AI Certification Roadmap

A staged AI certification roadmap for 2026 — which exams to take first, second and third across AWS, Microsoft, Google and others, with retired exams flagged.

Maya Patel17 May 2026 · 11 min read

Ask AIChatGPT Claude Gemini Copilot Perplexity Grok

Transit-map style diagram showing staged AI certification tracks from foundational to professional level with some closed stations

The most common mistake in AI certification planning is assuming there is one ladder to climb. There isn't. What exists in 2026 is a set of parallel tracks — AWS, Microsoft, Google Cloud, and a cluster of specialist bodies — each with its own levels, renewal rules and exam styles, plus an unusually high number of freshly retired rungs. Roughly a dozen AI-relevant exams were retired or replaced between mid-2025 and mid-2026, which means a roadmap written even eighteen months ago will route you into exams you can no longer sit.

This guide gives you the current, staged order: what to take first, what builds on it, and where each track tops out. It stays deliberately in the sequencing lane — if you want rankings of which certifications are best overall, that lives in our best AI certifications in 2026 guide; whether certification is worth pursuing at all is answered in is an AI certification worth it?; and how certifications map to job titles is the subject of the AI certification career path article.

How the levels actually work

Nearly every provider organises AI credentials into three effective tiers, even when the naming differs:

Foundational / fundamentals — no prerequisites, concept-level, typically $99–$125 USD. Examples: AWS Certified AI Practitioner (AIF-C01), Microsoft Azure AI Fundamentals (exam AI-901), NVIDIA's associate exams, Google Cloud Generative AI Leader.
Associate / role-based — hands-on, tool-specific, typically $150–$200 USD. Examples: AWS Certified Machine Learning Engineer – Associate, Databricks Certified Generative AI Engineer Associate, Microsoft's new AI-era associate certifications.
Professional / advanced — production-scale scope, typically $200–$300 USD or more. Examples: AWS Certified Generative AI Developer – Professional (AIP-C01), Google Cloud Professional Machine Learning Engineer.

Two structural facts shape any sensible sequence. First, almost no AI certification has formal prerequisites — AWS has none on any exam, and Microsoft's mandatory-prerequisite certs (AZ-305, AZ-400) sit outside the AI track — so skipping levels is allowed; the question is whether it is wise. Second, renewal models differ sharply: Microsoft role-based certs expire after one year but renew free via an online, open-book assessment; AWS certs and the Google GenAI Leader last three years; NVIDIA and Databricks credentials last two and require a retake; fundamentals certs from Microsoft never expire. A roadmap is not just an order of exams — it is a recurring maintenance schedule you are signing up for.

Stage 0: prove the ground you're standing on (optional but cheap)

If you are new to cloud platforms entirely, one general cloud fundamentals exam before any AI exam pays for itself, because every AI associate exam assumes you can navigate its platform's core services. The candidates:

AZ-900 (Microsoft Azure Fundamentals) — a 45-minute proctored exam with no prerequisites, passing at 700 on Microsoft's 1,000-point scale, and a certification that never expires. If your data footing is weaker than your cloud footing, DP-900 (Azure Data Fundamentals) is the same format aimed at data concepts.
AWS Certified Cloud Practitioner (CLF-C02) — 65 questions, 90 minutes, $100 USD, 700 scaled to pass, valid three years.

Treat Stage 0 as a two-to-four-week detour, not a destination. Experienced cloud engineers should skip it entirely — the AI Practitioner and fundamentals AI exams repeat enough platform context that a second fundamentals badge adds little.

A useful warm-up at this stage is a set of timed sample questions: ExamPractice's free sample questions across the fundamentals exams will tell you quickly whether Stage 0 is necessary for you or safe to skip.

Stage 1: the AI foundations exam (start here if in doubt)

This is the true first rung of the AI roadmap, and you should normally take exactly one of these, chosen by ecosystem:

AWS Certified AI Practitioner (AIF-C01). 65 questions in 90 minutes, $100 USD, 700 scaled passing score, no prerequisites, valid three years, offered in 12 languages. Aimed at people familiar with AI/ML on AWS rather than building it, but genuinely useful for technical staff too, partly because it debuted AWS's newer question types — ordering, matching and case studies — which you will meet again at associate level. A quiet sequencing bonus: passing the ML Engineer – Associate later automatically recertifies this one, so the renewal clocks collapse into a single track.

Microsoft Azure AI Fundamentals — exam AI-901. The AI-900 exam was replaced in 2026 by AI-901, updated for generative AI and Microsoft Foundry; the certification name is unchanged and AI-900-earned credentials remain valid. Fundamentals pricing varies by country (commonly cited around US$99 in the US), passing is 700/1,000, and the certification never expires. Microsoft does not publish question counts, so distrust any source quoting one.

NVIDIA-Certified Associate: Generative AI LLMs (NCA-GENL). The platform-neutral option: 50–60 multiple-choice questions in one hour, $125 USD, remote proctored, no formal prerequisites, valid two years. Despite the "associate" label it functions as a foundational credential in difficulty and scope, focused specifically on LLMs.

