Azure fundamentals: AI-900 vs AZ-900 vs DP-900, which one first?

Microsoft sells three Azure fundamentals exams: AZ-900 (Azure), AI-900 (AI), and DP-900 (Data). They're each $99 USD, each multiple choice, each about 60 minutes, and each marketed as the entry point into a different Microsoft track. Most people I know who take one end up taking at least two. A handful chase all three for the badge collage on LinkedIn.

So which one first, and is the rest of the set actually worth it? The short answer: AZ-900 first, because it's the foundation the other two assume you already have. Whether you stop there or keep going depends on what kind of work you do.

The three at a glance

Exam	Focus	Length	Best for	Difficulty
AZ-900	Core Azure services, pricing, governance	60 min	Anyone touching Azure	Easiest
AI-900	Azure AI services + ML concepts	60 min	Devs / analysts using Azure AI	Easy-medium
DP-900	Azure data services (SQL, Synapse, Cosmos)	60 min	Data analysts, BI, junior data engineers	Medium

All three are $99 in the US, all three follow the same Pearson VUE / online-proctored format, and all three are valid for life — Microsoft doesn't expire the fundamentals (it does expire the role-based associates and experts). That's a small but real thing. AI-900 you pass in 2026 is still a credential in 2030.

AZ-900: the entry point

AZ-900 covers cloud concepts, core Azure architecture (regions, availability zones, resource groups, ARM), the main service families (Compute, Networking, Storage, Identity, Databases), governance and compliance (Azure Policy, RBAC, locks), and pricing/SLAs.

It's the easiest of the three because the questions are conceptual and the vocabulary repeats across every other Microsoft exam. If you've used any cloud platform before, AZ-900 is roughly a one-week prep at 5–8 hours total. If you've never touched cloud, give it 3–4 weeks at the same pace.

Who AZ-900 is actually for:

• Anyone in a non-engineering role at a Microsoft shop — PMs, sales engineers, IT managers — who needs to be able to read an architecture diagram without bluffing.
• Career-switchers proving baseline cloud literacy before applying for cloud-engineering jobs.
• Engineers who learned AWS first and need to translate the vocabulary fast.

If you only take one of the three, take AZ-900. The other two assume you already know what a resource group is.

AI-900: the easy AI badge

AI-900 covers AI workload types (computer vision, NLP, document intelligence, generative AI), responsible AI principles, and Azure-specific services: Azure AI Vision, Azure AI Language, Azure OpenAI Service, Azure AI Search, Azure Machine Learning at the studio level. As of the late-2024 refresh it leans heavily on generative AI — Azure OpenAI, prompt design, RAG concepts.

The exam doesn't require you to write code. It doesn't require you to know what a transformer is, or how an LLM is trained, or what loss function a classifier uses. It's a service-catalog exam wrapped in AI vocabulary. If you've shipped a Copilot integration or used Azure AI Studio for a weekend, you can probably pass it cold.

Who AI-900 is actually for:

• Developers about to ship something on Azure OpenAI who want a structured tour of the service surface.
• Business analysts and PMs who need to evaluate AI vendors and don't want to be hoodwinked by sales decks.
• Junior data folks who want a friendlier first cert than DP-900.

Who it isn't for: people expecting it to teach them ML. It won't. AI-900 is to ML what AZ-900 is to distributed systems — orientation, not depth.

Time investment: 1–3 weeks at 5–7 hours per week, depending on how much Azure AI exposure you already have.

DP-900: the trickiest of the three

DP-900 is the data fundamentals exam. It covers relational concepts (SQL, normalization, indexes), non-relational concepts (Cosmos DB APIs, document/graph/key-value), analytics workloads (Synapse, Fabric, Power BI), and the operational shape of each on Azure.

It's the hardest of the three for two reasons. First, the surface area is wider — you need to know enough SQL to read a query, enough Cosmos to know which API maps to which data shape, and enough Synapse / Fabric to talk through a modern data platform. Second, Microsoft refreshed the exam in late 2024 to lean harder on Microsoft Fabric, which has its own vocabulary (lakehouses, OneLake, semantic models) that nobody had heard of two years ago.

Who DP-900 is actually for:

• Data analysts moving from on-prem SQL Server / Power BI to Azure.
• BI developers proving foundational data-platform literacy.
• Junior data engineers en route to DP-203 or DP-700 (the Fabric Analytics Engineer Associate, which has been Microsoft's hot 2025 cert).
• Software engineers who keep getting pulled into data-platform conversations and want the vocabulary.

Time investment: 3–5 weeks at 5–8 hours per week if you have SQL background, 6–8 weeks if you don't.

Difficulty ranked

Honest difficulty rank, easiest first:

1. AZ-900 — broad but shallow, mostly vocabulary.
2. AI-900 — narrower, mostly service identification, low cognitive load.
3. DP-900 — wider technical surface, more Fabric vocabulary that's still settling.

Pass rates aren't published but Microsoft Learn community polls put them all in the 75–85% first-attempt range, which is high for technical exams. The fundamentals are designed to be passable — Microsoft wants people in the funnel.

Should you take all three?

I'd push back on the all-three-for-the-badge approach. The fundamentals share enough vocabulary that the marginal value of the second and third drops fast. A single AZ-900 badge plus a role-based associate (AZ-104, AZ-204, DP-100, AI-102) signals more than three fundamentals.

That said, here are the cases where collecting more than one makes sense:

• AZ-900 + DP-900 if you're heading toward DP-203, DP-300, or DP-700. The data exams build on both.
• AZ-900 + AI-900 if you're a developer about to specialize on Azure OpenAI and want the conceptual scaffolding before AI-102.
• All three if your employer reimburses certs and you genuinely want a tour of Azure before picking a track. Three fundamentals plus all the studying is roughly 10–15 weeks of evening time and $297 in fees. Cheap for the breadth.

Don't take more than one of the three if you've already passed any role-based Azure cert. AZ-104 covers everything in AZ-900 plus a lot more; DP-100 swallows DP-900 whole; AI-102 makes AI-900 redundant. Going backward to collect the fundamentals after you've passed an associate is a vanity move.

Pick one and start

If you're brand new to Azure: AZ-900, then decide based on the work you actually do whether AI-900 or DP-900 is next.

If you're a data person: skip AZ-900, take DP-900, and be honest with yourself about whether DP-100, DP-203, or DP-700 is your real target.

If you're an AI-curious developer: AI-900 standalone is fine. You can backfill AZ-900 later if a job posting asks for it.

When you're ready to grind questions, the AZ-900 question bank is the right place to start, and the AI-900 and DP-900 banks are right there if you decide to keep going. Just don't stack three fundamentals as a substitute for one associate — that's the most common mistake I see on cert plans.