Azure AI-901 vs AI-103: Which Microsoft AI Exam to Take First

We earn a commission if you buy through links on this page. How we're funded.

AI-102 is retired. The replacement is Microsoft Certified: Azure AI Apps and Agents Developer Associate (AI-103).

Exam AI-900: Microsoft Azure AI Fundamentals is retired. The replacement is Exam AI-901: Microsoft Azure AI Fundamentals, which earns the same Azure AI Fundamentals certification but expects Python and familiarity with REST APIs and SDKs.

See Associate AI Engineer for Developers on DataCamp →

The table below compares 2 certifications on provider, level, realistic time, coding needed and best for.

Which should you take first — or should you skip the fundamentals exam?

Skip the fundamentals exam if you already write code for a living. Microsoft set no formal prerequisite between AI-900 and AI-102, and nothing on the AI-102 syllabus assumed you had passed the fundamentals paper first. A working developer who spends the fundamentals prep weeks inside the Azure AI services portal instead will walk into the associate exam — now AI-103 — better prepared and one exam fee lighter.

AI-900 was the one to take first — and possibly only — if you did not code. It was designed for business users, project managers and analysts in Microsoft-stack organisations, and for that audience it was a legitimate endpoint, not a stepping stone. Its successor, AI-901, expects basic Python, so it no longer serves that audience the same way; it still suits career changers who can write a little code and want a proctored credential before deciding whether to invest in the developer path. If you do not code, our guide to no-code AI certifications maps the rest of the territory.

What does each exam actually test?

AI-900 was a concepts exam. Its outline covered AI workload types, core machine-learning principles, and how Azure exposes vision, natural-language and generative-AI services — multiple-choice and its variants, no hands-on tasks. Its successor is not: AI-901 gives 40–45% to AI concepts and 55–60% to implementing AI solutions with Microsoft Foundry, including a lightweight chat client built with the Foundry SDK and a single agent.

The table below shows the fee and format of AI-901 and AI-103, as Microsoft publishes them. Every other exam we track is compared on our AI certification exams page.

AI-102 was a practitioner exam. It tested designing and implementing solutions with Azure AI services: calling and securing the APIs, building agents and search-backed applications, integrating vision and language services into working software, and deploying responsibly, with scenario questions that assumed real familiarity with the SDKs and the portal, in Python or C#. Its successor, AI-103, asks for experience developing apps in Python and centres on Microsoft Foundry.

AI-901 vs AI-103 side by side: what Microsoft's study guides say

Microsoft publishes a study guide for each exam with a skills outline and its weights, and the two outlines show how far apart the tiers sit. The table below quotes the AI-901 guide (skills measured as of 15 April 2026) and the AI-103 guide (skills measured as of 16 April 2026), both read on 16 October 2026. Fee and exam length are in the Microsoft table above.

The table below sets AI-901 against AI-103 on five questions: which retired exam each replaced, who Microsoft writes it for, how the marks are split, what score passes and whether the certification expires.

The outline answers the question this page is named for. AI-901 is one concepts domain and one build domain; AI-103 is five implementation domains, with planning and managing a solution and building generative and agentic solutions together carrying most of the marks. A developer who already ships Python against cloud APIs is studying AI-103's outline either way, so the fundamentals exam adds a fee and little else. Someone who has never deployed anything gets a real first step from AI-901's Foundry domain before attempting five.

How big was the jump from AI-900 to AI-102?

Bigger than the shared product family suggests. AI-900 rewarded a couple of weeks of structured reading; AI-102 rewarded months of actually building. The gap was not raw difficulty so much as kind: one was a knowledge check, the other a professional-skills exam, and clearing AI-900 comfortably said little about readiness for AI-102. The same logic applies to AI-901 and AI-103 — plan hands-on lab time for the associate exam, not just video hours.

Who should take the fundamentals exam and stop there?

AI-900 was the place to stop if you approved or managed AI work rather than built it. Azure AI Fundamentals can still be that endpoint, but its exam is now AI-901, which expects basic Python — check you are comfortable with that before booking it.

