AWS Roadmap 2027 for Data and AI Careers
An AWS roadmap for 2027 for people heading into data and AI roles: cloud foundations, the data lake and lakehouse stack with S3 Tables and Iceberg, streaming, machine learning with SageMaker AI, and generative AI with Bedrock and AgentCore, plus the 2025-26 launches that changed the plan and a nine-month timeline.
GloryTecks Team
Training & Placement Faculty
An AWS roadmap for 2027 for data and AI careers has five layers: cloud foundations, the data lake and lakehouse stack, streaming, machine learning, and generative AI with agents. The foundations are the same as ever: identity and access, networking, storage and cost. The upper layers changed a lot in 2025-26. AWS made Bedrock AgentCore generally available for running agents, added the Nova 2 model family, made S3 Vectors generally available as cheap vector storage, and brought Apache Iceberg version 3 to its analytics services and S3 Tables. If you learned AWS for data work a few years ago, the core still holds; the way data and AI services fit together is newer.
The launches that changed the 2027 plan
| Launch | Date | Why it matters |
|---|---|---|
| Amazon Bedrock AgentCore generally available | October 2025 | A managed runtime, gateway, memory, identity and observability for agents built with any framework |
| Nova 2 models, S3 Vectors GA, AgentCore evaluations and policy | December 2025 | Cheaper vector storage, new models and governance for agents |
| Iceberg v3 support across AWS analytics | 2026 | Deletion vectors, row lineage and new types in EMR, Glue and S3 Tables |
| Certification changes | 2026 | New Generative AI Developer Professional; ML Engineer Associate updated |
AgentCore's general availability announcement highlights VPC and PrivateLink support, eight-hour execution windows with session isolation and availability in nine regions, including Asia Pacific (Mumbai). AWS's re:Invent 2025 roundup covers Nova 2, S3 Vectors with up to 2 billion vectors per index, and AgentCore's new evaluations and policy controls.
Layer 1: cloud foundations
IAM users, roles and policies; VPCs, subnets and security groups; S3 storage classes and lifecycle rules; EC2, Lambda and containers at a basic level; CloudWatch for logs and metrics; and AWS Budgets and Cost Explorer. Learn to read a bill and explain what drives it.
Milestone: a small application deployed with least-privilege roles, logging and a budget alert.
Layer 2: the data lake and lakehouse
S3 as the data lake, the Glue Data Catalog, Glue ETL with PySpark, Athena for SQL on S3, Redshift for warehousing and EMR for large Spark jobs. Then open table formats: AWS's Iceberg v3 documentation describes support for deletion vectors, row lineage and default values in Spark on EMR, Glue ETL, SageMaker Unified Studio notebooks and the Glue Data Catalog, with S3 Tables supporting all of v3's new data types. Learn the medallion pattern of raw, cleaned and curated layers on Iceberg tables.
Milestone: a pipeline from raw files to curated Iceberg tables, queried with Athena and catalogued in Glue.
Layer 3: streaming and orchestration
Kinesis Data Streams and Amazon Data Firehose for event ingestion, Amazon MSK for managed Kafka, and change data capture from databases with AWS DMS. Orchestrate with Step Functions or Amazon MWAA, AWS's managed Airflow. Build pipelines that are idempotent and observable.
Milestone: database changes streamed into a lakehouse table, orchestrated end to end with alerts.
Layer 4: machine learning with SageMaker AI
Data preparation, training jobs, hyperparameter tuning, model registry, real-time and batch endpoints, pipelines and model monitoring. You do not need every SageMaker feature; you need one model taken from training to a monitored endpoint.
Milestone: a model trained, registered, deployed and monitored, with a retraining pipeline.
Layer 5: generative AI and agents with Bedrock
Bedrock model access and prompt management, Knowledge Bases for retrieval-augmented generation, Guardrails for safety, vector storage with OpenSearch, Aurora PostgreSQL with pgvector or S3 Vectors, and agents with tool use. Then AgentCore for running agents in production: the runtime, gateway for tools and MCP, memory, identity and observability, plus its evaluations and policy controls.