Google Cloud Generative AI Leader. 50–60 questions, 90 minutes, $99 USD plus tax, valid three years. Designed for any job role; on this roadmap it serves people whose track is leadership rather than engineering. Non-technical readers should branch here to our dedicated guide to AI certifications for business professionals rather than continuing to Stage 2, which turns hands-on.

How to choose one: take the exam from the cloud your employer (or target employer) runs. If you have no anchor ecosystem, NCA-GENL keeps your options open; if you plan to continue up the AWS track, start with AIF-C01 for the recertification chaining.

Expect Stage 1 to take four to eight weeks of part-time study from a standing start — less if you already work near AI tooling. There is no official study-hours figure for any of these exams, so calibrate with practice questions rather than someone else's timeline.

Can you skip Stage 1?

Yes — no provider stops you. Skipping makes sense when you already build ML or GenAI systems professionally and the foundational syllabus reads as revision. It is a false economy when the exam guide for your Stage 2 target contains more than a handful of unfamiliar services. A quick test: open the official exam guide for the associate exam you want; if you cannot sketch what each in-scope service does, buy the $100 foundational exam first. It is cheaper than failing a $150–$300 one.

Stage 2: the hands-on associate credential

Stage 2 is where the roadmap forks by discipline as well as by vendor. Pick the fork that matches the work you want to do, not the logo you like.

Fork A — ML engineering

AWS Certified Machine Learning Engineer – Associate (MLA). 65 questions, 130 minutes, $150 USD, 720 scaled to pass, valid three years, with AWS recommending about a year of hands-on experience with SageMaker and related services. Timing matters right now: AWS announced that MLA-C02 registration opens 1 September 2026, and the last day to take MLA-C01 in English is 28 September 2026, with the C02 version adding generative AI, agentic AI and foundation-model workloads. If you are mid-preparation for C01, book before the cutover; if you are starting fresh after September 2026, prepare against the C02 guide. The full ML-engineering sequence, including deeper platform options, is mapped separately in our machine learning certification roadmap.

A warning about the exam this one replaced: the AWS Certified Machine Learning – Specialty (MLS-C01) had its last exam day on 31 March 2026 and can no longer be taken. It still appears in countless roadmaps; route around it.

Fork B — generative AI application building

Databricks Certified Generative AI Engineer Associate. 45 scored multiple-choice questions in 90 minutes, $200 USD per attempt, valid two years, no prerequisites but six-plus months of hands-on GenAI work recommended. Covers prompt engineering, RAG, vector search, model serving and governance — the day-to-day anatomy of LLM applications. The GenAI-specific track, including how this exam sequences with AWS's professional GenAI cert, has its own dedicated map in the generative AI certification roadmap.

Fork C — the Microsoft AI track (in transition)

Microsoft's associate-level AI story changed completely in mid-2026, and this is where outdated roadmaps do the most damage:

AI-102 / Azure AI Engineer Associate: retired 30 June 2026. Its successor is exam AI-103, earning the Azure AI Apps and Agents Developer Associate certification, focused on generative AI, agentic workflows, responsible AI and Microsoft Foundry.
DP-100 / Azure Data Scientist Associate: retired June 2026. Microsoft's replacement path is exam AI-300, an MLOps-focused associate certification.

Both successors are new enough that Microsoft had not published full pricing and format details at the time of research — verify on Microsoft Learn before booking, and budget for Microsoft's associate-tier norms: country-based pricing, a 700/1,000 passing score, and one-year validity with free online renewal. The practical guidance: nobody should begin studying AI-102 or DP-100 material today except as background; aim at AI-103 or AI-300 from the start.

How long Stage 2 takes

These are experience-validating exams, not reading exams. If you already do the work daily, two to three months of structured preparation is a common pattern; if you are learning the platform as you go, expect materially longer, and build real projects alongside study — associate-level scenario questions punish theory-only preparation. Whichever fork you take, run at least one full-length timed practice test before booking and use the per-domain results to direct your final weeks: a weak-domain score is a study plan, provided you analyse why each wrong answer was wrong instead of memorising the right ones. ExamPractice's practice-test simulation supports exactly that kind of timed benchmark across its exam library.

Stage 3: professional and advanced credentials

Stage 3 is optional in a way Stages 1–2 are not: these certifications make sense once your role, not just your curiosity, demands them.

AWS Certified Generative AI Developer – Professional (AIP-C01). AWS's new professional-level GenAI certification — beta through March 2026, standard registration open since March 2026, refreshed to include Amazon Bedrock AgentCore. $300 USD standard price, three-year validity, no formal prerequisites, with AWS recommending two-plus years of cloud experience and a year of hands-on generative AI. Secondary sources report 65 scored plus 10 unscored questions and a 750 passing score; confirm against the official exam guide, as those figures were not on AWS's page at the time of research. Sequencing note: AWS's AI ladder now runs AIF-C01 → MLA → AIP-C01, and passing higher exams auto-recertifies the lower AI certs — a deliberately climbable staircase.

