The Modern Scaling Playbook: Heroku to AWS
A detailed post-mortem from Argos chronicles its production PostgreSQL migration from Heroku to AWS, offering a playbook for startups outgrowing managed platforms. Key lessons include extensive rehearsal, using logical replication for large datasets, and leveraging managed services like RDS to balance control and reliability.
Heroku's "PaaS tax" can result in costs 3-5 times higher than equivalent AWS infrastructure at scale. This cost difference becomes a primary driver for migration as startups scale their applications and user base. For example, a setup costing around $360 per month on Heroku could be as low as $114 on AWS with reserved pricing. The migration trend accelerated after Heroku eliminated its popular free tier in November 2022, which had been a go-to for developers to host personal projects and prototypes. The company cited rampant abuse as a key reason for the change, which pushed many users to explore more cost-effective or flexible alternatives. Beyond cost, AWS offers far greater flexibility and control over the infrastructure stack compared to Heroku's more opinionated, managed environment. Heroku, which is built on top of AWS, provides a curated selection of services, limiting language and environment choices. In contrast, AWS provides extensive options for compute, databases, and networking, allowing for more complex and resilient system architectures. This move reflects a broader DevOps trend where growing startups adopt Infrastructure as Code (IaC) and containerization to gain more granular control over their environments. Tools like Docker, Kubernetes, and Terraform become essential for managing the increased complexity that comes with the power of an IaaS provider like AWS. The shift in Heroku's strategy became more apparent after its acquisition by Salesforce in 2010 for over $200 million. More recently, in early 2026, Salesforce announced it would no longer offer new Heroku enterprise contracts, signaling a transition to a "sustaining engineering model" to focus more on enterprise AI. Startups now often begin with a Platform as a Service (PaaS) for speed and then architect a gradual migration to an Infrastructure as a Service (IaaS) provider. This allows them to balance initial development velocity with long-term scalability, cost-efficiency, and control as their technical needs and team expertise mature.
