Parts of a computer, and which ones run AI
A computer has seven main parts: a processor that follows instructions, a graphics chip for many small sums at once, memory for the work in progress, storage that keeps files, input, output and a network connection. AI runs mostly on the processor and graphics chip, either inside your own device or in a distant data center.
Last reviewed: 20 October 2026 7 min read
What are the main parts of a computer?
Every computer, from a phone to a laptop to a machine in a data center, has the same seven jobs to do. Each part does one of them.
| Part | Also called | What it does | When AI runs on your device | When AI runs in a data center |
|---|---|---|---|---|
| Processor | CPU, central processing unit | Follows the instructions in a program, one step after another. | Runs the app and handles the steps that do not suit the graphics chip. | Organizes the work and passes the heavy math to graphics chips. |
| Graphics chip | GPU, graphics processing unit | Does many small calculations at the same time. Built for drawing pictures on screen. | Does the multiply-and-add math of a small AI model. | Rows of graphics chips do the same math for large models, many at once. |
| Memory | RAM, working memory | Holds the program and data being used right now. Forgets everything when the power goes off. | Holds the AI model while it runs, so a model must fit in memory. | Holds parts of very large models, spread across many machines. |
| Storage | drive, SSD, flash, hard disk | Keeps files, apps and photos when the power is off. | Keeps the model file between uses. | Keeps the datasets and model files. |
| Input | camera, microphone, keyboard, touch screen | Brings information in. | Gives the AI the photo, sound or words to work on. | Your input is sent over the network first. |
| Output | screen, speaker | Shows or plays the result. | Shows the AI's answer. | The answer travels back over the network, then appears. |
| Network | Wi-Fi, mobile data, cable | Connects the computer to other computers. | Not needed for the AI step itself. | Carries your input there and the answer back. |
A computer also has a main circuit board that connects the parts and a power supply or battery that runs them.
How do the parts work together? Follow one photo
Here is one photo traveling through the parts, in an app that sorts pictures by what is in them.
- Input. The camera turns light into numbers: three numbers for each pixel, for red, green and blue.
- Storage. The photo is saved as a file, so it survives when the phone is switched off.
- Memory. When the app opens, the photo and the sorting model are copied from storage into memory, because memory is much faster.
- Processor. The processor runs the app's instructions: open the photo, shrink it, hand it to the model.
- Graphics chip. The model multiplies and adds millions of numbers to score the photo: "beach", "dog", "birthday cake". The graphics chip does many of these sums side by side.
- Output. The screen shows the photo in the "Dogs" album.
If the app sends the photo to a data center instead, two steps change. The network carries the photo away after step 2, a computer far away does steps 3 to 5, and the network brings the answer back before step 6.
Which part of a computer runs AI?
The graphics chip and the processor do the work, with memory holding the model while it runs.
Most AI models are built from huge numbers of multiplications and additions. A processor has a smaller number of powerful cores, each able to do almost anything. A graphics chip has many more, simpler processing units, and newer ones add circuits built for exactly this kind of matrix math. That is why the graphics chip does most of the heavy lifting.
Memory matters too. A model has to be loaded into memory to run, and memory is far faster than storage: main memory can move up to about 100 gigabytes a second, a fast flash drive about 1 to 3 gigabytes, and a hard disk about 100 to 200 megabytes. A model that does not fit in a device's memory cannot run well on that device.
What is the difference between memory and storage?
Memory (RAM) is the desk you work at; storage is the cupboard where things are kept. Memory is fast and holds only what is in use right now, and it is wiped when the power goes off. Storage is slower and keeps files for years, whether the power is on or off.
That is why a computer can have 16 gigabytes of memory and 1 terabyte of storage: the desk is small and the cupboard is big. For how big a gigabyte and a terabyte are, see From bytes to terabytes: how much data does AI use?.
What does "on-device" AI mean?
On-device AI means the model runs on the processor, graphics chip and memory of the device in your hand. The input, such as a photo or your voice, does not need to leave the device for the AI step.
Cloud AI, or AI in a data center, means your input travels over the network to a computer somewhere else. That computer runs the model and sends the answer back.
| On-device | In a data center | |
|---|---|---|
| Where the model runs | your phone, tablet or computer | computers in a building far away |
| Needs a network connection for the AI step? | no | yes |
| Does your input leave the device? | not for the AI step | yes, it is sent there |
| Model size | small enough to fit in the device's memory | can be much larger |
| Who can see the input | depends on the app | depends on the app and the data center's operator |
On-device does not automatically mean private. An app can still send other information over the network. The app's privacy policy says what it sends and keeps. For more, see What is data privacy?.
How much electricity do data centers use?
A lot, and the amount is rising. US data centers used about 4.4% of all US electricity in 2023, according to a US Department of Energy report. The same report links rising demand to the growth of data centers and of AI applications, alongside other causes.
So the choice between on-device and data center AI is not only about speed and privacy. It is also a question about energy: who uses it, who pays for it and where it comes from.
Going deeper, by age
Ages 8 to 10: name the jobs.
Point at a device and name one part for each job: something that brings information in, something that shows it, and something that keeps it. A camera is input; a screen is output; the photo gallery lives in storage.
Ages 11 to 13: count the sums.
A tiny model that scores a picture might multiply each pixel by a weight and add up the results. A 32 × 32 color picture has 3,072 numbers, so one score needs 3,072 multiplications and about as many additions. A graphics chip does thousands of these at the same moment. For what a weight is, see What is a neural network?.
Ages 14 to 18: follow the trade-offs.
Running a model on a device keeps the input local and works offline, but limits the model's size to the device's memory. Running it in a data center allows larger models, but sends the input away and depends on the network and on someone else's computers. Ask who chose where it runs, and who is affected by that choice.
What this page does and does not show
It shows:
- What each part of a computer does, and which parts do the work when AI runs.
- The difference between on-device AI and AI in a data center.
- Why memory size and the graphics chip matter for AI.
It does not show:
- How any particular phone, computer or app works. The photo-sorting app is an illustration written for this page.
- Every kind of chip. Some devices add other chips designed for AI math; the jobs are the same.
- How a model learns. That is a separate step from running it; see What is a model?.
Where does Mechna Academy fit?
Mechna Academy is a city of machines for curious children aged 8 to 18, opening first to ages 8 to 10.
Mechna does not collect student data. The app has no student accounts, and what a child does in the game stays on the device.
Questions
The processor follows instructions, the graphics chip does many small calculations at once, memory holds the work in progress and storage keeps files. Input parts such as a camera bring information in, output parts such as a screen show results, and the network connects the computer to others.
The graphics chip and the processor do the calculations, and memory holds the model while it runs. The graphics chip does most of the heavy math because it can do many multiplications and additions at the same time.
A CPU, the processor, has fewer but more powerful cores that can do almost any task in order. A GPU, the graphics chip, has many more, simpler processing units that do the same kind of sum on lots of numbers at once. That suits drawing pictures and running AI models.
Memory, or RAM, is fast and holds what the computer is using right now, and it is wiped when the power goes off. Storage is slower and keeps files, apps and photos whether the power is on or off.
It means the AI model runs on the phone, tablet or computer itself, so the input does not need to leave the device for the AI step. The other option is a data center, where the input is sent over the network to another computer and the answer comes back.
Mechna does not collect student data. The app has no student accounts, and what a child does in the game stays on the device.
Mechna Academy is a city of machines for curious children aged 8 to 18, opening first to ages 8 to 10.
For grown-ups: join the waitlistPlaces are limited. The waitlist is for parents and guardians 18 and older.