AI is very far away
You type a question on your phone, but the AI isn't inside your phone. It's very far away.
DATA CENTER
AI's building
Big AI models run in data centers: buildings full of server racks with GPUs, a special kind of chip. They use as much power as a whole neighborhood, and need constant cooling.
API
A door called an API
Apps send your question over the internet to that building, through a door called an API, and get the answer back. Most AI apps don't run their own models. They rent them through this door, as video 56 explained.
REGION
Which country your data is in
Which region the data center is in, meaning which country, matters too. Banks and hospitals often have to choose where their data lives because of regulations.
PEAK HOURS
Overloaded, try again later
At peak hours, millions of people ask at once. The system turns on more machines by itself, called autoscaling. If it can't keep up, you see the familiar message: overloaded, try again later.
ON-DEVICE
AI right on your phone
There are also small models that run right on your phone, called on-device AI. No internet needed, your data never leaves the device, but they're less smart. Many new phones use both: light tasks on the phone, hard ones sent to the cloud.
NEXT
The door needs a guard
With AI that far away, the door needs a guard. Part three: input guardrails.
This article is based on the video Cloud: where AI lives from the Mark học AI channel. Watch the video (in Vietnamese) to see the animations.