Do Data Centers Really Consume Too Much Water? The Burger Comparison

Data centers are under scrutiny. Arizona limits their construction due to water scarcity. Microsoft saw its water consumption increase by 34% between 2021 and 2022. The Guardian denounces xAI facilities violating environmental rules. The European Union now requires mandatory annual reports on water and energy consumption. Greenpeace regularly targets cloud giants. The message is clear: data centers drink too much water and threaten drinking water resources.

This narrative deserves nuance. Not to minimize the environmental impact of digital technology, but to put it in perspective with our other consumption patterns. A recent SemiAnalysis article compares the water consumption of Colossus 2, one of the world’s largest data centers, with that of a burger restaurant. The results are surprising. This article builds on that analysis with our perspective as cloud practitioners who have supported companies in their migration for 15 years.

Colossus 2: 346 Million Gallons Per Year

Colossus 2 is xAI’s data center in Memphis that powers future generations of Grok. With a critical IT capacity of 400 MW, it is one of the largest facilities in the world. SemiAnalysis calculated its annual water consumption by detailing each item: cooling, power generation, chip manufacturing.

The data center uses a hybrid cooling system with about 130 dry coolers and 135 adiabatic units. Dry coolers operate in a closed loop and consume very little water beyond initial filling. Adiabatic units use a water mist to pre-cool intake air, improving heat rejection during hot conditions while consuming less water than traditional evaporative cooling.

Assuming a PUE of 1.15, a utilization rate of 70%, and wet operation 30% of the time, SemiAnalysis arrives at 267 million gallons evaporated per year for cooling. Add 66 million gallons withdrawn but not evaporated for adiabatic equipment flushing. The data center currently uses simple-cycle turbines that do not consume water for power generation. Finally, the water footprint of chips (about 200,000 GPUs, 100,000 CPUs, 1.6 million HBM3e stacks) represents 64 million gallons amortized over 5 years, or 13 million per year.

Total: 346 million gallons per year, or about 950,000 gallons per day. This figure matches public estimates circulating about Colossus 2. It represents a WUE (Water Usage Effectiveness) of 0.51 liters per kWh, which is reasonable for a data center of this size with hybrid cooling.

An In-N-Out Restaurant: 147 Million Gallons Per Year

SemiAnalysis then calculated the water footprint of an average In-N-Out restaurant. The methodology distinguishes three types of water: green water (rain absorbed by soil), blue water (surface or groundwater usable), and grey water (wastewater). For a fair comparison with Colossus 2, only blue water is counted.

The analysis focuses on the Double-Double, In-N-Out’s iconic burger. Each ingredient is weighed and its water intensity (liters per kg) is calculated. Beef represents 95% of the total footprint. This water comes mainly from irrigation of feed crops (alfalfa, corn) that feed cattle. Water intensity varies greatly by region. The American Southwest, where In-N-Out sources, shows much higher intensity than other regions due to dry climate requiring more irrigation.

Integrating regional intensities for beef, California tomatoes, iceberg lettuce and other ingredients, SemiAnalysis arrives at 245 gallons per burger (927 liters). This is lower than public estimates of 650-700 gallons which likely include green water and use national averages rather than regional ones.

An average In-N-Out restaurant generates about $5.8 million in annual revenue. Assuming burgers represent 60% of revenue and an average price of $5.80, that gives 600,000 burgers per year. Multiply by 245 gallons per burger: 147 million gallons per year for a single restaurant.

The Game-Changing Ratio: 2.5 Restaurants for One Giant Data Center

Colossus 2 consumes 346 million gallons per year. An In-N-Out restaurant consumes 147 million gallons per year. The ratio is 2.5 to 1. One of the world’s largest data centers consumes as much water as 2.5 burger restaurants.

There are over 400 In-N-Out restaurants and hundreds of thousands of other burger restaurants in the United States. If data center water consumption is a problem, the food industry should receive far greater attention. But no one proposes limiting restaurant construction because of water scarcity in Arizona.

The usual argument is that food is a basic need, unlike data centers. This argument avoids the real question: what economic and social value does each gallon of consumed water generate? Does a data center training AI models used by millions for medical research, education, productivity or artistic creation produce less value than a burger?

