In-memory at scale. Built for hot data.
Memory-Optimized plans pack up to 8 GB of RAM per vCPU on DDR5 ECC. Built for Redis, Memcached, in-memory analytics, and large-buffer Postgres / MongoDB workloads — starting at $22/mo.
Hold the working set
in actual memory.
When your dataset fits in RAM, p99 latency drops by 100×. Memory-Optimized lets you keep it that way — affordably.
DDR5 ECC memory
Latest-generation DDR5 with on-die ECC. Higher bandwidth, lower latency, and silent error correction across the lifetime of every VM.
Up to 1:8 vCPU:RAM
Pack up to 8 GB of RAM per vCPU. Right-sized for in-memory caches, large connection pools, and analytic engines that hate paging.
Sub-µs RAM latency
Every Memory-Optimized host runs on dual-socket NUMA-aware boards. The hypervisor pins guest memory to local channels for maximum throughput.
NVMe swap & spill
Pure NVMe block storage with 7 GB/s reads as a swap fallback. Postgres temp tables, RocksDB compactions, and Redis AOF rewrites stay fast.
20 TB bandwidth
Generous outbound allowance on every plan. Inbound traffic is always free. Overage billed at $1/TB.
DDoS protection — free
Enterprise-grade DDoS mitigation included on all plans at no extra charge. Filters up to 1 Tbps of volumetric attacks.
Pay for the RAM you actually use
Every plan includes DDR5 ECC memory, NUMA-aware scheduling, NVMe SSD storage, and 24/7 support.
Questions? Compare all plans →
Built for workloads that live in RAM.
Keep your hot set in RAM, where it belongs.
Run Redis, Memcached, KeyDB, or Dragonfly with multi-tens-of-GB working sets. Pinned NUMA-local pages mean p99 stays in single-digit milliseconds at every scale.
- Up to 256 GB DDR5 ECC per VM
- NUMA-aware scheduling on dual-socket hosts
- NVMe AOF persistence at 7 GB/s
Stop paging. Start responding.
Postgres shared_buffers, MySQL InnoDB buffer pool, MongoDB WiredTiger cache — they all want RAM. Memory-Optimized gives you the headroom without paying for cores you don't need.
- 1:8 vCPU-to-RAM ratio out of the box
- Predictable latency for buffered reads
- Snapshot-based PITR backups daily
Real-time aggregates, no GC pauses.
Elasticsearch, OpenSearch, ClickHouse, DuckDB, and Spark — JVM and columnar engines all benefit from large heaps. Run on ECC RAM and forget about silent corruption.
- ECC error correction across full lifecycle
- Predictable allocator behaviour at high heap
- Pair with object storage for cold tiers
Frequently asked questions
Hot data deserves hot memory.
DDR5 ECC memory, NVMe SSD. Cancel anytime.