 2 每天可能消耗多达 100 万加仑的⽔”。这听起来很多，但当你试图查明这个数字是如何得出的时，通常很难找到详细说明。这就是为什么我们决定做⼀些计算，看看结果如何。 Let’sfirst define what we mean by water footprint, and what we will consider in ourcalculation. When studying a datacenter’s overall water usage, we can consider both directand indirect water footprint sources. More specifically, we will split water usage into direct(primarily cooling, initial loopfill and power generation) and indirect (water usage embeddedin the supply chain, mainly chip manufacturing, although you could also considerconstruction water, etc.). For simplicity, we will focus on the three most materials components: cooling, power generation, and chip manufacturing. ⾸先让我们定义⼀下所谓的⽔⾜迹，以及在计算中会考虑哪些内容。在研究数据中⼼的整体⽤⽔时，我们可以考虑直接和间接的⽔⾜迹来源。更具体地说，我们将把⽤⽔分为直接（主要是冷却、初始回路注⽔和发电）和间接（供应链中嵌⼊的⽤⽔，主要是芯⽚制造，尽管也可以考虑施⼯⽤⽔等）。为简单起见，我们将关注三种最重要的组成部分：冷却、发电和芯⽚制造。 You are all aware that chip power translates into heat that needs to be dissipated toguarantee good performance, and different cooling architectures can be used, as we’vecovered in previous articles. The most relevant point for our analysis here is the distinctionbetween dry, wet and adiabatic cooling systems. Dry cooling is largely closed loop (there’s anirrelevant amount of tiny evaporation in the pipes) and uses little water beyond the initialfill.Wet cooling uses evaporation in o