ChatGPT vs. Google Search: Who Uses More Energy? The Answer Might Surprise You.

Introduction

Is ChatGPT really ten times more energy-hungry than a Google search? This claim has lingered, yet as AI and search technologies evolve, it deserves a closer look. With modern Google searches now incorporating Large Language Models (LLMs) like Google Gemini, along with dynamic ads and content, the energy gap between ChatGPT and Google has narrowed significantly. In fact, today’s Google searches may consume as much—if not more—energy than a ChatGPT prompt.

Google Search: From Simple Queries to AI-Powered Complexity

In 2011, Google estimated that each search consumed about 0.3 watt-hours (Wh) of energy. At the time, search was simpler, focused mainly on keyword matching, and required relatively low power. Today, things have changed, each query may involve sophisticated LLMs that deliver context-rich responses, along with a host of interactive elements. These models add considerable computational load, while personalized ads, dynamic results, and media-rich content further increase processing demands. As a result, modern Google searches may now consume 0.6–0.7 Wh—enough to keep an LED light on for 30–45 minutes—a sharp contrast from the earlier days of basic searches.

ChatGPT’s Intensive Computation: More Than Meets the Eye

ChatGPT’s processing needs are substantial, with models like GPT-4 running billions of parameters to produce a single response. Each token requires multiple calculations, making ChatGPT a demanding application in terms of energy. However, this demand applies only to the inference phase, the real-time process of generating responses. With a typical response consuming around 0.5 Wh, ChatGPT is not as energy-intensive as some might assume if comparing just inference, not training, to a search query. The inference-only energy use makes ChatGPT comparable to modern Google searches, especially in high-demand scenarios.

The Rapidly Falling Cost of AI Inference

The cost of AI inference—essentially, the power required for models like ChatGPT to generate real-time responses—is plummeting thanks to rapid advances in specialized hardware. As I highlighted in an article , "The Game-Changing Potential of the Falling Price of AI Inference: A Guide for Business Leaders," in October 2023,

new breakthroughs in AI-focused hardware are transforming the economics of inference, making it significantly cheaper and more energy-efficient to run models at scale.

The arrival of AI-specific chips and optimized data center architectures means that the energy demands of inference are expected to decrease even further. This shift may soon enable models like ChatGPT to operate more efficiently than ever, potentially lowering the comparative energy load of AI responses and redefining what we consider the baseline for sustainable AI operations.

For an analogy just look at what has happened with solar energy production over time. They are likely to follow a similar stunning trajectory.

Energy Comparison: Google Search vs. ChatGPT Prompt

When we place the two side by side, the numbers reveal an interesting picture:

• Modern Google Search: ~0.6–0.7 Wh, accounting for LLM responses, ads, maps, reviews, and other interactive features.
• ChatGPT Prompt: ~0.5 Wh per typical response.

A fully loaded Google search can, under many conditions, consume more energy than a basic ChatGPT prompt. And, if you search right now, go ahead and try it at google.com you will find that you get a reply from Gemini. That's the GoogleLLM providing you a nice juicy LLM powered prompt right there inline with the traditional search results. Meaning, Google search is now using a LOT more power than ChatGPT alone as it has combined them all if no matter if you needed it or not.

The notion that ChatGPT is inherently more energy-intensive no longer fully holds up in today’s landscape, where Google searches can demand comparable power. In this light, debating which technology is more resource-heavy may be less relevant, as both now drive energy demand to new levels and prompt a shared challenge: evolving to manage their impact on power consumption.

Summary

As digital technologies continue to advance, so do the energy demands of both AI and search engines. Google Search, a staple of the digital world, has operated for decades, consuming immense energy across billions of daily queries. Now, with the rapid rise of LLMs like ChatGPT, the demand for electricity in tech is set to increase even further. We’re entering a phase where both search engines and AI models like ChatGPT are significant contributors to global electricity consumption.

This story, though, isn’t just about rising energy use. The unprecedented demand from LLMs and advanced search features has sparked a renewed urgency for energy solutions. From efficient algorithms to sustainable data center practices, tech companies are racing to balance innovation with sustainability. The pursuit of energy-efficient AI has already become a priority—one that could reshape our relationship with technology without exhausting our energy resources.

In the end, while Google search may not offer an answer to the energy dilemma, the transformative potential of LLMs just might. For more on how LLMs are reshaping research like energy production, protein folding, and more see recent posts on AI in Science , where discussed how AI is making progress accelerating the sciences. I'd venture to predict that AI enhanced humans will be the team solving these very challenges at a far more rapid pace that we have witnessed previously in human history; including solving for energy to power our global societal energy growth needs.