ElastiCache Serverless cost, engine by engine
ElastiCache Serverless drops the node-sizing exercise entirely: there is nothing to provision and no per-hour instance line. That makes the pricing simpler but easy to mis-estimate, because the engine you pick changes both the rate and the minimum you pay. Here is exactly what each engine costs in 2026 and where the real gaps are.
The two axes you actually pay for
With a node-based cluster you pay per node-hour for whatever instance type you provision, full or empty. Serverless throws that model out. You pay for two consumption meters and nothing else:
Data stored, measured in gigabyte-hours. If your cache holds 10 GB for a month, that is 10 GB times ~730 hours, or 7,300 GB-hours, whatever the traffic. Requests, measured in ElastiCache Processing Units (ECPUs). A simple GET or SET moving up to 1 KB consumes 1 ECPU; larger payloads and multi-key commands consume proportionally more. There is no instance charge underneath, so an idle Serverless cache costs only its stored data plus the small minimum.
That structure is why engine choice matters more here than on nodes. The rate per GB-hour and per ECPU is set by the engine, and so is the floor you pay before you have stored anything meaningful.
ElastiCache Serverless rates by engine (us-east-1, 2026)
| Serverless meter | Valkey | Redis OSS | Memcached |
|---|---|---|---|
| Data stored, per GB-hour | $0.084 | $0.125 | $0.125 |
| Compute, per million ECPUs | $0.0023 | $0.00340 | $0.00340 |
| Minimum billed storage per cache | 100 MB | 1 GB | 1 GB |
| Near-idle dev-cache floor / month | ~$6 | ~$91 | ~$91 |
| Per-node / per-hour charge | None | None | None |
| Cheapest Serverless engine | Yes | No | No |
Sources: AWS ElastiCache pricing (Serverless section, us-east-1) and AWS: Announcing ElastiCache for Valkey (33% lower, 90% lower minimum storage), checked July 2026. Monthly floors are computed at ~730 hours; verify region pricing before procurement.
Worked example: a 10 GB cache at 2,000 ops/sec
Take a cache holding 10 GB and serving a steady 2,000 operations per second, each moving about 1 KB (so 1 ECPU per op), over a 730-hour month. The arithmetic is entirely from the published rates above:
Storage: 10 GB times 730 hours is 7,300 GB-hours. On Valkey that is 7,300 times $0.084, or about $613/month; on Redis OSS or Memcached it is 7,300 times $0.125, or about $913/month.
Requests: 2,000 ops/sec times 3,600 seconds times 730 hours is about 5,256 million ECPUs. On Valkey that is 5,256 times $0.0023, or about $12/month; on Redis OSS or Memcached, 5,256 times $0.00340, or about $18/month.
Total: roughly $625/month on Valkey versus $931/month on Redis OSS or Memcached, a saving of about $306, or 33%, for the same workload. Storage dominates the bill here, which is typical: unless your traffic is extreme relative to your working set, the GB-hour rate is what moves the number, and that is exactly where Valkey is cheapest. This is a model over AWS's published rates, not a quote; your real bill depends on your actual stored bytes and request mix.
Where Memcached stops being the cheap option
On node-based ElastiCache, Memcached often wins on cost-per-usable-cache: it runs no replica, so every gigabyte of a Memcached node is cache you can use, while a Redis or Valkey node spends half its billed memory on a replica you also pay for. That logic vanishes on Serverless. There is no node and no replica line item, so the only thing that varies is the per-unit rate, and there Memcached matches Redis OSS at $0.125/GB-hour and $0.00340 per million ECPUs, both 33% above Valkey.
So the ranking flips. On Serverless, Valkey is the cheapest engine for any given workload, Redis OSS and Memcached tie for second, and the old rule of thumb that Memcached is the budget cache no longer holds. If you were reaching for Memcached Serverless specifically to save money, Valkey does the job for a third less. For the node-based version of this trade-off, see the ElastiCache Redis vs Memcached cost breakdown.
Serverless or a node-based cluster?
Serverless is not automatically cheaper than provisioning your own nodes; it is cheaper for a different shape of workload. It removes capacity planning, scales in and out instantly with no downtime, and bills only what you consume, which is ideal for spiky, bursty, or unpredictable traffic and for dev and test caches, where Valkey's 100 MB floor lands you at about $6/month instead of provisioning a node that sits mostly idle.
For a steady 24x7 workload with a known working set, a reserved node-based cluster is usually cheaper per stored gigabyte, because a one- or three-year reservation buys fixed capacity at a lower effective rate than paying per GB-hour on demand. The Valkey engine keeps its roughly 20% node-price advantage there too. The break-even turns on your peak-to-average ratio: the spikier and more idle-prone the cache, the more Serverless wins; the flatter and fuller it runs, the sooner a reserved node pays off. For the node-side numbers and the engine-freeze caveat (Redis OSS on ElastiCache is stuck at 7.1), see ElastiCache: Valkey vs Redis OSS.
FAQ
How is Amazon ElastiCache Serverless priced?
ElastiCache Serverless has no node to size. You pay on two axes: data stored, billed in gigabyte-hours (GB-hours), and requests, billed in ElastiCache Processing Units (ECPUs), where a simple read or write of up to 1 KB consumes 1 ECPU. There is no per-node or per-hour instance charge and no capacity to provision. On the Valkey engine, storage is $0.084 per GB-hour and compute is $0.0023 per million ECPUs (AWS ElastiCache pricing, us-east-1, checked July 2026).
Is Valkey cheaper than Redis OSS on ElastiCache Serverless?
Yes, by 33% on both billing axes. ElastiCache Serverless for Valkey is priced at $0.084 per GB-hour stored and $0.0023 per million ECPUs, while Redis OSS and Memcached are $0.125 per GB-hour and $0.00340 per million ECPUs (AWS ElastiCache pricing, checked July 2026). AWS states the Valkey engine is 33% lower than other supported engines on Serverless. Valkey also carries a 100 MB minimum billed storage per cache versus 1 GB for Redis OSS and Memcached, which matters most for small and dev caches.
What does a small ElastiCache Serverless cache cost per month?
At the storage floor with negligible traffic, a Valkey Serverless cache bills its 100 MB minimum: 0.1 GB times $0.084 per GB-hour times ~730 hours is about $6 per month, which matches the as-low-as-$6 figure AWS quotes. A Redis OSS or Memcached Serverless cache bills a 1 GB minimum at $0.125 per GB-hour, about $91 per month for the same near-idle cache. The gap comes from both the 33% lower rate and the 10x lower storage minimum on Valkey.
Is Memcached cheaper than Redis on ElastiCache Serverless?
No. On ElastiCache Serverless, Memcached is priced identically to Redis OSS ($0.125 per GB-hour and $0.00340 per million ECPUs) and carries the same 1 GB minimum, so Valkey is the cheapest of the three Serverless engines outright. This inverts the node-based picture, where a Memcached node can deliver more usable cache per dollar because it spends no memory on a replica. On Serverless there is no node and no replica to pay for, so Memcached's structural cost advantage disappears and its higher per-unit rate makes it the most expensive engine for a given workload.
Should I use ElastiCache Serverless or a node-based cluster?
Serverless removes capacity planning and scales instantly, and it is the cheapest option for spiky, unpredictable, or dev workloads, especially on Valkey with its $6/month floor. For a steady 24x7 workload with a known working set, a reserved node-based cluster is usually cheaper per stored gigabyte because you commit to fixed capacity at a lower effective rate. The break-even depends on how full and how busy the cache stays; model both against your peak-to-average ratio before committing.