Research: ChatGPT consumes up to an estimated 500ml of water for every five to 50 prompts; Microsoft reported its water use spiked 34% YoY in 2022, Google 20%
The cost of building an artificial intelligence product like ChatGPT can be hard to measure. But one thing Microsoft-backed OpenAI needed …
Context & Ripple Effects
This coverage establishes water consumption as a measurable, though highly variable, input to generative-AI infrastructure rather than an abstract environmental externality. Microsoft and Google had already disclosed year-over-year increases in water use, while later work found that a GPT-4 task's estimate differed sharply by data-center location, underscoring how local operating conditions can change the water footprint.
The issue gains importance as usage scales: OpenAI later reported more than 200M weekly ChatGPT users, making per-prompt resource estimates consequential even when their ranges are uncertain.
First-order effects
- The research puts water alongside power and compute as an operating input that Microsoft-backed OpenAI and other model providers must account for when running ChatGPT-scale services.
- Microsoft's 34% and Google's 20% reported increases make their data-center water disclosures a clearer point of comparison for customers, communities, and investors.
Second-order effects
- Providers and cloud customers have a stronger incentive to measure workload-level water use and to weigh data-center location when allocating AI inference workloads; the later location-specific GPT-4 estimates illustrate why a single per-query figure is insufficient.
- As AI usage rises, water availability can become a more material constraint in infrastructure siting and capacity planning, alongside the compute and electricity requirements already associated with AI expansion.
Third-order effects
- If comparable reporting becomes more common, AI infrastructure competition may increasingly hinge on the ability to deliver compute with lower local water intensity, not just model capability and chip supply.
- The broader shift is toward treating AI data centers as utility-scale infrastructure whose resource impacts vary by region and must be managed at the workload level.
The trend: Generative AI is turning inference into a resource-management problem in which water, power, and location increasingly shape the economics of scale.