AWS unveils its 192-core Graviton5 chip, with an up to 25% better compute performance than Graviton4, and says Graviton makes up 50%+ of AWS' new CPU capacity
Fifth generation chip provides the best price performance for a broad range of workloads in Amazon EC2. — Key takeaways
Context & Ripple Effects
AWS has steadily expanded its Arm server line from early cost-focused positioning to successive performance and capacity upgrades. The prior Graviton4 generation raised cores, memory bandwidth and performance, making this release the next step in a multi-generation effort rather than a one-off instance launch.
The notable change is deployment scale: AWS says Graviton now supplies more than half of its new CPU capacity. That makes the processor roadmap increasingly central to EC2’s price-performance proposition and AWS’s control over its compute supply.
First-order effects
- EC2 customers with compatible workloads gain access to a 192-core option that AWS says delivers up to 25% more compute performance than Graviton4, potentially improving workload density or price-performance.
- AWS shifts a majority of its incremental CPU capacity toward its own Arm design, deepening Graviton’s role in the services and instance choices it can offer.
Second-order effects
- Customers and software vendors that have not optimized for Arm face a stronger incentive to validate and maintain Arm-compatible deployments, as Graviton becomes a larger share of AWS’s new capacity.
- The move raises the competitive bar for general-purpose cloud CPU offerings: AWS can pair processor iteration with EC2 pricing and capacity decisions, extending the logic behind its earlier Graviton2-based compute instances.
Third-order effects
- If AWS continues directing new capacity to Graviton, custom silicon becomes less a differentiated instance option and more a structural lever in cloud compute economics.
- The longer-term constraint may shift from chip availability alone to the breadth of software portability across processor architectures; the pace of that shift depends on customer migration and workload compatibility.
The trend: Hyperscalers are turning in-house processors into a larger share of mainstream cloud capacity, using hardware-software integration to compete on general-purpose compute economics.