AWS re:Invent 2025 was all about AI.
Trainium 3 Chip
Amazon's third-generation in-house AI training chip: - FP8 compute: 4 PFLOPS (vs. H100's 2 PFLOPS) - Energy efficiency: 40% better than H100 - Largest cluster: 100,000 cards - Positioning: Not competing with NVIDIA on peak performance, but focusing on cost-effectiveness
Amazon Nova
Nova is AWS's own multimodal large model: - Supports text, image, and video understanding - Deep integration with AWS services (S3, DynamoDB, Lambda) - Positioned as "practical AI" rather than "the most powerful AI"
Amazon Q 2.0
An AI coding assistant for developers, competing with GitHub Copilot: - Supports code generation, review, and testing - Deep integration with the AWS console - Priced 30% lower than Copilot
Strategic Significance
AWS's AI strategy is clear: instead of pursuing the most powerful models, it aims to be the "most convenient AI cloud computing" platform. Developers can use AI without leaving the AWS ecosystem—covering the entire chain from training to inference to applications.
This may be the most effective strategy against Microsoft (Azure + OpenAI).
Comments
Comments are closed.