The NS0-901 AI Expert exam is 60 scenario-based questions in 90 minutes, and it costs US$250. It is half AI theory and half NetApp tooling: 14 of its 31 objectives name a NetApp product or solution, and in Domain 5, AI common challenges (22%), every single objective does.
That split is the most useful thing to know before you plan your study. Candidates with a data science background tend to know the theory and miss the products. Storage admins tend to know the products and miss the theory. This entry sorts the official objectives into those two piles, domain by domain, and then builds a four-week plan around them. Every exam figure here was checked on NetApp's and Pearson VUE's own pages on 27 October 2026.
What is the NS0-901 AI Expert exam?
NS0-901 is the exam behind the NetApp Certified AI Expert certification. According to the NetApp Certified AI Expert page, it is meant for technical professionals who build infrastructure for AI workloads and have six to 12 months of hands-on experience with AI. NetApp says you should understand its AI solutions, general AI concepts, the AI lifecycle, AI software and hardware architecture, and the common challenges teams run into.
The key facts:
- Exam: NS0-901, NetApp Certified AI Expert (often shortened to NCAE).
- Format: 60 scenario-based questions in 90 minutes, which is about 1.5 minutes per question.
- Price: US$250. NetApp prices its Expert-level exams above the usual US$200.
- Passing score: NetApp does not state one on its AI Expert page or its policy page.
- Validity: 24 months once earned.
- Retakes: after a first failure you can retake straight away; after more than one failure you wait 14 days.
The question count and length come from the Pearson VUE NetApp exam page, which announced the exam as available to employees, partners and customers.
How are the five NS0-901 domains weighted?
The lifecycle domain is the largest, at 27%, and the challenges domain is next, at 22%. Here are all five, with the number of objectives under each and roughly how many of the 60 questions each weight works out to:
- Domain 1, AI overview (15%): 6 objectives, about 9 questions.
- Domain 2, AI lifecycle (27%): 7 objectives, about 16 questions.
- Domain 3, AI software architectures (18%): 5 objectives, about 11 questions.
- Domain 4, AI hardware architectures (18%): 7 objectives, about 11 questions.
- Domain 5, AI common challenges (22%): 6 objectives, about 13 questions.
The question counts are my own arithmetic from the weights, rounded. NetApp does not publish a per-domain count.
Which parts are AI theory and which are NetApp products?
Of the 31 objectives, 17 are general AI knowledge and 14 name a NetApp product or solution. If you assume each objective inside a domain carries the same weight, the NetApp share of the exam comes to roughly 48%. The split changes a lot from one domain to the next.
- Domain 1 (0 of 6 NetApp): pure theory. Training versus inferencing, AI versus machine learning versus deep learning, supervised, unsupervised and reinforcement learning, industry uses, AI with HPC and analytics, and on-premises versus cloud versus edge.
- Domain 2 (2 of 7 NetApp): mostly theory. Predictive versus generative AI, transformers, hallucinations, and RAG versus fine-tuning. The two NetApp objectives cover data preparation (BlueXP classification, XCP, CopySync) and inferencing with NetApp data mobility.
- Domain 3 (3 of 5 NetApp): the tipping point. MLOps platforms and notebooks versus pipelines are general. The NetApp DataOps Toolkit, Trident for Kubernetes, and the BlueXP GenAI Toolkit and Workload Factory are not.
- Domain 4 (3 of 7 NetApp): data lakes, CPU versus GPU versus TPU versus FPGA, Ethernet versus InfiniBand, and file versus object protocols are general. Storage platforms (C-Series, A-Series, EF-Series, StorageGRID), SuperPOD with E-Series and BeeGFS, and AIPod, FlexPod AI and OVX are NetApp-specific.
- Domain 5 (6 of 6 NetApp): all product. Sizing, traceability, data movement, cost, security and GPU utilisation are each tied to named NetApp features.
Which NetApp products should you know for NS0-901?
You need to know what each named product does in an AI pipeline, not how to configure it step by step. Group them by the job they do and they become much easier to remember:
- Moving and preparing data: XCP, CopySync, SnapMirror, FlexCache and BlueXP classification.
- Tracing data, code and models: Snapshots and cloning, so a team can show exactly which data trained which model version.
- Running AI software: the NetApp DataOps Toolkit (Python, on Kubernetes or standalone), Trident for Kubernetes storage, and the BlueXP GenAI Toolkit and Workload Factory for using private data with generative AI.
- Storage platforms: C-Series, A-Series, EF-Series and E-Series, StorageGRID, and BeeGFS for parallel file system work.
- Reference stacks: SuperPOD, BasePOD, AIPod, FlexPod AI and OVX.
- Cost and protection: storage efficiencies, FabricPool, Data Infrastructure Insights, Keystone, Autonomous Ransomware Protection and Multi-Admin Verification.
A good test of readiness is to explain each one in a single sentence that starts with the AI problem it solves. For example: "When a training job needs the same dataset in two sites, SnapMirror or FlexCache moves or caches it."
What general AI knowledge does NS0-901 expect?
It expects the working vocabulary of someone who supports data scientists, not of a researcher. You won't need to derive anything, but you do need to tell similar ideas apart under time pressure:
- Training needs data, code, compute and time. Inferencing needs the model loaded into memory, so model size matters.
