A second-year computer science student I know paid £340 for an online “AI bootcamp” last spring. She completed four of the eleven modules. The free course she started in June, Stanford’s CS229 lectures watched at 1.5x speed between shifts, she actually finished.
That pattern is common. Free AI courses for students aren’t scarce. There are hundreds of them, and some are better than the paid ones. The bottleneck isn’t access, it’s choosing badly and then quitting quietly in week three.
What “free” actually means on AI course listings
Before you sign up for anything, work out which kind of free you’re getting. Coursera and edX both let you audit most courses at no cost, which means full video lectures and readings but no graded assignments or certificate. Certificates usually cost between $49 and $99. Coursera’s financial aid form takes about fifteen minutes and gets approved for a lot of students, so it’s worth trying.
Other platforms are genuinely free end to end. Microsoft Learn, Google’s Machine Learning Crash Course, Kaggle Learn and DeepLearning.AI’s short courses don’t charge you at any point. GitHub-hosted curricula such as Microsoft’s AI for Beginners are free forever because they’re open source repositories.
One warning: a course being free doesn’t mean your time is free. A 40-hour course you abandon is worse value than a 6-hour course you complete and use in a project.
Six free AI courses students actually finish
This list is deliberately short. All of these have a defined end point, which matters more than production values.
- CS50’s Introduction to AI with Python (edX) — Harvard’s follow-on to CS50. Seven problem sets covering search algorithms, constraint satisfaction, optimisation and neural networks. Free to audit, roughly 12 hours of lectures.
- Google’s Machine Learning Crash Course — about 15 hours across 25 lessons, built around TensorFlow. The interactive widgets and the real-world framing make it the best starting point if you’ve never touched ML.
- Microsoft’s AI for Beginners — 24 lessons over 12 weeks on GitHub, with runnable notebooks and quizzes. Good if you prefer reading and experimenting over watching video.
- DeepLearning.AI short courses — one to two hours each on specific tools like prompt engineering or retrieval-augmented generation. Ideal for a single Sunday afternoon.
- Kaggle Learn — micro-courses of four to five hours, free certificates, and a dataset community attached. Start with Intro to Machine Learning.
- MIT 6.S191 — MIT’s fast-paced deep learning bootcamp, released publicly each year. If you want the full story, there’s a detailed breakdown of what MIT 6.S191 covers and who it suits.
Pick one. Not two. Adding a second course in week one is the single most reliable way to finish neither.
Student credits most people never claim
Training is free, but compute isn’t, and compute is where AI projects quietly die. Use your student status.
The GitHub Student Developer Pack gets you a bundle of tools plus cloud credits. Azure for Students gives you $100 of credit with no card required. Google Cloud and AWS both run student programmes with starter credits. Google Colab’s free tier includes a T4 GPU with usage limits, which is enough to fine-tune a small model or run a handful of training experiments per week.
If a tutorial asks you to rent a GPU instance for eight hours, look for a smaller dataset first. Most student projects fail on scope, not on hardware.
The bit that separates a certificate from a skill
Nobody has ever been hired because of a Kaggle completion badge. They get hired because they can describe a problem they solved and defend the choices they made.
So build one small thing with every course you take. If you’re doing the Google crash course, train a model on a dataset you care about: campus bus delays, local air quality readings, your own Spotify history. Aim for something you can explain in three sentences, with one number attached, like “my model predicted rent prices with a mean error of £62 a month.”
If you want help telling a strong free course from a weak one before you commit, this guide to finding free AI courses and actually finishing them lays out the signals to check in the first twenty minutes.
How to study so the material sticks
Four hours of passive video a week produces almost nothing. Three hours with retrieval practice produces a lot.
After each lecture, close the tab and write down what the model does, what it’s bad at, and one question you still can’t answer. Then answer that question before the next session. Space your reviews: revisit day one’s notes on day three and day ten. It feels slower and it isn’t.
Using a language model as a tutor works well here, provided you make it quiz you rather than explain to you. There’s a practical walkthrough of using AI to study smarter without outsourcing your thinking that’s worth twenty minutes before your next revision session.
Pair the AI material with data skills
Almost every AI-adjacent job ad mentions SQL. It’s unglamorous and it’s the fastest way to become useful on a team. Learn joins, window functions and aggregation, then practise on a real dataset.
Snowflake runs a free training platform that covers exactly this, and it’s more practical than most university data modules. This walkthrough of Snowflake University’s free training tracks explains what’s inside and how long each level takes. Two weekends will get you to a point where you can pull and clean your own data, which makes every AI project afterwards easier.
An eight-week plan that fits around a degree
This assumes about five hours a week, which is realistic during term time.
- Weeks 1–3: Google’s Machine Learning Crash Course. Two hours of video, three hours of notebooks per week.
- Week 4: Pick your project dataset. Spend the whole week on cleaning and exploring it. Expect this to be boring and take longer than you think.
- Weeks 5–6: Train a baseline model, then one improvement. Write a short README with your results.
- Week 7: CS50 AI’s first two problem sets, to fill gaps in search and optimisation.
- Week 8: Write up the project properly and publish it. Then choose your next course.
Publish something every two weeks, even if it’s imperfect. A public commit history is a better signal of ability than a stack of certificates nobody can verify.
Where free AI courses actually lead
They lead to three places, roughly in order of likelihood. A better grade on a project you’d have struggled with otherwise. A portfolio piece that gets you a first internship conversation. Or, occasionally, a genuine appetite for the maths that pushes you toward a formal programme, in which case it’s worth reading up on how to choose an AI degree with real cost and salary numbers before you apply anywhere.
What free courses can’t give you is a deadline, a tutor who notices you’re stuck, or a peer group that makes showing up easy. Replace those deliberately. Join a study Discord, tell a friend what you’re working on, or book a weekly slot in your calendar where you show someone your progress. The content is free and abundant. The accountability is the part you have to build yourself, and it’s the reason the student I mentioned finished CS229 on a laptop with a cracked screen while her paid bootcamp gathered digital dust.