Cloud skills an AI engineer needs: a matrix by level
In this article (5 sections)
Related programme: the FDE for Freshers course (6–7 months; EMI from ₹8,000/month · full fee shared on your counselling call). See also open Forward Deployed Engineer jobs in India, updated daily.
An AI engineer needs enough cloud skill to deploy a model-backed service safely and keep it running: identity and access, secret handling, basic networking, choosing compute, storage for documents and vectors, containerised deployment, cost control and monitoring. You don't need to be a cloud architect, but you do need to be able to answer "where does this run, who can reach it, what does it cost and how will we know it broke?" The matrix below breaks each area into three levels so you can place yourself and pick what to learn next.
The matrix
Levels: L1 — can use it with guidance; L2 — can set it up for your own service; L3 — can design it and explain trade-offs to others.
| Area | L1 | L2 | L3 |
|---|---|---|---|
| Identity and access | Log in with roles; understand users vs service accounts | Create a service identity with least-privilege permissions for your app | Design access across environments; review others' permissions |
| Secrets | Know API keys must not be in code | Store keys in a secrets manager and read them at runtime; rotate one | Set rotation and access policies; handle a leaked key |
| Networking | Know public vs private endpoints | Put a service behind HTTPS; restrict inbound access | Private connectivity to model endpoints and databases; explain data paths to security |
| Compute | Run a container or function someone else set up | Choose between serverless, containers and VMs for a workload | Size GPU vs CPU inference; plan for scale and cold starts |
| Storage | Upload and read files from object storage | Set up a database or vector store with backups | Choose storage for documents, embeddings and logs with retention rules |
| Deployment | Deploy with an existing pipeline | Write a Dockerfile and a CI pipeline that tests and deploys | Environments, rollbacks and infrastructure as code |
| Model services | Call a managed model API | Configure a managed model endpoint with quotas and logging | Compare managed, hosted and self-run options on cost, latency and data terms |
| Cost | Know that tokens, GPUs and storage cost money | Set a budget alert; estimate cost per request | Build a cost model; find and cut the biggest cost driver |
| Monitoring | Read logs in the console | Structured logs, basic metrics and one alert for your service | Dashboards for latency, errors, cost and quality; incident response |
| Data residency | Know where your data is stored | Pick regions deliberately | Explain residency and transfer to a client's legal and security teams |
Where to aim by role
| Role | Aim for |
|---|---|
| Fresher applying for AI engineer or FDE roles | L2 in secrets, compute, deployment and monitoring; L1 elsewhere |
| Data scientist moving to AI engineering | L2 in deployment, model services, cost and monitoring |
| Experienced engineer targeting client-facing FDE work | L3 in networking, data residency, cost and monitoring |
For client-facing roles, networking and data residency rise in importance because they are where enterprise security reviews spend most of their time. The IT services delivery to FDE guide shows how infrastructure experience carries over.
Place yourself honestly
For each row, ask: have I done this myself, on a system that ran for more than a day? Reading about it or watching a tutorial counts as L1 at most. Keep evidence — a repository, a screenshot of an alert you configured, a cost estimate you wrote.
A quick way to practise the cost row is to write the estimate before you deploy. A worked example with hypothetical prices:
# All prices hypothetical; replace with your provider's current price list.
requests_per_day = 1_200
input_tokens, output_tokens = 1_800, 350 # per request, measured on test set
price_in_per_million = 40.0 # hypothetical ₹ per 1M input tokens
price_out_per_million = 160.0 # hypothetical ₹ per 1M output tokens
hosting_per_month = 3_000.0 # hypothetical ₹ for container + storage
per_request = (input_tokens * price_in_per_million +
output_tokens * price_out_per_million) / 1_000_000
monthly = per_request * requests_per_day * 30 + hosting_per_month
print(f"per request ₹{per_request:.3f}, monthly ₹{monthly:,.0f}")
assert round(per_request, 3) == 0.128
assert round(monthly) == 7608Two things make this useful: token counts measured on your own test set rather than guessed, and hosting included, because it is often forgotten. The script shows the arithmetic only; the ₹ figures are placeholders.
A project that covers L2 across the matrix
Deploy one small retrieval or extraction service and meet these criteria:
- Runs in a container behind HTTPS on a cloud of your choice.
- Uses a dedicated service identity with only the permissions it needs.
- Reads its model API key from a secrets manager; nothing sensitive in the repository.
- Stores documents in object storage and has a backup or rebuild procedure written down.
- Has a CI pipeline that runs tests before deploying.
- Emits structured logs with request ID, latency and token counts.
- Has one alert (for example, error rate above a threshold) and a budget alert.
- Includes a one-page cost estimate like the one above, then a comparison with the actual bill after a week.
Tear it down afterwards to stop charges, and keep the write-up.
Certifications and the matrix
A cloud certification can structure your L1-to-L2 learning, but check the current exam version on the official page before starting — several AI-related exams changed in 2026. See what you can verify about AI certifications.
The FDE for Freshers programme covers production engineering and cloud deployment as a phase of its curriculum, alongside LLM and agent work. It is delivered live online, with optional in-person labs at our Noida centre.
Frequently asked questions
Which cloud should I learn first for AI engineering?
The one your target employers use. Read ten job descriptions and count which cloud they name. The underlying concepts — identity, networking, storage, containers, monitoring — transfer, so the first cloud you learn well makes the second much easier.
Pankit Kumar has 10 years in Data Science & AI, building and shipping production systems in regulated pharma and clinical environments. He is a freelance trainer at Boston Institute of Analytics, AnalytixLabs and Scaler, and has taught this material to thousands of working professionals.
This article is part of our FDE for Freshers programme — 6–7 months. Build your engineering foundations, then take AI from discovery to delivery.
Explore FDE for Freshers