AI-ChatGPT vs Tech-Water-Impact
In an era where technology's environmental footprint is under increasing scrutiny, it's essential to understand the water usage across various technologies. This page delves into a comprehensive analysis comparing the water impact of AI, particularly OpenAI's ChatGPT, with traditional technologies such as car washes and household energy consumption. By examining real-world data and scaling it to broader contexts, we highlight that while ChatGPT does consume water, its usage is significantly more efficient compared to other common technologies. This comparison aims to provide a balanced perspective on the water footprint of AI versus other everyday technologies.
AI and Tech Companies such as OpenAI Google, Microsoft, and Amazon also operate massive data centers with similar water usage for cooling. ChatGPT's water use is part of a broader trend in AI and tech industry practices.
Historical Context: Traditional industrial technologies (e.g., power plants, manufacturing) have historically used and often wasted more water than modern data centers.
Efficiency Improvements process: OpenAI and other tech companies continuously work on improving the efficiency of their data centers, adopting greener technologies to reduce water and energy usage.
Renewable Energy and Cooling Innovations: Many data centers are now powered by renewable energy sources and use innovative cooling solutions (like liquid cooling and using natural airflows) to minimize environmental impact.
How Is Water Utilized in the Development, Operation, and Running of AI/ChatGPT Applications?
● Direct Water Usage:
This refers to water directly used in the operation of a technology, such as cooling systems for data centers.
Data Centers: AI models, including ChatGPT, are hosted on servers in data centers that consume water primarily for cooling purposes. The energy-intensive process of running these servers generates heat, and water is often used in cooling systems to dissipate this heat.
Training Models: Training large AI models involves significant computational power, which indirectly translates to water usage through the energy required to power and cool the computational infrastructure.
● Indirect Water Usage:
This includes the water footprint associated with manufacturing, maintenance, and energy production required to run the technology.
Water Usage in OpenAI's ChatGPT
● Qualitative Aspects:
1. Data Centers:
Cooling Systems: Data centers, where ChatGPT models are hosted, use significant amounts of water for cooling purposes. This is critical for maintaining optimal operating temperatures of the servers that run these AI models.
Energy Consumption: The computational power required to train and run AI models like ChatGPT is substantial, translating to high energy usage. Cooling these systems efficiently often involves water-cooled chillers or evaporative cooling methods.
2. AI Model Training:
Resource Intensive: Training large AI models involves intensive computational tasks, which generate a considerable amount of heat. The cooling systems in data centers use water to manage this heat, ensuring the equipment functions properly.
3. Operational Efficiency:
Advanced Cooling Technologies: OpenAI, like many tech companies, uses advanced cooling technologies to minimize water usage while maximizing cooling efficiency. This includes using air cooling where possible and implementing water-saving measures in their data centers.
● Quantitative Aspects:
This includes the water footprint associated with manufacturing, maintenance, and energy production required to run the technology.
1. Water Usage Metrics:
» Water Usage Effectiveness (WUE): This metric measures the liters of water used per kilowatt-hour (kWh) of energy consumed by data centers. A typical WUE value might range from 0.1 to 1.5 liters per kWh, depending on the cooling technology and climate conditions.
» PUE (Power Usage Effectiveness): While not directly a water metric, PUE is related as it measures the efficiency of data centers' energy usage, impacting the cooling needs and thus water consumption. Lower PUE values (closer to 1.0) indicate more efficient data centers.
2. Case Studies and Data:
» General Data Center Water Use: On average, a medium-sized data center might consume hundreds of thousands of gallons of water per day for cooling purposes. For example, a 15 MW data center can use up to 360,000 gallons of water per day.
» OpenAI Specifics: While exact numbers for OpenAI's water use are not publicly available, we can estimate based on industry averages. Assuming OpenAI's infrastructure aligns with typical large-scale data centers, their water usage for cooling could be significant but comparable to other large tech firms.
Technical Comparison: ChatGPT and the (Urban) Car Wash Industry
Los Angeles, California U.S. case-study
ChatGPT session water use reference: 1.2 liters, equivalent to 5.25 to 5.75 text pages (letter-size sheet). 5.5 pages will be taken as a single reference. Also, incorporating the population of Los Angeles, estimating the number of ChatGPT users, and calculating the average session to estimate the total number of ChatGPT sessions.
Analysis for Los Angeles, USA
Population Data
Population of Los Angeles County: approximately 10 million people.
ChatGPT Usage Estimation
User Penetration:
Assume 20% of the population uses ChatGPT (a reasonable estimate given the adoption of AI technologies or just a percentage of reference).
Number of ChatGPT Users:
10
,
000
,
000
×
0.20
=
2
,
000
,
000
10,000,000×0.20=2,000,000 users.
Usage Frequency:
Assume each user has an average of 5 sessions per month.
Sessions per Year per User:
5
×
12
=
60
5×12=60 sessions.
Total Annual Sessions:
2
,
000
,
000
×
60
=
120
,
000
,
000
2,000,000×60=120,000,000 sessions.
Average Session Length:
From previous data, 1.2 liters of water is used per 5.5 pages of conversation.
Assume the average session produces 5.5 pages of conversation.
Water Usage per Session: 1.2 liters.
Total Annual Water Usage for ChatGPT
Total Water Usage:
Annual ChatGPT Sessions: 120,000,000 sessions.
Total Water Usage:
120
,
000
,
000
×
1.2
=
144
,
000
,
000
120,000,000×1.2=144,000,000 liters per year.
Car Wash Data Calculation
Total Registered Vehicles and as value of reference : 7 million in Los Angeles County. (regardless if the value is higher and/or not all vehicles are used and/or the size of the vehicle)
Car Wash Frequency and Water Usage
The following data are values as a reference to calculate.
