AWS ML Engineer Associate (MLA-C01): what to expect from the new cert

MLA-C01 went GA in August 2024 and quietly became the AWS ML cert most worth your time. It replaced the old Machine Learning Specialty (MLS-C01), which was retired in 2024 — though MLS-C01 is still showing up on outdated study guides because the internet doesn't update on AWS's schedule. If you came here trying to figure out whether to study MLS-C01 or MLA-C01, the answer is MLA-C01. The old cert is dead.

What's interesting is the shift in scope. MLS-C01 leaned heavy on the data-science side — algorithm choice, hyperparameter tuning, the math of when to use XGBoost vs a neural net. MLA-C01 swings hard the other direction. It's an engineering exam. SageMaker deployment, MLOps pipelines, monitoring drift, debugging endpoints. If you wanted a cert that tests whether you can train a model, this isn't it. If you wanted a cert that tests whether you can ship and operate one, MLA-C01 is exactly that.

Format

65 questions, 170 minutes, $150, scaled passing score 720/1000. Four domains:

• Data Preparation for ML (28%)
• ML Model Development (26%)
• Deployment and Orchestration of ML Workflows (22%)
• ML Solution Monitoring, Maintenance, and Security (24%)

That distribution is misleading on first read. "ML Model Development" sounds like training models, but in MLA-C01 it's actually about choosing built-in SageMaker algorithms and configuring training jobs. You're not asked to design a transformer from scratch.

Who this cert is for

Honestly, three groups:

Data engineers who got dragged into ML. You built the pipelines. Now your team needs SageMaker endpoints behind an API Gateway and you're the one wiring it up. MLA-C01 maps to this work tightly.

Backend engineers shipping ML features. You're not training models. A data scientist hands you a model artifact and you need to deploy it, monitor it, retrain it, and rollback if it drifts. This is the cert for that.

Cloud engineers pivoting toward ML platforms. You're already comfortable with IAM, VPCs, S3 lifecycle, CloudWatch. Now you need to learn the SageMaker shapes. MLA-C01 is a focused way to do it.

It is not the cert for data scientists who do model R&D. They want a different cert — possibly nothing AWS-specific, possibly the GCP Professional ML Engineer if they want any cloud signal at all. AWS's pure-ML cert is gone; AIF-C01 (AI Practitioner) is foundational and lighter; AIP-C01 (GenAI Developer Pro) is the new heavy-ML cert at the Professional tier.

What's actually on it

SageMaker, top to bottom. SageMaker Studio, training jobs (built-in algorithms, BYO container, Script Mode), processing jobs, batch transform, model registry, endpoints (real-time, serverless, asynchronous, multi-model), shadow tests, A/B model variants, autoscaling. Also SageMaker Canvas, SageMaker JumpStart for foundation models, SageMaker Pipelines for orchestration.

You don't need to know every built-in algorithm by heart. You need to know roughly when to use linear learner vs XGBoost vs DeepAR vs Object2Vec, and how to configure training jobs to use Spot, distributed training, and managed warm pools.

Data preparation on AWS. AWS Glue, Glue DataBrew, EMR, Kinesis Data Streams / Firehose / Analytics, Athena, SageMaker Data Wrangler, SageMaker Feature Store. Lots of data engineering bleeds into this exam. If you've taken DEA-C01, ~25% of MLA-C01 will feel familiar.

MLOps patterns. SageMaker Pipelines, AWS Step Functions, EventBridge for triggers, CodePipeline integration, blue/green model deployment, canary rollouts, model registry approvals, CI/CD for ML. Not as deep as DOP-C02 on pure CI/CD but solid coverage.

Monitoring and drift. SageMaker Model Monitor (data quality, model quality, bias drift, feature attribution drift), SageMaker Clarify for bias and explainability, Model Dashboard. CloudWatch metrics for endpoints. This is one of the underrated topics — most candidates skim it because monitoring sounds boring, then 12 questions on the real exam are about it.

