How Azure Cosmos DB Runs Under the Hood | Andrew Liu | Azure Cosmos DB Conf 2026
Andrew Liu explains what happens inside Azure Cosmos DB from provisioning through serving requests at scale, focusing on the service’s internal deployment model, replication approach, partitioning mechanics, and how the serverless Request Unit model works under the covers.
Overview
Andrew Liu walks through Azure Cosmos DB’s internal architecture, focusing on what happens between provisioning a database and it being ready to handle production traffic at global scale.
Deploying the database engine onto hardware
- How the Cosmos DB database engine is deployed onto hardware in datacenters worldwide.
- Mentions hardware partners such as AMD as part of the underlying infrastructure story.
Replication and high availability
- How replica sets are used.
- How quorum-based replication is used to deliver high availability.
Data placement and partitioning
- How consistent hashing determines where data lands.
- Why choosing a good partition key matters for how data is distributed.
Serverless and Request Units (RUs)
- How the pay-per-request serverless model was built.
- What a Request Unit (RU) represents “under the covers.”
Sharing throughput with RU pooling
- How RU pooling can be used to share throughput capacity across a Cosmos DB fleet.
Links from the description
- Cosmos Conf 2026 playlist: https://aka.ms/CosmosConf26Playlist
- Cosmos Conf Challenge (DP-420 voucher for first 500 completions): https://aka.ms/CosmosDBConfChallenge
- Post-event survey: https://aka.ms/CosmosConf2026Survey
- Conference website: https://aka.ms/azurecosmosdbconf
- Speaker page: https://developer.azurecosmosdb.com/conf/#speaker/andrew-liu
- Andrew Liu on LinkedIn: https://www.linkedin.com/in/aliuy/
- Azure Cosmos DB on YouTube: https://www.youtube.com/@AzureCosmosDB
- Azure Cosmos DB on X: https://x.com/AzureCosmosDB
- Azure Cosmos DB on LinkedIn: https://www.linkedin.com/company/azure-cosmos-db/