Milestone: a document assistant with citations, guardrails, an evaluation set and an agent that calls one internal tool through AgentCore.
A nine-month plan
| Month | Layer | Output |
|---|---|---|
| 1 | Foundations | Secured, monitored small application with a budget |
| 2-3 | Data lake and lakehouse | Raw-to-curated Iceberg pipeline with Glue and Athena |
| 4 | Warehousing | Redshift model and a dashboard on top |
| 5 | Streaming and orchestration | CDC stream into the lakehouse, orchestrated |
| 6 | Machine learning | Trained, deployed and monitored model |
| 7-8 | Generative AI and agents | Evaluated Bedrock assistant and an AgentCore agent |
| 9 | Certification and portfolio | One associate exam and three documented projects |
Cost habits to build from day one
AWS bills for what you leave running, so build these habits in month one rather than after a surprise bill. Set a budget alert before you create anything. Tag every resource with a project name so you can see what costs what. Stop or delete practice clusters, endpoints and notebooks at the end of each session; SageMaker endpoints and EMR clusters keep charging while idle. Prefer serverless and pay-per-query services such as Athena and Lambda for learning projects. And for generative AI work, log token usage per request from your first prototype, because model calls are often the largest cost in an AI application. Being able to explain the monthly cost of your portfolio project is a strong interview answer.
Choosing a specialisation within AWS
- Data engineering: layers 1-3 in depth, Data Engineer Associate, and Redshift or EMR performance tuning.
- ML engineering: layers 1, 2 and 4 in depth, ML Engineer Associate, and MLOps practices.
- Generative AI engineering: layers 1 and 5 in depth, plus enough of layer 2 to manage document pipelines; Generative AI Developer Professional once you have production experience.
Learning AWS at GloryTecks
The four-month AWS Data Engineering course covers S3 data lakes, Glue, Athena and EMR with Iceberg, Redshift, Kinesis and MSK, DMS and orchestration with Step Functions and MWAA. The two-month AWS Generative AI with Amazon Bedrock course covers Bedrock models and APIs, prompt management, embeddings and vector stores, Knowledge Bases, Guardrails, agents and evaluation. For foundations, the AWS Solutions Architect Associate course covers IAM, VPC, compute, storage, databases and resilient architectures. All run in classroom batches in Ameerpet and live online. Our AWS data engineering projects guide has portfolio ideas, and the 2026 AWS roadmap for data engineers covers the data layers in more depth.
How long does it take to learn AWS for data and AI roles?
What is Amazon Bedrock AgentCore?
Which AWS services should a data engineer learn in 2027?
What are S3 Vectors?
Should I learn AWS, Azure or Google Cloud in 2027?
GloryTecks Team
Training & Placement Faculty
The GloryTecks faculty team mentors learners across Data Science, AI and Data Engineering, with a focus on real-time projects and placement readiness in Hyderabad.
Want to master AWS?
Learn Data Engineering at GloryTecks in Hyderabad, with trainers who have industry experience, real-time projects and placement assistance.
Related articles
AWS Certifications in 2027: Which Exam to Take for Data and AI Roles
AWS reshaped its certifications in 2026: the Machine Learning Specialty is gone, a Generative AI Developer Professional exam arrived, the ML Engineer Associate moved to a new version and SysOps became CloudOps. Here are the current exams, the changes and the right path for data and AI roles in 2027.
AWS Roadmap 2026 for Data Engineers
What to learn on AWS as a data engineer, in the order data moves through a pipeline: S3 and IAM, Glue and Athena, Glue jobs, Redshift, Step Functions, streaming and governance, with something to build at every stage and a clear test for when you are done.
Amazon S3 Explained: Storage, Classes and Best Practices
How Amazon S3 storage classes differ — minimum storage duration, minimum billable size, retrieval time — and a four-question method for putting each dataset in the right class, with lifecycle and security defaults for a data-engineering bucket.