March Networks Moved 5,600 TB of Video to AWS — and Cut Storage Costs by 79%
Cloud Platform EngineeringEnterprise video surveillance generates data at a scale most cloud cost discussions don’t account for. One retail customer in March Networks’ deployment ran 580+ cameras producing approximately 5,600 terabytes of archived footage — and paid around $1.7 million per year to store it on-premise. The cloud migration path cut that to $347,000 annually. The cost structure change wasn’t achieved through clever optimization; it came from tiering.
The Architecture
March Networks built its cloud platform on a tiered S3 strategy: Amazon S3 Standard and S3 Standard-Infrequent Access for footage that needs to be accessible, and S3 Glacier for long-term archival. The split aligns storage cost with actual access patterns — recent footage is queried regularly for incident review, while archived footage from six months ago may never be accessed at all. Paying premium storage rates for footage that will sit untouched is where on-premise economics collapse at scale.
Supporting infrastructure draws from the standard AWS platform engineering toolkit: Amazon SQS for asynchronous messaging between distributed sites, Amazon CloudWatch for monitoring and alerting, and PostgreSQL with Amazon ElastiCache for Redis to manage metadata. AWS Security Token Service handles authentication without requiring long-lived credentials distributed across thousands of camera endpoints.
The AI layer sits on top: integration with Amazon S3 Vectors and Amazon Bedrock enables semantic search and natural-language queries across the video dataset. Instead of reviewing footage by timestamp and camera location, operators can ask “show me incidents near the loading dock last Tuesday” and retrieve the relevant clips. That capability isn’t theoretical — it’s production behavior for organizations that have migrated at this scale.
The Scalability Argument
The on-premise model has a fixed cost problem: adding cameras, extending retention windows, or onboarding new locations requires hardware procurement and provisioning cycles. Each expansion is a capital project. The cloud model turns that into an elastic capacity decision — storage scales automatically as cameras are added or retention periods change, with no hardware lead time.
For a distributed enterprise with retail locations, restaurants, banks, or transportation networks across thousands of sites, that elasticity changes the economics of video retention entirely. Retention periods that were kept short because of storage costs can be extended because the marginal cost of additional months is predictable and low. Incident investigation that previously required on-site access can now happen centrally.
The So What
March Networks’ architecture is straightforward by enterprise cloud standards — S3 tiering, standard supporting services, Bedrock for AI queries. What makes it notable is the domain: video surveillance has been a laggard in cloud adoption because of the data volume, the regulatory considerations around footage retention, and the distributed nature of camera infrastructure.
The numbers from this deployment suggest those barriers have been addressed well enough to produce 79% cost reductions at scale. For engineering teams evaluating cloud-native video infrastructure, the tiering strategy is the core lesson: don’t pay active storage rates for archival access patterns. The cost difference at petabyte scale is the difference between a viable and non-viable cloud migration.
Source: AWS Architecture Blog
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →