How AI uses our drinking water - BBC World Service

Key Concepts:

• AI Water Consumption: The amount of water used by AI models like ChatGPT for processing and cooling.
• Data Centers: Facilities housing powerful computer chips that run AI calculations.
• Liquid Cooling: A method of cooling computer chips using water or other coolants.
• Water Neutrality: A state where a company balances its water usage by replenishing the same amount back to the environment.
• AI Electricity Demand: The increasing amount of electricity required to power AI data centers.

AI Water Consumption Estimates

• Sam Altman (OpenAI CEO) estimates a single ChatGPT interaction uses 1/15 of a teaspoon of water. He claims there are 1 billion messages sent to ChatGPT every day.
• Experts are skeptical of Altman's estimate, suggesting it might only apply to tiny models.
• One expert estimates 10-50 queries to a medium-sized large language model (like GPT-3) consumes roughly 500ml of water, including cooling and electricity generation.
• OpenAI declined to provide further details on Altman's estimate when asked by the BBC.

Why AI Uses Water

• AI prompts require complex calculations performed by specialized computer chips in data centers.
• Training AI models also requires intense work from these chips.
• The hardware generates significant heat, requiring cooling to prevent damage.

Liquid Cooling Process

• Coolant is piped over processing chips to absorb heat.
• The heated coolant is sent to a heat exchange unit, where water reduces its temperature.
• The coolant recirculates to cool the servers.
• The now-hot water is piped to cooling towers, where fans and water vapor dissipate the heat.
• Up to 80% of the water evaporates in this process.
• The remaining water is recirculated before being discharged back into a water source.

Environmental and Social Concerns

• Liquid cooling uses mostly drinking water, extracting it from circuits needed for irrigation, human consumption, and hygiene.
• Communities are protesting data centers due to stress on water sources and electricity grids in Spain, India, Chile, Uruguay, and parts of the US.
• Generating electricity for data centers also requires water, especially for power plants like coal, gas, and nuclear.
• The International Energy Agency projects a 400% increase in electricity demand for AI-optimized data centers by 2030, equivalent to the UK's annual electricity consumption.
• Water is also needed for manufacturing semiconductor chips used in AI.
• Water is used for the refination of the critical raw materials that are needed to create the hardware of AI.

Data Transparency and Corporate Pledges

• Google, Meta, and Microsoft release annual figures on data center water usage (billions of liters), but don't specify how much is due to AI.
• Many tech giants, including Google, Microsoft, and Meta, have pledged to be water neutral by 2030.

Potential Solutions and Innovations

• Companies are trialing ways to cool data centers without evaporating water.
• Using generated heat to warm homes.
• Exploring moving data centers under the sea, to the Arctic, or even into space (satellites for backup tasks).

Notable Quotes:

• Sam Altman (OpenAI CEO): "1/15 of a teaspoon" (water usage per ChatGPT interaction).
• Expert: "At this point, there was just not enough information for me to agree with or trust the number."
• Expert: "Basically, if you write an email or ask some questions, if you have 10 to 50 queries, you're going to be consuming roughly 500ml of water."

Technical Terms:

• Large Language Model (LLM): A type of AI model trained on vast amounts of text data.
• Data Center: A facility housing computer systems and associated components.
• Coolant: A fluid used to absorb and transfer heat.
• Heat Exchange Unit: A device used to transfer heat between two fluids.
• Terawatt Hours: A unit of energy equal to one trillion watt-hours.
• Semiconductor Chips: Integrated circuits used in electronic devices.

Synthesis/Conclusion:

The AI revolution has a significant water footprint, stemming from the cooling of data centers and the electricity required to power them, as well as the manufacturing of the hardware. While precise figures are difficult to obtain, the issue is gaining attention, with tech companies pledging water neutrality and exploring innovative cooling solutions. The industry is still young, and collaborative efforts are needed to minimize water and energy consumption for a more sustainable future.