AWS, Google, Microsoft and Oracle are all investing heavily in ARM-based servers, and Apple has moved its entire computing lineup to ARM. For anyone running cryptographic code in the cloud, that makes ARM performance a practical question.
Power and cost
Energy efficiency
ARM processors deliver exceptional performance per watt. Data centers consume 1-2% of global electricity, so this matters a great deal.
| Metric | x86 Server | ARM Server | Advantage |
|---|---|---|---|
| Typical TDP | 250W | 180W | 28% lower |
| Performance/Watt | 1.0x | 1.4x | 40% better |
| Annual Power Cost | $2,190 | $1,577 | $613 saved |
Based on 24/7 operation at $0.10/kWh
For organizations running thousands of servers, these savings add up to millions of dollars a year.
Licensing and cost
ARM's licensing model lets companies build custom silicon for their specific workloads. Amazon's Graviton, Apple's M-series and Ampere's Altra all show what that flexibility makes possible.
AWS Graviton3 instances typically cost 20-40% less than comparable x86 instances, while delivering equivalent or better performance for many workloads.
Cryptographic instructions on ARM
Modern ARM processors have dedicated hardware for cryptographic operations: AES instructions that run an encryption round in a single cycle, SHA instructions for hashing, and polynomial multiplication for fast authenticated encryption. With these, ARM processors match or exceed x86 performance on cryptographic workloads.
OpenCrypt benchmarks
We benchmarked OpenCrypt across platforms:
| Operation | Intel Xeon | AMD EPYC | Graviton3 | Apple M2 |
|---|---|---|---|---|
| AES-256-GCM | 4.5 GB/s | 4.8 GB/s | 4.2 GB/s | 4.6 GB/s |
| SHA-256 | 2.1 GB/s | 2.3 GB/s | 1.9 GB/s | 2.4 GB/s |
| ML-KEM Keygen | 42 μs | 38 μs | 45 μs | 35 μs |
| ML-DSA Sign | 89 μs | 82 μs | 95 μs | 78 μs |
ARM processors achieve competitive performance while consuming significantly less power.
ARM chips from the cloud providers
AWS Graviton
Amazon's Graviton processors have evolved quickly:
- Graviton1 (2018): first generation, modest performance
- Graviton2 (2020): 40% better price-performance than x86
- Graviton3 (2021): 25% faster than Graviton2, DDR5 memory
- Graviton4 (2024): 30% better compute performance
AWS reports that Graviton instances now power a significant portion of its internal infrastructure, including security-sensitive workloads.
Google Axion
Google announced Axion, its custom ARM processor, in 2024. It is built on ARM's Neoverse V2 design and targets both general compute and specialized workloads, including cryptographic operations.
Microsoft Cobalt
Microsoft's Cobalt 100 brings ARM to Azure, giving customers another option for running cryptographic workloads with improved efficiency.
Migrating to ARM
What works well
- Containerized workloads: Docker and Kubernetes run without issues.
- Modern languages: Go, Rust, Python and Node.js have excellent ARM support.
- Databases: PostgreSQL, MySQL and MongoDB all perform well.
- Cryptographic operations: native hardware acceleration is available.
Potential problems
- Legacy x86 dependencies: some older software may need updates.
- JIT compilers: make sure your runtime has ARM optimizations.
- Binary dependencies: native extensions have to be recompiled for ARM.
OpenCrypt on ARM
OpenCrypt detects ARM capabilities automatically and uses the optimal code path:
use opencrypt::aead::{AesGcm256, Aead};
// Same API on all platforms
let cipher = AesGcm256::new(&key);
let ciphertext = cipher.encrypt(&nonce, &plaintext, &aad)?;
// Automatically uses:
// - AES-NI on x86
// - AES/PMULL instructions on ARM
// - Portable fallback elsewhere
Moving between platforms requires no code changes.
AWS instance costs
For a typical web application handling TLS termination:
| Scenario | x86 Instance | ARM Instance | Monthly Savings |
|---|---|---|---|
| Small | c6i.large ($62/mo) | c7g.large ($44/mo) | $18 (29%) |
| Medium | c6i.xlarge ($124/mo) | c7g.xlarge ($88/mo) | $36 (29%) |
| Large | c6i.4xlarge ($496/mo) | c7g.4xlarge ($352/mo) | $144 (29%) |
AWS on-demand pricing, US East region
With hundreds of instances, migrating to ARM saves tens of thousands of dollars a month.
Security features
ARM processors have security features comparable to x86:
- Memory tagging: ARM MTE helps detect memory safety issues.
- Pointer authentication: PAC prevents certain exploit techniques.
- Speculation barriers: mitigations for Spectre-class vulnerabilities.
- TrustZone: hardware isolation for sensitive operations.
For cryptographic implementations, the key considerations are:
- Constant-time execution: ARM provides timing-safe instructions.
- Side-channel resistance: hardware crypto avoids cache-based leaks.
- Key isolation: TrustZone can protect cryptographic keys.
OpenCrypt's ARM implementation keeps the same security guarantees as our x86 code. It needs no configuration changes on ARM, and our CI/CD pipeline tests on both x86 and ARM to ensure consistent behavior and performance across platforms.
Industry analysts project that ARM will capture 30% of the server market by 2027, up from under 10% today.