Upstash vs AWS ElastiCache: Serverless Redis Pricing and Performance in 2026

Upstash Redis and AWS ElastiCache both provide managed Redis. Almost everything else is different, e.g. how you connect or the pricing structure.

In this guide I wanna do an up-to-date 2026 comparison with current numbers and a live throughput run to see how both perform.

What is the difference between Upstash Redis and AWS ElastiCache?

ElastiCache is AWS hosting the Redis OSS or Valkey engine on nodes inside your VPC (virtual private cloud, AWS's private network), billed per node-hour or per metered serverless usage. Upstash is a serverless Redis-compatible database with its own engine, reachable from anywhere over HTTPS or TCP.

With ElastiCache we pick node types, replica counts, subnets, and security groups, and your clients must run inside (or peer with) the VPC. With Upstash we create a database and get a URL and a token and can just start.

How much throughput can Upstash Redis handle?

Upstash Redis handles 16,000 commands per second on the Fixed 100GB and 500GB plans.

Enterprise plans go beyond 100,000 commands per second with dedicated resources.

I checked the limits from a Node.js client in the same AWS region as the database. Sequential PING commands came back with a median of 1.54 ms. Normal GETs with 32 concurrent workers ran at 4,735 requests per second with a median of 5.89 ms.

Pipelined SETs through the REST pipeline endpoint, 100 commands per request and 16 in flight, sustained 16,471 commands per second:

PING (20 sequential):  p50=1.54ms
Plain GET (32 workers, 10s):  47365 requests = 4735.75 req/sec, p50=5.89ms p99=14.32ms
Pipelined SET (batch 100, 16 in flight, 25s):  412700 commands = 16471.82 commands/sec
Rate limits: 0  Request errors: 0  Command errors: 0

The same pipelined load pushed from the Virginia client against a second database in Frankfurt held above 10,300 commands per second across 680,000 commands, again with zero 429 errors. This is really nice.

In every run, requests past the plan limit waited a bit longer instead of failing.

What does Upstash Redis cost in 2026?

Upstash Redis has a free tier (256 MB, 500,000 commands per month), pay-as-you-go at $0.20 per 100,000 commands, Fixed plans from $10 to $1,500 per month, and custom-priced Enterprise plans. An idle database on pay-as-you-go costs $0.

The Fixed plans, which bill a flat monthly price with unlimited commands:

Pay-as-you-go adds $0.25 per GB of storage after the first free GB, and bandwidth is free up to 200 GB a month, then $0.03 per GB. For production guarantees, the Prod Pack add-on ($200 per month per database) brings a 99.99% uptime SLA (service level agreement), multi-zone high availability, encryption at rest, SOC-2 report access, and Prometheus, Grafana, and Datadog integration.

Enterprise adds the custom limits above plus VPC peering, HIPAA, and SSO.

What does AWS ElastiCache cost in 2026?

ElastiCache bills per node-hour, or on Serverless per GB-hour of storage plus compute units called ECPUs. The cheapest node is a cache.t4g.micro running Valkey at $0.0128 per hour, about $9.34 per month. Serverless meters a minimum of $6.13 per month on Valkey and $91.25 on Redis OSS, even when idle.

The common on-demand nodes in us-east-1:

AWS prices Valkey 20% below Redis OSS on nodes, and about a third below on Serverless. On Serverless, each kilobyte transferred consumes one ECPU: a read of 3.2 KB costs 3.2 ECPUs. The idle minimums come from minimum metered storage: Redis OSS meters at least 1 GB, so 1 GB × $0.125 per GB-hour × 730 hours = $91.25 a month for an empty cache.

The quiet costs add up too. Replicas bill as full nodes. Cross-AZ traffic costs $0.01 per GiB in each direction, backups cost $0.085 per GiB-month, and running an end-of-life Redis OSS version adds an extended support surcharge of 80%, rising to 160% in year three.

Reserved nodes discount up to 55% with all-upfront payment. Some cluster examples are in our full ElastiCache pricing breakdown.

Which is cheaper for a small or idle workload?

Upstash. An idle Upstash database costs $0 on pay-as-you-go, and 5 million commands a month cost $10. Every ElastiCache option bills a minimum whether you send one command or none:

The free tiers differ in kind. Upstash's free tier is permanent: 256 MB and 500,000 commands a month. AWS accounts created after July 15, 2025 get $100 in credits plus up to $100 more for activating other AWS services, and the credits run out.

Which is cheaper at high, steady traffic?

It depends on where the traffic runs. ElastiCache node pricing ignores command volume, so an in-VPC app pushing 100 million commands a month through one t4g.micro pays $9.34. An Upstash Fixed plan is flat too: $20 a month covers unlimited commands at up to 10,000 per second.

Clients already inside a VPC is ElastiCache node pricing's best case. If you're already on AWS infrastructure, I think ElastiCache is very good option.

But once the workload is higher or unpredictable, 50,000 requests per second on ElastiCache Serverless would be $915 to $1,359 a month, while Upstash covers 16,000 commands per second and 10 TB of bandwidth for a flat $800 and moves past that on Enterprise.

Can you reach ElastiCache from a serverless or edge function?

No. ElastiCache has no public endpoint. Clients connect over TCP from inside the VPC. Edge runtimes like Vercel or Cloudflare Workers don't allow TCP connections, so they can't connect to ElastiCache directly. Upstash uses plain HTTPS through its REST API, which works from Workers, Vercel Edge, and browsers (Upstash also supports TCP btw).

The REST API is also request-based rather than connection-based, so AWS Lambda functions skip connection pool management entirely. Zapier ran into this in production: their rate limiting ran in Vercel edge middleware, which couldn't open TCP connections to their existing Redis instance, so they moved global rate limiting and malicious IP blocking to Upstash over HTTP.

How does global replication compare?

Upstash replicates to read regions you pick from 18 regions across AWS and GCP, with no fixed limit on how many. ElastiCache Global Datastore is active-passive and allows at most two secondary regions, so writes go through one primary region.

In my tests i got reads under 1 ms and writes under 5 ms from the same region at the 99th percentile, and reads and writes under 50 ms from the same continent. You can run the read test yourself at latency.upstash.com. Each read region costs half the Fixed plan's monthly price, and on pay-as-you-go a write bills once per region it replicates to, so 100,000 writes with one read region cost $0.40.

Where does ElastiCache win?

On network cost inside AWS, and AWS-native tooling. Same-AZ traffic between EC2 and ElastiCache is free, and metrics and access control go through CloudWatch and IAM.

If you're on AWS already with a majority of products, I think it makes sense to stay on AWS for Redis/Valkey, too.

Which should you pick?

ElastiCache is better when everything runs inside one AWS VPC and you have someone to manage nodes. Upstash is better if you're using another runtime to connect, want a better developer experience or are using agents to write your code (see our Upstash Redis skill!s). Both Upstash and ElastiCache handle extremely high throughput and low latency.

Supabase compared ElastiCache, Redis Cloud, and DynamoDB before picking Upstash for its global edge caching, global replication, the REST API, and predictable cost at high bandwidth.

If your workload looks like that, low-latency reads from many regions and clients that run outside a VPC, the same reasoning applies. If you prefer one busy backend in us-east-1, a reserved ElastiCache node is good too.