Ketan JoshiJun 10

The planned gas-fired capacity **JUST FOR DATA CENTRE ON-SITE USE** in the US is the same as *all* planned gas capacity in India, Africa, Latin America, Southeast Asia, Japan, Korea and China *combined*.

But sure, 'ChatGPT is just like 10 seconds of watching Netflix' 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠 🫠

https://www.iea.org/reports/world-energy-investment-2026

Show threadJorge Luis ⁂Jun 10
@ketan this is really alarming but Netflix is part of that pot. They run their services on this infrastructure.
Show threadpaelneverJun 10
@jorge @ketan
Nope please, don't spread misinformation. The service that any on-demand streaming platform like netflix provide for millions of people (usually 2~3 million clients) can be handled by a single CDN server. That server can stream up to 200Gb/s of video spending less that 400 W of electric power. Also those servers don't need big datacenters to work, they use to be placed at the telecommunication nodes of the internet service providers in order to have better connectivity.
Show threadJorge Luis ⁂Jun 10
@paelnever @ketan it was not my intention to compare them in terms of power demand - just wanted to mention that data centres ≠ chatbots. Netflix doesn't only stream media, they also run their own models on their user data.
Show threadpaelnever
@jorge
Ok but the scale of the operation is crucial here. I am very aware of the technical advances in AI and never heard before about netflix doing something in that field. Probably they don't build whole datacenters for it. You can do a lot of things at enterprise level with a setup of servers running in a single rack that drains about 20~30KW depending on how advanced are the chips inside the servers. I also run models and train LORAs with my own data on a device in my home that consumes 60 watts, no kidding.
Jun 10 at 1:58pmWeb