• Business, operations and project roles in Microsoft-stack organisations — the fundamentals stamp plus daily Copilot fluency covers the job.
• Managers who approve AI projects but never build them — pair it with a strategy course rather than the associate exam.
• Anyone collecting a first proctored credential for CV recency — AI-900 was among the easiest respected certifications available, though AI-901's build domain makes it a harder sit.

Who should skip the fundamentals exam for the associate tier?

Go straight to the associate tier, now AI-103, if you already write code in an Azure shop: the fundamentals exam adds little you do not already know.

• Working software developers in Azure shops — the fundamentals exam adds nothing a developer does not already know or can absorb in a weekend. Our software engineers guide covers the wider credential stack.
• Data engineers adding the AI serving layer to their pipelines — the deeper rationale is in our data engineers guide.
• Engineers with another cloud's associate-level credential — the concepts transfer; only the Azure specifics need study. Our AWS vs Azure vs Google comparison maps the equivalences.

If your stack question is AWS-versus-Azure rather than fundamentals-versus-associate, start with our AWS AI Practitioner vs Azure AI Fundamentals comparison instead — it settles the vendor question before this one.

What do they cost — and how does renewal work?

AI-900 was priced as a fundamentals exam, and Microsoft prices its successor AI-901 by the country where it is proctored; AI-102 cost more, as associate-level Microsoft exams do, and its successor AI-103 sits at that same tier. Renewal differs too: Microsoft fundamentals certifications do not expire, while associate certifications like AI-103 expire annually and renew through a free online assessment. Factor the renewal habit in — an expired associate cert quietly disappears from your verified profile.

How do you prepare — and can preparation be free?

Entirely free in both cases if you use Microsoft's own materials. Microsoft publishes a study guide for each of the current exams, AI-901 and AI-103, and a practice assessment for each on AI Skills Navigator (Microsoft lists AI-901's as free); the Microsoft Learn paths written for AI-900 and AI-102 are still useful background, but they follow the retired outlines. For either current exam, add real portal time: a personal Azure account on free-tier credits gives you enough room to build against the Foundry services both now test. The only unavoidable spend is the exam fee itself — and employers with Microsoft partner agreements often hold voucher pools, so ask before paying.

None of that needs a paid course. If you are a developer who wants structured practice before the associate tier, the one we would use is DataCamp’s Associate AI Engineer for Developers: about 29 hours of Python on the OpenAI API, embeddings, a vector database, LangChain and the Model Context Protocol. It is not AI-103 preparation — it never touches Azure or Foundry, the services AI-103’s study guide is built around — and it ends in a DataCamp certificate of completion, not a Microsoft certification.

Where the AI-900-first default goes wrong

The training industry sells ladders, and every rung is revenue. Cert-path content pushes everyone through fundamentals first because two exams sell more than one — not because the sequence serves every candidate. For developers, the AI-900-first default costs weeks and a fee to certify things they already know; the honest advice is to skip it, and you will rarely hear that from anyone selling prep courses.

The same scepticism applies in the other direction. Business users are increasingly told AI-102 is the 'real' certification and the fundamentals badge is somehow embarrassing. It is not. Most people in Microsoft-stack organisations will never build an AI service and do not need a practitioner exam to do their jobs well — for them, a fundamentals badge plus visible daily fluency is a complete, defensible answer. Take the rung that matches the work, and treat anyone who insists you need both as a salesperson until proven otherwise. The wider evidence on when credentials pay off is in our analysis of whether AI certifications are worth it.

Verdict

If you write code, skip the fundamentals exam and take AI-103, the credential Microsoft names as the replacement for AI-102. Plan from Microsoft's AI-103 study guide rather than an AI-102 plan: the side-by-side table above shows where its marks now sit, with generative AI and agentic solutions the largest domain. AI-900 is retired too: Azure AI Fundamentals now comes through AI-901, which expects basic Python, so if you do not code at all it is no longer the no-code endpoint it was — our no-code guide above covers the alternatives. In-betweeners — career changers and technical-adjacent analysts who can write a little Python — take AI-901 first as a low-cost confidence builder, then decide about the associate tier with real information. If you are still weighing Microsoft against the wider field, our 2026 ranking puts both exams in context, the AI certification roadmap sequences the whole journey, and our free Picker tool matches a certification to your background in about a minute.

Ready to start?

Frequently asked questions