Uber implements "Zero Growth Stack" for infrastructure and AI cost optimization

Zero Growth Stack Strategy

Uber has implemented a "Zero Growth Stack" strategy to manage its infrastructure more efficiently. This approach aims to separate capacity growth from business demand, allowing the company to reduce its physical hardware footprint while maintaining service scalability. The strategy focuses on automated runtime optimization and AI lifecycle management to mitigate traditional resource scaling overheads.

Go Runtime Optimization with GOGCTunner

A key component of this initiative is the optimization of Go runtime performance. Uber developed GOGCTunner, a library that automates the adjustment of the GOGC value, which controls garbage collection frequency. This dynamic tuning addresses services with varied memory footprints, reclaiming 70,000 CPU cores across 30 mission-critical services. GOGCTunner balances heap size against memory utilization thresholds to prevent out-of-memory events.

Generative AI Integration and Cost Challenges

Uber integrated generative AI into its software development lifecycle to enhance developer velocity, with 92% of engineers using these agents monthly and 31% of new code being AI-authored. The architecture includes a Platform Layer (Michelangelo AI), a Context Layer (MCP Gateway), a Specialised Layer (Minion and Shepherd agents), and a Review Layer (uReview and Code Inbox).

However, the rapid adoption of AI tools led to significant economic friction. AI-related costs increased sixfold since 2024 due to token-based billing models, with individual developer costs reaching $2,000 monthly by early 2026.

AI Cost Governance and Efficacy Auditing

To address the escalating costs, Uber transitioned from unconstrained AI adoption to strict cost governance. The company implemented usage caps, limiting AI tool expenditure to $1,500 per developer. Uber is also auditing the efficacy of AI-generated code to quantify its direct impact and ensure cost-effectiveness.