You've seen the LinkedIn posts: "Just cleared AI-900", a badge, 400 likes. Then you open a job posting for an AI role and it asks for a project and Python, not a certificate. So which is it? Are AI certifications worth it in India in 2026, or is it money you could spend on API credits and a decent laptop?
Short answer: a certification is worth it in a few specific situations, and mostly not otherwise. This post lists the certifications that exist (taken from each vendor's own page, checked on 2026-10-08), what each one actually tests, and a decision rule for when to pick the portfolio instead. I'm not giving salary figures. I didn't look for any on the vendor pages and I won't invent one.
Which AI certifications exist, and what do they cost?
Everything in this table comes from the vendor's certification or exam page, read on 2026-10-08. Prices are in US dollars as listed; tax and your country's pricing may change the real amount.
| Certification | Level | Price (as listed) | Length / questions | Valid for |
|---|---|---|---|---|
| AWS Certified AI Practitioner | Foundational | $100 | 90 min, 65 questions | 3 years |
| Microsoft Certified: Azure AI Fundamentals (exam AI-901) | Fundamentals | Varies by country; no figure shown | Not stated on the page I read | Not stated |
| Google Cloud Generative AI Leader | Business-level | $99 | 90 min, 50-60 multiple choice | 3 years |
| AWS Certified Machine Learning Engineer - Associate (MLA-C01) | Associate | $150 | 130 min, 65 questions | 3 years |
| Google Cloud Professional Machine Learning Engineer | Professional | $200 | 2 hours | Renewal rules on a separate FAQ |
| Databricks Generative AI Engineer Associate | Associate | $200 | 90 min, 45 scored questions | 2 years |
| NVIDIA Generative AI LLMs Associate | Associate | $125 | 1 hour, 50-60 multiple choice | 2 years |
Sources: AWS AI Practitioner, AWS ML Engineer Associate, Microsoft AI-901, Google Generative AI Leader, Google Professional ML Engineer, Databricks, NVIDIA.
Some caveats I'd rather say up front:
- AWS's AI Practitioner page doesn't print an exam code as a labelled field. The study links on it say AIF-C01, so that's what I'd search for, but check at booking.
- AWS's ML Engineer page showed two versions: the current MLA-C01 ($150, 65 questions) and a beta MLA-C02 at $75 beta pricing with 85 questions and 170 minutes, which adds Amazon Bedrock to the recommended experience. If you're about to book, read that page first, because the beta may be gone or promoted by the time you do.
- Microsoft's price depends on the country where the exam is proctored, and the page doesn't show India's number. Don't trust a blog that quotes one without a date.
- Google's ML Engineer page says "Languages: English, Japanese" and recommends "3+ years of industry experience". That's a recommendation, not a prerequisite, but it tells you who the exam is written for.
I couldn't open every vendor's India-specific pricing or the exam provider's booking pages in this check. Treat the dollar numbers as an indication.
What does each certification actually prove?
Be careful here, because the titles oversell. Going by what the vendor pages say they test:
Fundamentals tier (AWS AI Practitioner, Azure AI Fundamentals, Google Generative AI Leader). These prove you know the vocabulary and the vendor's product map. Microsoft's AI-901 page says the exam assesses "Identify AI concepts and capabilities" at 40-45% and "Implement AI solutions by using Microsoft Foundry" at 55-60%, and that candidates should know Python syntax and be familiar with Azure resources. So it isn't purely conceptual. Google's Generative AI Leader page says it's for "anyone in any job role, with or without hands-on technical experience" and describes business-level knowledge of Google Cloud's gen AI offerings. That is the clearest "this is not a builder's credential" statement of the lot.
Associate and professional tier (AWS ML Engineer Associate, Google Professional ML Engineer). AWS recommends at least one year of experience with SageMaker and related services. These are the ones where having built things on the platform helps in the exam. Pass one of these and a hiring manager at a company on that cloud can at least infer you've touched the tooling. It's still a multiple-choice exam, so it doesn't prove you can ship.
