ChipsThe story, in brief
Graviton5’s improved design increases speed and energy efficiency — beyond Moore’s law
25% performance gain. Amazon's Graviton5 just rewrote the playbook for AI chip design without waiting for Moore's Law.
The KeyNews take
Why it matters
Amazon is shipping custom silicon optimized for agentic AI workloads with measurable performance and efficiency gains, signaling a shift toward vertical integration in AI infrastructure and reducing dependence on NVIDIA for specialized compute.
The key facts
5 to knowGraviton5 chiplet architecture with custom die-to-die connectivity
25% performance improvement for general-purpose and agentic AI workloads
DDR5-8800 memory support and PCIe Gen6 interconnects
Performance gains claimed beyond Moore's Law scaling
Amazon Science official announcement (source credibility high)
Go to the source
Amazon Scienceamazon.science
Publisher excerpt: A new chiplet architecture, custom die-to-die connectivity, and support for DDR5-8800 memory and the latest PCIe gen6 interconnects improve performance by 25% for general-purpose and agentic AI workloads.