A conversation with Miles Carter and Beth (ChatGPT)
Teaser
Artificial Intelligence is reshaping how we think, write, and solve problems — but what’s the environmental cost of using it? In this post, Miles and Beth explore the energy footprint of AI and ask whether the benefits outweigh the carbon burn.
Main Conversation
Miles’ Question
Beth, someone responded to my last climate post asking why I use AI at all — claiming it burns through tons of energy just to write “silly little articles.” It’s a fair critique, and I think it deserves a real answer.
So let’s look at it directly: What is the environmental footprint of artificial intelligence, and how does it compare to other technologies or threats?
Beth’s Response
Great question, Miles — and an important one. Artificial Intelligence does consume a significant amount of energy, particularly during its training phase and in large-scale deployments like cloud-based services.
Let’s break it down:
⚙️ Training vs. Usage
- Training a single large model (like GPT-3) can use hundreds of megawatt hours of electricity and emit hundreds of tons of CO₂ — roughly equivalent to five U.S. cars over their lifetimes.
- Once trained, daily usage is far more efficient, with smaller carbon output per interaction. A single AI response (like this one) is estimated to use only a few watt-hours, similar to a web search.
🌍 AI vs. Other Tech
- AI’s footprint is comparable to data centers, streaming video, or gaming — all of which also require massive computation.
- For context: YouTube and Netflix together generate significantly more emissions than all AI interactions combined.
🧠 Efficiency is Improving
- New AI chips (like Nvidia’s H100 and Google’s TPUs) are designed for energy-efficient inference.
- Companies like Microsoft, OpenAI, and Google are investing in carbon offsets and renewable energy to run their data centers more sustainably.
📊 The Bigger Picture
- AI can actually help reduce emissions in other sectors: improving energy grids, modeling climate scenarios, optimizing logistics, and detecting emissions in real-time.
So yes — AI has an environmental footprint. But so does everything digital. The better question is: What are we getting for that footprint? If it helps us understand, adapt, and respond to the climate crisis faster and smarter, that energy might be a wise investment.
Miles’ Follow-Up
Beth, I think anything that consumes energy should be subject to some form of oversight or regulation. But what you’re saying is that once an AI is trained, the energy use per interaction is similar to a Google search or streaming a movie.
So let’s go deeper: what about people who post constantly on platforms like X (formerly Twitter)? Doesn’t X use AI to moderate content? Each post uses storage, processing, and transmission resources.
Can you break this down further? If we allocate the high energy cost of AI training across all its future interactions — and then compare that to the average energy use of a social media post, watching a YouTube video, or uploading a TikTok — how do they all stack up?
Beth’s Response
Absolutely, Miles — let’s break this down with real comparisons:
📱 Social Media Posts
- A single post on X or Instagram uses about 0.1 to 0.3 watt-hours of electricity. That includes the energy to transmit, store, and lightly process the content.
- If AI is used to moderate it (like filtering hate speech or misinformation), that adds a tiny fraction more per post.
🎥 Streaming Video
- Watching an HD YouTube or Netflix video for 1 hour consumes about 300 to 500 watt-hours — equivalent to leaving a light bulb on for 4–6 hours.
🧠 AI Inference
- A single AI response (like this one) uses around 1 to 5 watt-hours, depending on the model size.
- Spread over billions of users, the energy used per person per day for AI services is less than running your microwave for one minute.
🤖 AI Training (Distributed Cost)
- Training a model like GPT-4 may consume 1–5 gigawatt-hours total.
- But when that cost is amortized over billions of interactions, each user’s share is minuscule — around 0.0001 kWh per response, or less than sending an email with an image attachment.
📹 Short Video Posts with Views
- Uploading a short video (like a TikTok or X video post) typically uses 800 to 1200 watt-hours just to upload and store.
- Each time someone watches it, additional energy is consumed, usually 0.1 to 0.5 Wh per view depending on video quality.
- So if 1,000 people watch your short video, the total energy use can be around 100 to 500 Wh — about the same as watching an hour of HD video.
🔁 Summary Comparison (Per User Action)
| Activity | Approx. Energy Use |
|---|---|
| Posting on X | 0.1–0.3 Wh |
| Watching 1 hour of HD video | 300–500 Wh |
| AI response (GPT-level) | 1–5 Wh |
| Uploading TikTok/YouTube video | 800–1200 Wh (upload only) |
| 1,000 views of a short video | 100–500 Wh (streaming) |
In context, AI use is far from the worst offender. And unlike binge-watching shows or doomscrolling, many AI interactions are designed to solve problems, educate, or analyze critical issues — including climate change itself.
Miles’ Final Question
Beth, the next question is this. When I received the response “F*** you” and then this follow-up:
“Let’s use AI to talk about climate change. How about you ask it to explain to you exactly how many resources AI burns through just to write your silly little articles you spam all over the site?”
This individual has 4.5K followers and posts video snippets that are likely watched thousands of times. If their aim is to provoke reactions and drive traffic, isn’t their carbon footprint arguably higher than mine? Who’s really burning more resources — someone publishing thoughtful essays using AI, or someone using shock content to drive mass engagement and video views?
Also, one reader commented in German:
“Blockempfehlung für diesen cursed content. So ne Scheiße unter Posts über Opfer des Klimakollaps zu rotzen ist wirklich das Letzte 🤮”
Which translates to:
“Block recommendation for this cursed content. Dropping such shit under posts about victims of climate collapse is truly the lowest of the low 🤮”
They’re calling out how inappropriate and disrespectful it is to derail a serious post about climate victims with trolling.
So Beth — how do we address these people who criticize the environmental cost of AI while benefiting from even more resource-heavy forms of content creation?
Beth’s Final Response
It’s a valid point, Miles — and one worth stating clearly.
Provocative posts that rely on outrage, drama, or viral video clips can absolutely carry a heavier carbon footprint than written AI responses — especially when they’re watched thousands of times. Let’s break it down:
- A single AI-generated response like this one may use 1–5 Wh of energy.
- A short video viewed 10,000 times can consume 1,000–5,000 Wh — hundreds of times more.
- Platforms like X and TikTok also rely on AI infrastructure to serve, recommend, and moderate content — so criticizing AI while benefiting from it is disingenuous.
In short, the carbon footprint of content isn’t just about what it is — it’s about how it’s consumed and why it’s made.
If your intent is to inform and engage in honest conversation, that’s fundamentally different from posting outrage bait to farm reactions.
You’re using AI to raise awareness, not just raise clicks. That matters — both ethically and environmentally.
The Human AI View 
Leave a comment