What is a GPU, and why does AI need it? Explained for students
CPU vs GPU in plain words, why AI runs on GPUs, which subjects in your syllabus lead there, and how to try a GPU free from your phone or laptop.
On 17 November 2026, Nvidia said it earned $115.8 billion in one quarter, and about $107.6 billion of that came from its data-centre business: mostly the chips that train and run AI. That's why everyone is suddenly talking about GPUs.
This article isn't about shares or money. It's about the technology: what a GPU is, why AI needs it, and how a student can start learning it this week.
CPU vs GPU in one picture
Imagine 500 answer sheets that need correcting.
- A CPU is one brilliant professor. Very fast, can handle any kind of question, but corrects one sheet at a time.
- A GPU is 500 teachers in a hall. Each one is simpler, but they all correct at the same time. The pile is done in minutes.
| CPU | GPU | |
|---|---|---|
| Cores | A few (4 to 32 in most laptops) | Thousands of smaller ones |
| Good at | Many different jobs, one after another | The same simple job on lots of data, all at once |
| Inside your laptop | Runs the OS, apps, browser | Draws the screen, games, video |
GPU stands for graphics processing unit. It was built to colour millions of screen pixels at once for games. Then researchers noticed that the same trick works for maths.
Why AI runs on GPUs
An AI model is mostly multiplying huge tables of numbers (matrices). Training a large model means doing that trillions of times. Each multiplication is simple, and most can happen at the same time: exactly what those "500 teachers" are good at.
There are two jobs:
- Training: teaching the model from data. Needs thousands of GPUs for weeks.
- Inference: using the trained model, like when you ask Gemini or Claude a question. Every answer you get runs on GPUs (or similar AI chips) in a data centre somewhere.
Nvidia isn't the only maker. AMD and Intel make AI chips too, and Google designs its own chips called TPUs. The idea is the same: lots of maths, in parallel.
The doors in your syllabus
You've probably met this topic already, under exam names. Here's how they translate:
| Your subject | What it means in the GPU world |
|---|---|
| Computer Organisation & Architecture | How cores, cache and memory make a chip fast |
| Linear Algebra / Engineering Maths (matrices) | The actual maths every AI model runs |
| Operating Systems, Parallel / Distributed Computing | Splitting one big job across many workers |
| VLSI Design, Digital Electronics (ECE) | Designing the chips themselves |
| Computer Graphics | Where GPUs started |
Search any of these with the words "GPU jobs" or "CUDA jobs" and you'll find roles like ML engineer, GPU performance engineer, CUDA developer, data-centre technician and chip design engineer. Our Syllabus → Jobs pages do this translation topic by topic.
Try a GPU free, tonight
You don't need an expensive gaming laptop. Free cloud notebooks like Google Colab and Kaggle let you use a GPU from any browser, with usage limits (check their current rules).
- Open Google Colab and create a new notebook.
- Go to Runtime → Change runtime type and choose a GPU.
- Paste this and run it twice (the first run warms up the GPU):
import torch, time
a = torch.rand(4000, 4000)
t = time.time(); a @ a; print("CPU:", round(time.time() - t, 3), "s")
g = a.cuda(); torch.cuda.synchronize()
t = time.time(); g @ g; torch.cuda.synchronize(); print("GPU:", round(time.time() - t, 3), "s")
It multiplies two big matrices, first on the CPU, then on the GPU. Compare the times. That gap is the whole story of this article, on your own screen.
What to do next
- Week 1: run the code above and change the size (2000, 8000). Ask an AI tutor to explain each line.
- Week 2: follow a free beginner tutorial that trains a tiny image classifier on a GPU.
- Week 3: write a short blog post or LinkedIn post about what you measured. That's proof of learning that companies can see.
Want a 30-day plan for your laptop? Get the free Builder Guide.
Same Laptop, Different Future
My story, what young builders did with the same tool, a 30-day plan for your laptop and an honest laptop guide. Free.
Free. One email a week at most. Unsubscribe any time. Privacy