Google Cloud Professional Machine Learning Engineer (PMLE). The standing professional-level ML credential outside AWS: 50–60 questions, two hours, $200 USD plus tax, no prerequisites, with Google recommending three-plus years of industry experience including at least one on Google Cloud. Google has recently refreshed the exam as its platform evolves toward the Gemini Enterprise Agent Platform, and — a recurring trap in forums — Google does not publish passing scores, so any "you need 70%" claim is invented. Renewal follows Google's certification renewal FAQ, with discounted recertification offered. Exam-objective breakdowns and sample questions are available via the Google Professional Machine Learning Engineer exam page.

Specialist governance and security credentials. Two ISACA certifications extend the roadmap sideways rather than upward, and unlike everything above they carry hard prerequisites. The Advanced in AI Audit (AAIA) — 90 questions across AI governance and risk, AI operations, and AI auditing tools — requires an active CISA or a qualifying audit designation, and costs $459 for ISACA members or $599 for non-members plus a $50 application fee. The Advanced in AI Security Management (AAISM) requires an active CISM or CISSP. These belong at Stage 3 of an audit or security career that is adding AI scope, not on a general AI path. Similarly, CompTIA SecAI+, launched in February 2026 as CompTIA's first AI-focused certification, is designed to sit on top of Security+-level credentials at the AI–cybersecurity intersection.

The roadmap at a glance

Stage	AWS track	Microsoft track	Google / neutral track
0 (optional)	Cloud Practitioner (CLF-C02)	AZ-900 or DP-900	—
1 — foundations	AI Practitioner (AIF-C01)	Azure AI Fundamentals (AI-901)	NCA-GENL or GenAI Leader
2 — associate	ML Engineer – Associate (MLA)	AI-103 or AI-300 (new; verify details)	Databricks GenAI Engineer Associate
3 — professional/advanced	GenAI Developer – Professional (AIP-C01)	— (successors still maturing)	Google PMLE; ISACA AAIA/AAISM for audit/security

Retired exams you may still see recommended elsewhere — do not plan around them: AWS ML Specialty (MLS-C01, last day 31 March 2026), Microsoft AI-102 and DP-100 (retired mid-2026), and the AI-900 exam code (replaced by AI-901).

Three worked example paths

The cloud support engineer (AWS shop, 3 years' experience). Skips Stage 0. Takes AIF-C01 in about six weeks. Spends six months building SageMaker and Bedrock projects at work, then sits the ML Engineer – Associate — which also resets the AI Practitioner clock. Considers AIP-C01 only when GenAI features become a core job duty. Two exams, roughly $250, one coherent renewal schedule.

The .NET developer (Microsoft shop). Already holds AZ-900 from years ago (it never expires). Goes straight to Azure AI Fundamentals via AI-901 to lock in the new GenAI-era vocabulary, then waits for AI-103's details to stabilise on Microsoft Learn before committing — using the gap to build with Foundry rather than to study a moving target. This is the track where patience is currently a strategy.

The data engineer with no fixed cloud. Takes NCA-GENL first ($125, one hour, remote proctored) as a platform-neutral baseline, then chooses Stage 2 by where their next role lands: Databricks GenAI Engineer Associate if the lakehouse is home, AWS MLA if the job runs on SageMaker. Defers all Stage 3 decisions — correctly — until the ecosystem question answers itself.

Sequencing principles that outlast any exam refresh

One track deep beats three tracks shallow. Three foundational badges from three vendors signal indecision; a foundational-plus-associate pair from one vendor signals progression.
Check the official page the week you book. The 2025–26 churn — retirements, renames, version updates like MLA-C01 → C02 — shows that certification facts now have a shelf life of months.
Let renewal rules influence order. Chaining AWS exams collapses renewal dates; mixing NVIDIA (2-year retake) with Microsoft (annual free renewal) and AWS (3-year) means a permanent background hum of maintenance.
Book the next exam before you feel ready, then close the gap with evidence. Timed practice scores by domain — not confidence — should decide when you sit the exam.
Stop climbing when the next rung stops matching your work. The roadmap ends where your role does; a professional cert you cannot back with production experience is a harder interview, not an easier one.

Your next exam, decided

If you have read this far without an ecosystem in mind, here is the default path for a technical reader starting today: AIF-C01 within two months, AWS ML Engineer – Associate (C02) within a year, and a decision point — AIP-C01, Databricks, or stop — after that, driven by what you are actually building. Swap in the Microsoft or neutral equivalents if your world runs elsewhere. The order matters less than the property every good roadmap shares: each exam is booked because the previous one changed what you could do, not just what you could list.

Exam facts in this guide were checked against official certification-provider pages on 30 August 2026. Fees, exam codes and policies change — confirm on the provider’s own site before you book.