Q3 Earnings: Andy Jassy on why demand for Amazon’s custom chips keeps growing

Summary

• Amazon's chips business grew triple-digit percentages year over year.
• Revenue commitments for Graviton & Trainium each increased more than 20x year over year.
• Graviton serves as a tailwind for AI tasks like post-training, reinforcement learning, & agent tool use, offering 30-40% better price-performance than alternatives.
• Customers saved more than $2 billion in Q3 using custom silicon.
• TorchNeuron was open-sourced to allow developers to run existing PyTorch code on Trainium without rewriting.
• Backlog for prior-generation Graviton instances more than doubled quarter over quarter following the Graviton 5 launch.
• Short-term capacity for prior-generation Trainium instances sells at a premium compared to the first half of the year.
• Amazon intends to make Trainium available outside of AWS late in 2027.
• Inference accounts for well over half of AWS's AI revenue & remains a growing segment.
• Enterprises are still early in using inference at scale.