*Automatic Car Washes: 45% (aprox.)
*Manual Car Washes: 35%
(aprox.)
*Home Car Washes: 20% (aprox.)
*people washig the car every 2 weeks average.
As a sensitive analysis the calculation of car-washing will be carry out compating with ChatGPT of 5, 10, 30 sessions per month (per person who uses Chat GPT)
Population of Los Angeles: 10 million. Percentage Using ChatGPT: 20% (regardless of whether the actual number is lower). ChatGPT User Population:
10
,
000
,
000
×
0.2
=
2
,
000
,
000
10,000,000×0.2=2,000,000
ChatGPT Water Usage with 5 Sessions per Month
Given:
Water Usage per Session: 1.2 liters
Sessions per Person per Month: 5
Sessions per Person per Year:
5
×
12
=
60
5×12=60
Total Annual Sessions:
2
,
000
,
000
×
60
=
120
,
000
,
000
sessions per year
Annual Water Usage for ChatGPT:
120
,
000
,
000
×
1.2
=
144
,
000
,
000
liters/year
ChatGPT Water Usage with 10 Sessions per Month
Given:
Water Usage per Session: 1.2 liters
Sessions per Person per Month:
10
Sessions per Person per Year:
10
×
12
=
120
10×12=120
Total Annual Sessions:
2
,
000
,
000
×
120
=
240
,
000
,
000
sessions per year
2,000,000×120=240,000,000 sessions per year
Annual Water Usage for ChatGPT:
240
,
000
,
000
×
1.2
=
288
,
000
,
000
liters/year
240,000,000×1.2=288,000,000 liters/year
ChatGPT Water Usage with 30 Sessions per Month
Given:
Water Usage per Session: 1.2 liters
Sessions per Person per Month: 30
Sessions per Person per Year:
30
×
12
=
360
30×12=36
Total Annual Sessions:
2
,
000
,
000
×
360
=
720
,
000
,
000
sessions per year
Annual Water Usage for ChatGPT:
720
,
000
,
000
×
1.2
=
864
,
000
,
000
liters/year
Comparison with Car Wash Water Usage
1. Calculating the Number of Cars in Los Angeles:
estimated the number of cars in Los Angeles to be around 7 million (reference value). However, not all registered vehicles are active. In most cases, the number of active vehicles is slightly lower than the total number of registered vehicles due to various reasons like vehicles being out of service, in storage, or not in regular use.
To account for this, let's assume that around 10-15% are registered vehicles in Los Angeles which are not active. Here's how that affects the calculation, choosing one value 15% (or 85% active):
85%
7,000,000×0.85=5,950,000 active vehicles
2. Determining the Frequency of Car Washes
Based on your updated information:
Frequency of Washing: Every 2 weeks
Total Washes per Year:
52
weeks
2
=
26
2
52 weeks
=26 washes per year per car
3. Breakdown of Car Wash Types
Using the new percentages:
Automatic Car Washes: 45%
Manual Car Washes: 35%
Home Car Washes: 20%
4. Calculate Water Usage per Type of Car Wash
Using standard estimates:
Automatic Car Wash: 40 liters per wash
Manual Car Wash: 200 liters per wash
Home Car Wash: 300 liters per wash
5. Total Annual Water Usage for Car Washes
Let's calculate the total water usage for each type of car wash and then sum them up:
Automatic Car Washes:
Number of Cars
×
Frequency
×
Percentage
×
Liters per Wash
=
5,950,000×0.45=2,677,500
Manual Car Washes:
Number of Cars
×
Frequency
×
Percentage
×
Liters per Wash
=
5
,
950
,
000
×
0.35
=
2
,
082
,
500
5,950,000×0.35=2,082,500
Home Car Washes:
Number of Cars
×
Frequency
×
Percentage
×
Liters per Wash
=
5,950,000×0.20=1,190,000
6. Annual Water Usage by Type:
Automatic Car Wash:
2,677,500×26×40=2,786,400,000 liters/year
Manual Car Wash:
2,082,500×26×200=10,819,000,000 liters/year
Home Car Wash:
1,190,000×26×300=9,282,000,000 liters/year
Total Annual Water Usage for Car Washes:
2
,
786
,
400
,
000
+
10
,
819
,
000
,
000
+
9
,
282
,
000
,
000
=
22,887,400,000 liters/year
Comparison ChatGPT vs Car Wash Water Usage
Summary of Comparisons:
ChatGPT
5 Sessions/Month: 144 million liters/year
10 Sessions/Month: 288 million liters/year
30 Sessions/Month: 864 million liters/year
Car Wash Water Usage (Adjusted for Active Vehicles):
Total: 22.887 billion liters/year
Conclusion:
After adjusting for the number of active vehicles, the total annual water usage for car washes in Los Angeles is 22.887 billion liters/year, which remains much higher than the water usage for ChatGPT under any session scenario. To equal the water usage of car washes in Los Angeles (22.887 billion liters/year), each of the 2 million (Assumed or reference value, regardless of whether the actual value is lower.) ChatGPT users (using 10 sessions value to compare) would need to engage in approximately 795 sessions per month. This is a significantly higher number than typical usage scenarios, reinforcing that even with frequent use, ChatGPT's water consumption is still considerably lower (relatively) than that of widespread activities like car washing.
The comparison between ChatGPT and car washing water usage is a technical one based on reference values. It doesn't favor or oppose either industry. Other sectors, including electricity and home water use, also involve water consumption and, in some cases, waste.
August/2024 version