Vendor-neutral-ish (NVIDIA, Databricks). NVIDIA's Generative AI LLMs Associate lists topic weights: Core Machine Learning and AI Knowledge 30%, Software Development 24%, Experimentation 22%, Data Analysis 14%, Trustworthy AI 10%. Databricks recommends 6+ months of hands-on experience with the tasks in the exam guide. Both are tied to a vendor's ecosystem more than their names suggest.
What none of them prove: that you can write a retrieval pipeline that works on messy documents, debug a flaky agent, or evaluate output quality. Those are what interviews for AI engineering jobs go after, and that's what a project shows. If that's your goal, the interview questions on RAG, agents and evals are a better preview of the hiring bar than any syllabus.
When is a certification worth paying for?
My judgment, not data. Pay for one when at least one of these is true:
- Your target employer sells or runs on that cloud. Consultancies and service companies often care about partner-level credentials, because their contracts can depend on how many certified people they have. I haven't verified how any specific Indian company handles this, so ask a recruiter or a current employee at the company rather than assuming.
- You're switching into tech without a tech résumé. A badge helps a little when there is nothing else on the page. It matters less once you have a project.
- Your current employer reimburses it. If someone else pays and gives you study time, the maths changes completely. Take the one your team's cloud uses.
- You learn better with a deadline. A booked exam date makes some people actually finish the course. That's a fair reason, as long as you call it that.
- You're in a non-engineering role and want credible vocabulary. For a manager, marketer or founder who has to talk to engineers, the Google Generative AI Leader or AWS AI Practitioner syllabus is a reasonable structured overview. Honest limit: you can also get most of the vocabulary free from the vendor's training pages.
When does a portfolio beat a certificate?
When the job is "build things with LLMs", which is most of what AI engineer postings are. A reviewer opening a repository sees evidence a multiple-choice score can't give: how you structured the code, what you tested, what you chose not to build. If you only have time or money for one, spend it on a project.
A simple rule: if you can't yet build a working RAG app or a tool-using agent, a fundamentals cert won't cover that gap. Start with building a RAG chatbot over your own PDFs, then pick up portfolio projects that get interviews. A cert then adds something. Before that, it mostly adds a line on a résumé.
There's also an asymmetry in cost. A certification fee is a one-off outflow that expires on the vendor's schedule: two or three years, according to the pages above. A project stays in your GitHub, keeps being useful, and you can update it. The credential has to be renewed to keep its value. The repository doesn't.
What about the career context in India?
I've deliberately left out salary uplift and hiring-demand numbers, because I couldn't find a primary source for either, and a certificate-seller's blog doesn't count. If you want the career-path view, the site already has a roadmap for Indian developers moving into AI engineering and a look at what prompt engineering roles pay and where. Read those for the planning side, and keep the usual scepticism about any salary range you can't trace to a survey with a method.
A decision checklist before you pay
Work through these in order and stop at the first "no" that surprises you.
- Can I name the employer or job type that asks for this certificate? If not, find one posting that does before paying.
- Does my target employer use this vendor's cloud? Check three job postings for the platform name.
- Have I read the current exam page, not a blog? Exam versions move: Microsoft's Azure AI Fundamentals certification page now lists AI-901 as the required exam, so guides that still say AI-900 may be out of date.
- Is there a free or discounted route? Some vendors run training or voucher programmes. I didn't verify any, so look on the vendor's training page and your college's or employer's partnership list.
- Do I have a project I could finish in the same time? If yes, do that first.
- What's my plan for renewal? Two or three years goes quickly.
And one honest note on exam prep: AI can write you practice questions from the official exam guide, but it can also invent service features that don't exist. Check every answer against the vendor's documentation. The same discipline applies to using AI for exam study generally.
What I could not verify
- India-specific exam prices and any local discounts for all seven certifications.
- Microsoft's exam price, passing-score details for the others, and the number of questions on AI-901.
- Whether AI-900 has a published retirement date (the AI-901 page shows "Retirement date: none" for itself, and the certification page lists AI-901 as the required exam).
- Any employer preferences or salary effects. I treated these as unknown.
Everything else above was read directly off the vendor pages on 2026-10-08. Come back to those pages before you book, because prices and exam versions move.