Security. IAM for SageMaker, KMS encryption for training data and model artifacts, VPC mode for SageMaker, network isolation, SageMaker Role Manager, PrivateLink for endpoints, audit logging.

What's not heavily tested (good news)

• Pure ML theory. You don't need calculus. You don't need to derive backprop. You're not asked which optimizer beats which.
• Deep learning architecture design. No questions on choosing transformer head counts.
• Statistical hypothesis testing. Old MLS-C01 had this; MLA-C01 dropped most of it.
• Generative AI specifics. That's AIF-C01 (foundational) and AIP-C01 (Pro). MLA-C01 mentions JumpStart and Bedrock but at the integration level, not at depth.

How to study

Resources are still maturing — the cert is only ~20 months old. As of April 2026:

• AWS Skill Builder has the official MLA-C01 learning path. It's good and free with an account.
• Stephane Maarek has an MLA-C01 course on Udemy that's well-paced.
• Adrian Cantrill hadn't released a full MLA-C01 course as of late 2025 — check whether one exists yet. If it does, it'll be the deepest option.
• Tutorials Dojo has practice exams and explanations. Quality is solid.
• AWS official practice exams on Skill Builder. Closest to the real thing.

Hands-on matters more than for SAA-C03. Spin up a SageMaker Studio domain, train a built-in algorithm on a Kaggle dataset, deploy to a real-time endpoint, hit it with curl, then deploy the same model to a serverless endpoint and notice the cold-start. Set up Model Monitor on the endpoint and trigger a drift alert. That whole exercise costs $5–$15 in AWS bills and teaches you 30% of the exam.

Time budget:

• Data engineer or backend with ML exposure: 6–8 weeks at 10 hrs/week.
• Cloud engineer without ML background: 12 weeks. Build something real.
• Data scientist learning the AWS side: 6–10 weeks, mostly on the deployment and ops topics.

Career signal: still emerging

This is where I'm going to be honest about uncertainty: MLA-C01 is too new for clean salary data. levels.fyi and Glassdoor don't yet cluster it as a distinct credential — most "ML engineer" roles in 2026 still list MLS-C01 (which doesn't exist anymore) or list no AWS cert at all. The job postings that do mention MLA-C01 are concentrated in mid-to-large companies with formal ML platform teams.

What I can say from informal data and conversations: senior ML engineers in major US metros land $180k–$280k base in 2026, with TC pushing $400k+ at FAANG-tier shops. MLA-C01 doesn't move that number much on its own — experience does. The cert is probably worth $5k–$15k in the same way SAA-C03 is for SA roles: a recruiter-facing signal that you've passed a knowledge bar.

What MLA-C01 does signal cleanly: that you can take a model from a notebook to a production endpoint without breaking things. Which, if you're a hiring manager for an ML platform team, is exactly what you want to know.

Comparison: MLA-C01 vs alternatives

• AIF-C01 is foundational. Take it if you want a generic AI signal for non-engineering roles. Doesn't replace MLA-C01.
• AIP-C01 is the Pro-tier GenAI Developer cert. Heavier on generative AI integration (Bedrock, prompt engineering, RAG patterns). Take it if your job is shipping LLM features. Stack it after MLA-C01 if you want both ML and GenAI signal.
• DEA-C01 (Data Engineer Associate) overlaps about 25% with MLA-C01 on the data prep side. If you do both, take DEA-C01 first.
• GCP Professional ML Engineer (PMLE) is the closest equivalent on Google Cloud. Heavier on Vertex AI and TPU specifics. Not interchangeable but similar shape.

Bottom line

MLA-C01 is a focused, fairly difficult Associate exam that maps closely to a real job — ML platform engineer / MLOps engineer. If you do that work, take it. If you're trying to break into ML from a backend background and want one cert that signals competence to recruiters, this is it. The salary data is still emerging but the trajectory is clearly up — ML engineering roles are hiring fast in 2026.

If you're studying, browse the MLA-C01 question bank on CertLabPro or start a timed exam. And spin up a SageMaker endpoint this week. The hands-on work is what makes the exam feel obvious.