tech
How Amazon weaned Alexa off Anthropic's pricey models to slash AI costs
Internal documents show Amazon worked to lower Alexa+ AI costs by reducing the use of Anthropic models and getting more work from every GPU.
TL;DR
- Amazon is overhauling Alexa to reduce reliance on Anthropic's AI models and cut costs.
- The company aims to quadruple the number of customer transactions per unit of computing capacity.
- Projected AWS cloud costs for Alexa+ were estimated at $1.7 billion in 2026, nearly triple the previous year.
- Initiatives include reducing calls to Claude models, using caching, and expanding deterministic handling for predictable requests.
- Amazon is also focusing on increasing the work each GPU can perform through software upgrades and evaluating its own Trainium chips.
- The strategy aligns with a wider industry trend of optimizing AI inference costs.