AWS Graviton5 generally available

Amazon said its AWS Graviton5 CPU is generally available and just in time as enterprises are looking at multiple CPU options for AI inference workloads.

Graviton5 was previewed at re:Invent 2025 and the Amazon has a strong backlog of demand. Meta, Snowflake and Uber have procured compute based on Graviton5.

Key points:

• Graviton5 runs 25% faster than its predecessor.
• The custom chip has 192 cores and 33% lower inter-core latency for AI agent workloads.
• Apps and ML inference is 35% faster on Graviton5 relative to the previous version. Databases are 30% faster.
• Amazon EC2 M9g and M9gd instances run on AWS Graviton5.

Amazon CEO Andy Jassy has said Graviton and Trainium are critical to the company. If AWS sold its chips to AWS and other third parties, the custom semiconductor business would have an annual revenue run rate of about $50 billion.

"There’s so much demand for our chips that it’s quite possible we’ll sell racks of them to third parties in the future," said Jassy. Having our own hotly demanded AI chip opens up many possibilities, but perhaps none larger than the ability to lower costs for customers and secure better economics for AWS. At scale, we expect Trainium will save us tens of billions of capex dollars per year, and provide several hundred basis points of operating margin advantage versus relying on others’ chips for inference.”

In his shareholder letter, Jassy said Graviton and Trainium instances are nearly fully subscribed.

Related:

• Google Cloud, AWS, Microsoft Azure: The AI vertical integration race
• AWS growth accelerates to 28% in Q1
• Meta gobbles up AWS Graviton capacity
• Anthropic, Amazon links tighten: Anthropic to spend $100 billion over 10 years on AWS capacity
• Snowflake expands AWS partnership, acquires Natoma, delivers strong Q1