Durable Consolidates Infrastructure, Ships 360B Tokens Annually with 6-Person Engineering Team
Key Takeaways
- ▸Durable serves 1.1B AI tokens daily (360B annually) across ~3M customers with just 6 engineers, demonstrating extreme operational efficiency
- ▸Infrastructure consolidation reduced costs by 3-4x versus self-hosting while improving reliability and multi-tenant security
- ▸AI-specific infrastructure challenges—model orchestration, tenant isolation, and per-customer cost attribution—drove the decision to move away from self-hosted solutions
Summary
Durable, an AI-powered business builder platform serving approximately 3 million customers, has consolidated its infrastructure to handle 360 billion tokens annually—equivalent to 1.1 billion tokens per day. The small team of six engineers achieved 10x operational leverage by consolidating from a multi-service, self-hosted architecture into a unified, multi-tenant platform designed specifically for AI workloads. The migration reduced infrastructure costs by 3-4x compared to self-hosting while maintaining the ability to safely and reliably serve millions of individual customer businesses simultaneously.
Founded on the mission to make business ownership as frictionless as employment, Durable enables entrepreneurs to launch and operate businesses in minutes using AI agents that handle tasks like SEO, content creation, and operations. The company's infrastructure challenges intensified as AI agents became a core platform component, requiring solutions for model orchestration across multiple providers, strict tenant isolation to prevent context leakage, and per-customer cost attribution for usage-based pricing. By choosing to build on a unified platform rather than continue managing disparate self-hosted services, Durable's leadership team eliminated the operational overhead that had become a "second product" in itself.
- A unified platform approach enabled 10x leverage across engineering, product, and design teams by eliminating operational friction
Editorial Opinion
Durable's infrastructure consolidation story highlights a critical inflection point for AI-native platforms: at scale, the complexity of self-hosting AI workloads becomes a liability, not an asset. The company's ability to serve 360B tokens annually with a tiny engineering team suggests that purpose-built, multi-tenant AI infrastructure platforms are becoming table-stakes for companies competing on both speed and unit economics. This mirrors the broader shift toward managed AI inference, but Durable's experience shows the challenges are uniquely acute when AI agents and strict tenant isolation are core requirements.
