Machine Learning Certification Roadmap

Stage 1: data and platform fluency

ML engineering exams assume you can already operate a cloud platform and move data through it competently. Stage 1 is where self-taught modellers most often discover their gap: they can train a model in a notebook but cannot design the pipeline that feeds it or the infrastructure that serves it.

You have two honest options here.

Option A: certify the data layer

AWS Certified Data Engineer – Associate (DEA-C01) — 65 questions, 130 minutes, $150 USD, passing score 720 scaled, valid three years. AWS suggests two to three years of data engineering experience and one to two years hands-on with AWS. For an ML-bound candidate this is not a detour: feature pipelines, data quality and orchestration are a large share of real ML engineering work, and the overlap with the ML Engineer Associate syllabus is substantial. On Google's side, the Professional Data Engineer (40–50 questions, 2 hours, $200) plays the same role at professional level.

Option B: certify general platform operations

Google Associate Cloud Engineer — 50–60 questions, 2 hours, $125, valid three years, with six or more months of hands-on Google Cloud experience recommended. If you are targeting the Google Professional Machine Learning Engineer at Stage 2, this is the natural warm-up: it forces fluency with the console, IAM, networking and deployment mechanics that the professional exam takes for granted.

When to skip Stage 1 entirely: you already work daily on the platform you intend to certify on. A developer with a year of SageMaker experience gains little from DEA-C01 before MLA; a data scientist already deploying on Vertex-era Google Cloud tooling can go straight at the professional ML exam. Certifications should close gaps, not decorate strengths.

Stage 2: the core ML engineering credential

This is the stage that defines the roadmap, and in 2026 there are three serious anchor options. All three are proctored exams from major platform vendors, none has formal prerequisites, and each maps to a different working context.

AWS Certified Machine Learning Engineer – Associate (MLA)

The closest successor to the retired ML Specialty. Current version MLA-C01: 65 questions, 130 minutes, $150 USD, passing score 720 scaled, valid three years, with roughly a year of ML engineering on AWS (SageMaker and related services) recommended. It uses the newer ordering, matching and case-study question types alongside multiple choice.

Timing note for late 2026: AWS has announced the MLA-C02 update. Registration for the C02 beta opens 1 September 2026, and the last day to take MLA-C01 in English is 28 September 2026 (Korean, Japanese and Simplified Chinese versions of C01 continue until C02 reaches those languages). MLA-C02 adds generative AI, agentic AI and foundation-model workloads. If you are exam-ready now, book C01 before its English cut-off; if you are months away, prepare against the C02 exam guide from the start rather than studying for a retiring version.

Google Cloud Professional Machine Learning Engineer (PMLE)

50–60 questions, 2 hours, $200 plus tax, delivered online-proctored or at a test centre. Google recommends three or more years of industry experience including at least one on Google Cloud, publishes no passing score (results are pass/fail only), and — unusually — states the exam does not directly assess coding, while recommending minimum Python and SQL proficiency to read code snippets. Professional-level Google certifications are valid for two years, with renewal handled per Google's renewal FAQ. The 2026 refresh matters: content has shifted towards the Gemini Enterprise Agent Platform, so prep material built around older Vertex AI-only framing is now partially stale. Working through Professional Machine Learning Engineer practice questions against the current exam guide is a sensible way to find out which of your study sources have kept up.

Databricks Certified Machine Learning Associate → Professional

If your employer runs Databricks, the platform's own two-level ML track is the highest-relevance path available, and it is the only anchor option with a built-in associate-to-professional progression on the same platform. Databricks certifications are valid for two years, with recertification by retaking the then-current exam. Exam-specific details vary by credential, so check the current guides — and note that ExamPractice hosts pages for both the Databricks Machine Learning Associate and the Machine Learning Professional exams.

Choosing your anchor: a comparison

Factor	AWS ML Engineer – Associate	Google Professional ML Engineer	Databricks ML Associate/Professional
Level	Associate	Professional	Associate, then Professional
Format	65 questions, 130 min	50–60 questions, 2 hours	Varies by exam — check current guide
Cost	$150 USD (regional variation)	$200 USD plus tax	Check Databricks certification pages
Passing score	720 scaled (100–1,000)	Not published (pass/fail)	Not published
Recommended experience	~1 year ML engineering on AWS	3+ years industry, 1+ on Google Cloud	Hands-on Databricks ML experience
Validity	3 years	2 years	2 years
Best for	Engineers in AWS shops; broadest job-market coverage	Experienced practitioners in Google Cloud environments	Teams standardised on Databricks / Mosaic AI
Renewal path	Retake or higher AWS exam; also auto-recertifies AIF-C01	Renewal within Google's eligibility window	Retake current exam

No universal winner exists here, and picking against your employer's stack is the roadmap's most common self-inflicted wound. Certify the platform you can practise on daily; the concepts transfer, but the exam scenarios are platform-specific.

Stage 3: specialise upward

With an anchor credential earned, three directions make sense — and which one depends on where your role is drifting.

Towards generative AI engineering: AWS Certified Generative AI Developer – Professional (AIP-C01). New in 2026, professional level, $300 USD, valid three years, covering foundation-model integration, retrieval-augmented generation, vector databases and production generative AI on AWS, with content refreshed to include Amazon Bedrock AgentCore. AWS recommends two or more years of cloud experience and at least a year of hands-on generative AI work. Some format details circulating online (question counts, a 750 passing score) come from secondary reporting rather than the official exam page, so verify against AWS's exam guide before you plan around them. If most of your Stage 3 interest lies in this direction, the dedicated generative AI certification roadmap sequences the wider LLM credential landscape.

