Best AI Certifications for Data Scientists

The two certifications data scientists spent the last five years recommending to each other no longer exist. AWS retired its Machine Learning – Specialty exam on 31 March 2026, and Microsoft retired DP-100 and the Azure Data Scientist Associate certification on 1 June 2026. If a list you are reading still ranks either as a takeable exam, it is out of date — and so, probably, is the rest of its advice.

For a working data scientist in late 2026, the strongest current options are the Google Cloud Professional Machine Learning Engineer, the AWS Certified Machine Learning Engineer – Associate, and the Databricks Certified Generative AI Engineer Associate, with NVIDIA's Generative AI LLMs Associate as a lighter-weight complement. This article ranks those choices for someone who already models for a living and wants a credential that deepens and formalises that work — not a deployment-heavy engineering track (that lane belongs to our guide to AI certifications for software engineers) and not an entry-level primer (see the beginner AI certifications list for those).

What changed, and why it matters for your shortlist

Between early 2025 and mid-2026, the AI certification market went through its biggest reshuffle to date. Three changes reshape a data scientist's shortlist directly:

• AWS Certified Machine Learning – Specialty (MLS-C01) is retired. The last exam day was 31 March 2026. Existing holders keep an active certification for three years from the date they earned it, but nobody can book it now. AWS points candidates to the Machine Learning Engineer – Associate as the closest successor, alongside the AI Practitioner, Data Engineer – Associate, and the new Generative AI Developer – Professional.
• Microsoft's DP-100 and the Azure Data Scientist Associate certification are retired (1 June 2026, per Microsoft's retirement listings). Microsoft's suggested replacement path is a new MLOps-focused associate certification via exam AI-300 — but official exam details for AI-300 were still sparse at the time of writing, so treat it as one to watch rather than one to book blind.
• Generative AI credentials became real proctored exams. Databricks, NVIDIA, Google and AWS all now run invigilated generative-AI certifications, which pull the credential landscape closer to the work many data scientists actually do day to day.

The practical consequence: a data scientist choosing a certification in 2026 is choosing between exams that all lean at least somewhat towards engineering and operations. The question is which one keeps modelling judgement — framing problems, evaluating models, reasoning about data — closest to the centre.

How this list was chosen

Every pick below had to clear three bars. First, it must be an active, proctored certification you can register for today — no retired exams, no course-completion certificates (a Coursera badge is not a proctored credential, a distinction covered in AI certifications explained). Second, it must reward statistical and modelling depth, not just platform navigation. Third, it must be defensible on a data scientist's CV — issued by a provider hiring managers recognise.

The top AI certifications for data scientists in 2026

1. Google Cloud Professional Machine Learning Engineer

The Professional Machine Learning Engineer (PMLE) is the closest thing left to a canonical data-science certification. Google Cloud's exam runs 50–60 multiple-choice and multiple-select questions over two hours, costs $200 plus tax, and has no prerequisites — though Google recommends three or more years of industry experience including at least one year on Google Cloud. Results are pass/fail only; Google does not publish a passing score, so ignore any blog quoting one.

Why it suits data scientists specifically: the exam does not directly assess coding, but Google recommends enough Python and SQL to read code snippets, and the scenarios reward judgement about problem framing, data preparation, model evaluation and responsible AI — the parts of the role a modeller already owns. Be aware the exam has been refreshed: Google notes updates reflecting the transition from Vertex AI-era content towards the Gemini Enterprise Agent Platform and recent product renames, so older prep material that speaks only in Vertex AI terms will lag the current exam guide. The certification is issued at Professional level and renewal follows Google's renewal FAQ, with Professional-tier certifications valid for two years per Google's published policy.

Once you have worked through the current exam guide, a set of Professional Machine Learning Engineer practice questions is a sensible way to check which domains still expose gaps before you spend the $200.

2. AWS Certified Machine Learning Engineer – Associate (MLA)

With the ML Specialty gone, the Machine Learning Engineer – Associate is AWS's flagship for ML practitioners. The current MLA-C01 exam has 65 questions in 130 minutes, costs $150 USD (regional pricing varies), and passes at 720 on AWS's 100–1,000 scaled system. It has no prerequisites; AWS suggests about a year of hands-on work with SageMaker and related services. It is also one of the first AWS exams to use the newer ordering, matching and case-study question types alongside multiple choice.

Timing matters here. AWS has announced a version update: registration for the MLA-C02 beta opens 1 September 2026, the last day to take MLA-C01 in English is 28 September 2026, and the C02 revision adds generative AI, agentic AI and foundation-model workloads. If your preparation is nearly complete, book MLA-C01 before the cut-off; if you are starting from scratch, prepare against the C02 objectives instead. Either way the certification is valid for three years, and earning it automatically recertifies the foundational AI Practitioner.

For a data scientist, MLA is more pipeline-flavoured than PMLE — expect data engineering, deployment and monitoring content — but it remains the right AWS-side choice if your employer runs on AWS. Deeper AWS-versus-Google sequencing questions belong to the machine learning certification roadmap.

3. Databricks Certified Generative AI Engineer Associate

If your data-science work is drifting towards LLM applications — and for many teams it is — the Databricks Certified Generative AI Engineer Associate is the most modelling-adjacent of the generative-AI credentials. The exam is 45 scored multiple-choice questions in 90 minutes, costs $200 per attempt, is proctored online or at a test centre with no aids permitted, and covers prompt engineering, retrieval-augmented generation, vector search, model serving and governance on Databricks. Databricks does not publish a passing percentage. The certification lasts two years, after which you retake the then-current exam; English, Japanese, Brazilian Portuguese and Korean are offered.

There are no formal prerequisites, but Databricks recommends six or more months of hands-on generative AI experience, and the scenario-based questions reflect that. Databricks also offers other data-science and machine-learning credentials, including an ML-focused associate exam — check the current lineup and exam guides on Databricks' certification pages, as details for those were outside what we verified for this article.

4. NVIDIA-Certified Associate: Generative AI LLMs (NCA-GENL)

The NCA-GENL is the quickest meaningful add-on on this list: 50–60 multiple-choice questions in one hour, $125, remote-proctored from home, no formal prerequisites, valid for two years. NVIDIA does not publish a passing score. It tests understanding of generative AI and LLM fundamentals rather than platform operations, which makes it a low-cost way for a data scientist to put a proctored generative-AI credential alongside a deeper cloud certification rather than instead of one. NVIDIA's programme also includes multimodal and infrastructure associate exams and professional-level NCP exams if you want to go further on that track.

5. Worth watching: Microsoft's AI-300 and the post-DP-100 path

Azure-based data scientists are in an awkward interim. DP-100 is gone, AI-102 retired on 30 June 2026 in favour of a new Azure AI Apps and Agents Developer Associate exam (AI-103), and Microsoft's replacement for the data-scientist track is a new MLOps-focused associate certification via exam AI-300. Official pricing, format and duration for AI-300 had not been published on a Microsoft page we could verify at the time of writing, so we are not ranking it — but if your organisation is Azure-committed, watch Microsoft Learn for the AI-300 exam page before defaulting to a cross-cloud alternative. Microsoft's fundamentals-level Azure AI Fundamentals certification continues (now earned via exam AI-901 rather than AI-900), though it is pitched well below a working data scientist's level.