AWS launches its ARM-based Graviton 3 processor, claiming it will be 25% faster than Graviton 2 and three times faster for ML workloads
Context & Ripple Effects
Graviton 3 is the third step in a line AWS started in 2018, when it introduced its first custom-designed Arm server processor claiming 45% lower costs for some workloads. The Graviton2 update in 2019 reframed the chip from a cost play to a performance claim — 40% better than comparable x86 instances at 20% less — and Graviton 3 extends that framing into machine learning, where AWS claims three times the speed.
First-order effects
- AWS customers running general-purpose workloads get an upgrade path promising 25% more performance per instance, while ML-heavy customers get a dedicated pitch at three times Graviton 2's speed.
Second-order effects
- Every generation of Graviton has been priced against comparable x86-based instances, so each release tightens the screw on x86 cloud pricing and gives customers leverage to renegotiate or migrate off incumbent x86 fleets.
Third-order effects
- If the cadence holds — Graviton 2 to 3 to the later generations AWS went on to ship — in-house Arm silicon moves from niche alternative to default capacity inside a major cloud, shifting server-chip power from merchant vendors toward hyperscaler design teams.
The trend: Hyperscalers are turning custom Arm processors from a cost experiment into their primary compute platform, with each generation widening the gap over rented x86.