Towards MLOps: Microsoft's DP-100 successor. Microsoft's replacement path for the retired Azure Data Scientist Associate is a new MLOps-focused associate certification via exam AI-300. As of August 2026 its official exam details are thin — treat any published question counts or pricing sceptically and rely on Microsoft Learn's certification pages as they fill in. For candidates in Azure-centric organisations, this is the one to watch rather than the one to book today. Remember Microsoft's distinctive maintenance model if you go this route: role-based certifications expire after one year but renew free through an unproctored online assessment.

Towards breadth: a second-platform credential. An AWS-certified ML engineer adding the Google PMLE (or vice versa) signals genuine portability. This is a strong move for consultants and contractors, and largely wasted effort for engineers settled in a single-vendor environment.

For a ranked evaluation of these individual credentials rather than their sequencing, the round-up of the best machine learning certifications owns that comparison.

A realistic 12-month sequencing example

Abstract roadmaps hide the scheduling decisions, so here is how the stages combine for a common profile: a backend developer with three years' experience, daily AWS exposure, and side-project ML only.

• Months 1–2: Skip Stage 0 (already technical). Close the data-layer gap with focused study of the DEA-C01 objectives — pipelines, orchestration, storage trade-offs — and sit the exam if the employer funds it, or study the objectives without sitting it if budget is tight. The knowledge, not the badge, is what Stage 2 requires.
• Months 3–6: Prepare for the AWS ML Engineer – Associate against the MLA-C02 exam guide (given the September 2026 transition). Build two small end-to-end projects on SageMaker — training, deployment, monitoring — because the case-study question format rewards people who have felt the workflow, not memorised it.
• Month 6: Benchmark with a full-length timed practice test, then spend the remaining weeks on the two weakest domains rather than re-reading strengths. ExamPractice's timed practice-test simulation is built for exactly this benchmark-then-target loop; analysing which domains you miss matters more than the headline score.
• Months 7–12: Sit MLA. Then accumulate real generative AI build experience before even considering AIP-C01 — AWS's recommended year of hands-on generative AI work for the professional exam is a genuine signal of its depth, not boilerplate.

The pattern generalises: one anchor exam per six-month block, projects before practice tests, practice tests before booking.

Common roadmap mistakes in 2026

• Studying for a retired exam. Course platforms still sell MLS-C01 and DP-100 material. Check the provider's own certification page before buying anything — as of August 2026, neither exam can be taken.
• Treating course certificates as certifications. Completion certificates from online course platforms are not proctored credentials and do not occupy a stage on this roadmap. The full distinction is explained in AI certifications explained.
• Stacking foundational certs. One Stage 0 credential is a signal; three is a substitute for Stage 2, and hiring managers read it that way.
• Certifying against the wrong cloud. The exam you can rehearse at work beats the exam with the better reputation.
• Ignoring renewal economics. AWS certifications last three years and higher exams auto-renew lower ones; Google professional certs last two years; Databricks requires a retake every two years; Microsoft renews annually but free. Sequencing three vendors' credentials without a calendar plan leads to expiries you only notice after the 30-day grace has passed.
• Booking before benchmarking. A timed, full-length practice run is the cheapest way to discover you need six more weeks — far cheaper than a $150–$300 retake.

Where to start this week

If one instruction survives this article, make it this: pick your Stage 2 anchor first, then work backwards. The anchor determines which foundations matter, which platform you practise on, and which Stage 0 exam (if any) earns its place. A developer in an AWS shop starts by reading the MLA-C02 exam guide; a Google Cloud practitioner starts with the current PMLE guide; a Databricks user starts with the ML Associate guide. Browse the exam pages in the amazon exams hub or google exams hub to see sample questions for the exams on your shortlist, and let the gap between what you can answer and what the exam asks set your timeline.

The 2026 reshuffle was disruptive, but it left the machine learning path clearer than it has ever been: literacy, platform fluency, an engineering associate credential, then a deliberate specialisation. Whether certification is the right investment for your situation at all is a separate question — one the sibling article Is an AI certification worth it? answers directly.

Frequently asked questions

Can I still take the AWS Machine Learning Specialty exam?

No. The last day to sit MLS-C01 was 31 March 2026. Existing holders keep an active certification for three years from the date they earned it. New candidates should target the ML Engineer – Associate instead, with the Generative AI Developer – Professional above it.

Do I need the AI Practitioner before the ML Engineer Associate?

No — AWS exams have no prerequisites. AIF-C01 suits candidates from non-engineering backgrounds; working developers can go straight to MLA, and passing MLA automatically recertifies AIF-C01 anyway if you hold both.

What replaced DP-100 for Azure candidates?

Microsoft's suggested replacement path is a new MLOps-focused associate certification via exam AI-300. Official exam details were still sparse as of August 2026, so check Microsoft Learn's certification pages before committing study time.

How long is each certification valid?

AWS: three years. Google Cloud: two years for professional-level certifications, three for foundational and associate. Databricks: two years, renewed by retaking the current exam. Microsoft role-based certifications: one year, renewed free via an online assessment; fundamentals certifications never expire.

Is the Google PMLE still a Vertex AI exam?

Not purely. Google's page notes the exam has been updated to reflect the transition towards the Gemini Enterprise Agent Platform and recent product renaming, so vet your study materials for currency.