- Building a model starts from data and code. Fine-tuning starts from an existing model plus your own data and code.
- RAG looks up fresh or private data at answer time. Fine-tuning bakes knowledge into the model itself.
- Notebooks such as Jupyter are for experiments. Pipelines are for repeatable production work.
- RDMA and GPUDirect Storage exist to keep data flowing to GPUs without the CPU getting in the way.
- Warehouses, data lakes and lakehouses are three ways to gather data before it feeds a model.
If any of these pairs feels shaky, fix it before you start on the NetApp products. Several Domain 5 objectives build on them, such as sizing storage and compute around GPU memory.
What does a 4-week NS0-901 study plan look like?
Plan for about 36 hours over four weeks, split roughly by the domain weights: Domain 1 gets 5 hours, Domain 2 gets 10, Domain 3 gets 6, Domain 4 gets 7 and Domain 5 gets 8. The order runs from theory towards products, so that each NetApp tool arrives after the problem it solves.
- Week 1 (9 hours): all of Domain 1 (5 hours), then the first half of Domain 2 (4 hours): predictive versus generative AI, transformers and hallucinations.
- Week 2 (9 hours): the rest of Domain 2 (6 hours): RAG versus fine-tuning, model-building and inferencing requirements, and the data preparation tools. Then start Domain 3 (3 hours) with MLOps platforms and notebooks versus pipelines.
- Week 3 (10 hours): finish Domain 3 (3 hours) with the DataOps Toolkit, Trident and BlueXP. Then all of Domain 4 (7 hours): compute, networking, protocols, storage platforms and the reference stacks.
- Week 4 (8 hours): Domain 5 in full: sizing, traceability, data movement, cost, security and GPU performance. Finish with a timed set of practice questions at 1.5 minutes each.
At the end of each week, write the product-in-one-sentence list from memory for everything you covered. Anything you can't explain goes to the top of the next week.
The same weight-to-hours method works for other blueprints. This blog applies it to a Cisco field technician exam in 800-150 FLDTEC exam topics and a five-week plan.
What mistakes do NS0-901 candidates often make?
Most lost marks come from studying only one half of the exam. These are the patterns to watch for:
- Treating it as a pure AI theory test. Domain 5 alone is 22% of the exam, and every objective in it names a NetApp product.
- Treating it as an ONTAP admin test. Domain 1 is 15% with no NetApp product in it at all. Domain 2 is 27% and mostly theory.
- Mixing up RAG and fine-tuning. The exam lists them side by side, so expect a scenario that asks which one fits.
- Learning products as a list of names. Scenario questions describe a problem first. Learn each product by the problem it solves.
- Ignoring the clock. 60 scenarios in 90 minutes leaves no room to reread long questions. Practise reading for the deciding detail.
What does an NS0-901-style question look like?
Here is one practice question, written for this entry in the style of the objectives. It is not taken from the exam.
Question: An audit asks a data science team to show exactly which dataset was used to train the model version that went live six months ago. The data sits on NetApp storage. Which approach best supports this?
- A. Retrain the model on today's dataset and compare the outputs.
- B. Keep a Snapshot copy of the dataset for each training run, and clone it when it needs to be examined.
- C. Move the dataset to an object store after each training run.
- D. Add more GPUs so retraining is faster.
Answer: B. Domain 5 ties code, data and model traceability to Snapshots and cloning. A Snapshot copy freezes the dataset as it was for that run. A clone lets the team open that frozen copy without touching the live data.
Watch: a short video on the NS0-901 exam
This short video from NWExam's YouTube channel walks through NS0-901 sample questions.
Who should take the NetApp AI Expert certification?
It suits storage, infrastructure and platform engineers who already support AI teams, and AI engineers who run their pipelines on NetApp storage. NetApp's guide is six to 12 months of technical experience with AI.
It also counts towards a bigger credential. The AI Data Infrastructure Certified Expert badge asks for the AI Expert exam, the NetApp Certified Data Administrator ONTAP exam, and one of two NVIDIA professional exams: AI Infrastructure or AI Operations.
How do the price, retakes and renewal work?
The exam costs US$250, and the certification stays active for 24 months. The rules sit in NetApp's certification program policies, which also explain that a candidate who passes must wait 12 months to sit the same exam again. If you fail once you can book again right away; after a second failure there is a 14-day wait.
Frequently asked questions
1. How many questions are on the NS0-901 AI Expert exam?
The exam has 60 scenario-based questions, and you get 90 minutes, about 1.5 minutes per question.
2. How much does the NS0-901 exam cost?
It costs US$250. NetApp prices its Expert-level exams, including the AI Expert exam, above its usual US$200 exam price.
3. Which NS0-901 domain carries the most weight?
Domain 2, AI lifecycle, carries 27%. Domain 5, AI common challenges, is next at 22%, and every one of its objectives names a NetApp product.
4. Do I need ONTAP experience for NS0-901?
NetApp's AI Expert page lists no prerequisite exam, but 14 of the 31 objectives name NetApp products such as Snapshots, SnapMirror, FlexCache and Trident. NetApp recommends six to 12 months of technical experience with AI.
5. How long does the NetApp Certified AI Expert certification last?
It stays active for 24 months after you earn it. If you fail the exam once, you can retake it straight away; after more than one failure, you wait 14 days.