AWS introduces its own custom-designed Arm server processor, AWS Graviton Processor, and claims 45% lower costs for some workloads
The little ecommerce company has been beavering away. Tweets: James Wang / @jwangark : 6/ Vertical integration strikes again—Amazon launches its own custom ARM server CPUs for AWS. Every cloud vendor now has a custom chip strategy: Amazon-ARM, Microsoft-FPGA, Google-TPU. Chip-ISA, no matter how dominant, is no longer a moat. http://www.geekwire.com/... Alex Schleber / @alexschleber : Custom silicon the mega trend of the next decade...? There are significant #infosec implications too... cf. Apple http://twitter.com/... Owen Williams / @ow : Did....Amazon just get into the CPU manufacturing game? http://www.cnbc.com/... Andy Jassy / @ajassy : New A1 instances are 1st to be powered by custom #AWS Graviton processors, based on @Arm architecture. Excited to pass cost savings back to customers, reducing them by up to 45% for scale-out workloads (like microservices & web servers): http://www.cnbc.com/.... #reInvent Nitin Borwankar / @nitin : I think AWS is following the Amazon strategy of having a billion SKUs http://twitter.com/...
Context & Ripple Effects
AWS entering its own CPU design in late 2018 was the moment cloud vertical integration reached the processor itself: as one observer quoted in the coverage put it, every major cloud vendor now runs a custom silicon strategy — Amazon on Arm, Microsoft on FPGAs, Google with TPU — and a dominant instruction-set architecture no longer functions as a moat. The claim attached at launch was concrete: up to 45% lower cost than comparable x86 instances for scale-out workloads.
The bet compounded annually from there. AWS followed with Graviton2 in 2019, then successive generations, until by the time of Graviton5 in 2025 the company said Arm chips accounted for more than half of all new AWS CPU capacity — meaning this first chip was the opening move in flipping the default server architecture of the largest cloud.
First-order effects
- AWS customers running scale-out workloads gain an immediately cheaper instance option priced up to 45% below comparable x86 offerings, making Arm a real choice rather than an experiment.
- Intel's x86 franchise loses its assumed position inside AWS: the largest cloud buyer can now route general-purpose compute demand to its own silicon instead.
Second-order effects
- Rival clouds face pricing pressure on commodity compute, since AWS can now set instance prices against its own chip economics rather than a supplier's margin — pushing Microsoft and Google to lean harder on their own custom hardware plays.
- Server-software ecosystems must treat Arm as a first-class target, because skipping it means ignoring the fastest-growing slice of AWS capacity.
Third-order effects
- If the pattern holds — and the later generations suggest it did — cloud compute consolidates around vertically integrated providers whose cost floor is set by their own chip roadmaps, structurally weakening merchant CPU vendors and making instruction-set choice a procurement detail rather than a strategic lock-in.
The trend: Cloud providers are absorbing the server CPU into their own vertical stacks, turning custom silicon from differentiator into table stakes.