SemiAnalysis pushes the calculation further. With the same assumptions on Colossus 2, plus some technical parameters on prefill and decode throughput, the data center can generate up to 3.9 quadrillion output tokens per year. That represents 8.9 million tokens per gallon of water. At 245 gallons per burger, a single burger equals 2.7 billion output tokens.

Assuming 30 queries per day with an average output of 375 tokens, one burger equals 668 years of daily Grok usage. You can use Grok 30 times a day, every day, for over six centuries with the water footprint of a single burger.

Why the Debate Is Biased

The debate on data center water suffers from three biases. The first is the absence of comparison. Numbers are presented absolutely without context. 346 million gallons per year seems huge until you compare it to agriculture, which represents 70% of global freshwater consumption. A kilo of beef requires between 2,000 and 15,000 liters of water depending on regions and production methods. A kilo of tomatoes requires 214 liters. The food industry consumes orders of magnitude more water than the digital industry, but it is rarely at the center of environmental criticism.

The second bias is ignorance of technical nuances. Not all data centers consume the same amount of water. Cooling architecture changes everything. A dry cooling system consumes very little water beyond initial circuit filling. An open evaporative system consumes much more. An adiabatic system like Colossus 2 sits in between. Microsoft announced in December 2024 that all its new data centers designed from August 2024 will use closed-loop cooling with zero water evaporation. These facilities will come online in late 2027.

Location also matters. Building a data center with evaporative cooling in Arizona, where water is scarce, is problematic. Building it in Ireland or Norway, where water is abundant and climate is cool, poses far fewer problems. AWS, Google and Microsoft publish detailed reports on their water consumption by region and reduction efforts. These reports show that hyperscalers optimize their architectures based on local constraints.

The third bias is the lack of consideration for recycling and reuse. xAI is building a water treatment plant in Memphis that will recycle municipal wastewater to supply Colossus 2 cooling. This plant will produce more water than the data center consumes, technically making it a net-zero consumption facility. This type of initiative is rarely mentioned in critical articles.

What This Means for Your Business

If you hesitate to migrate to cloud for environmental reasons, these numbers should reassure you. Cloud data centers are generally more efficient than on-premise servers for three reasons.

The first is pooling. A server in your office running at 15% utilization consumes almost as much energy and cooling water as a server running at 80%. In a cloud data center, resources are shared among thousands of customers, achieving much higher utilization rates. Fewer servers for the same workload means less total consumption.

The second is optimization. AWS, Google and Microsoft invest heavily in energy and water efficiency because it directly reduces their operating costs. They use state-of-the-art cooling systems, thermal optimization algorithms, and renewable energy sources. Your SMB has neither the resources nor the expertise to achieve this level of efficiency with internal servers.

The third is location. Hyperscalers build their data centers in regions where energy is abundant and renewable, and where water is available. They avoid (or should avoid) water-stressed areas. Your server in a Mulhouse office uses local electricity and water without particular optimization.

We have supported companies in Alsace in their cloud migration for 15 years. None have seen their environmental footprint increase after migration. Most see a significant reduction in total energy consumption thanks to cloud resource pooling and optimization.

Looking at Real Problems

This article does not claim that data centers have no environmental impact. They do. Data center water and energy consumption is increasing with AI and cloud growth. This growth must be monitored and regulated, especially in water-stressed areas.

But the current debate lacks perspective. Pointing fingers at data centers while ignoring that agriculture consumes 70% of global water, that the food industry generates massive water footprints, and that our daily consumption habits (meat, clothing, transport) have far greater impacts is inconsistent.

SemiAnalysis figures show that one of the world’s largest data centers consumes as much water as 2.5 burger restaurants. A single burger equals 668 years of daily AI model usage. These comparisons do not aim to absolve the digital industry of its responsibilities. They aim to put orders of magnitude in perspective.

If you worry about the environmental impact of your IT infrastructure, migrating to cloud is probably the best decision you can make. Hyperscalers have the resources, expertise and economic incentives to optimize their water and energy consumption. Your SMB has neither. And if you really want to reduce your water footprint, start by looking at what you eat for lunch.

This article is inspired by the analysis “From Tokens to Burgers: A Water Footprint Face-Off” published by SemiAnalysis in January 2026, with LCMH’s perspective and experience supporting companies in their cloud migration.