Amazon Redshift Pricing: Redshift Cost, AWS Redshift Serverless Pricing and RG Node Rates
Redshift bills four things at once: node hours, managed storage, scanned bytes and scaled capacity. Every rate below came out of the AWS Price List API this week, not out of another blog post.
Verified 22 August 2026 from the AWS Price List API, us-east-1 · nothing estimated · Last updated August 2026
Ask your data a question:
Writing SQL… Running (read‑only)… SQL Agentsql wrote
▋
Click a question. Agentsql writes the SQL, runs it read-only, and answers.
Direct answer
Amazon Redshift sells compute two ways. Provisioned clusters bill per node-hour: in us-east-1 the current Graviton RG generation runs $0.3801 for rg.large, $0.7602 for rg.xlarge, $3.04267 for rg.4xlarge and $9.128 for rg.12xlarge, while the older RA3 generation runs $0.543, $1.086, $3.26 and $13.04. Redshift Serverless instead bills $0.375 per RPU-hour, per second, with a 60 second minimum. Storage is $0.024 per GB-month on both, about $24.00 per TB. The single most valuable fact on this page is that every RG node is priced at exactly 70% of its vCPU-matched RA3 equivalent, and that RG's on-demand rate is identical to the price of a one-year No Upfront RA3 reservation, so the new generation gives you the commitment discount without the commitment. All figures read from the AWS Price List API on 22 August 2026.
›_ amazon redshift pricing, the published rates
RG node rates, and exactly how much less they cost than RA3.
us-east-1 (N. Virginia), on demand, read on 22 August 2026 from the AWS Price List API offer file published 14 August 2026. Monthly figures assume 730 hours.
| RG node | vCPU | Memory | Per hour | Per node per month | RA3 equivalent | Ratio |
|---|---|---|---|---|---|---|
| rg.large | 2 | 16 GiB | $0.3801 | $277.47 | ra3.large at $0.543 | 0.70000 |
| rg.xlarge | 4 | 32 GiB | $0.7602 | $554.95 | ra3.xlplus at $1.086 | 0.70000 |
| rg.4xlarge | 16 | 128 GiB | $3.04267 | $2,221.15 | ra3.4xlarge at $3.26 | 0.70000 per vCPU |
| rg.12xlarge | 48 | 384 GiB | $9.128 | $6,663.44 | ra3.16xlarge at $13.04 | 0.70000 |
AWS markets RG as "30% lower price per vCPU" than RA3. The price list shows that is not a marketing "up to". On all three vCPU-matched pairs the ratio is exactly 0.70000, and rg.4xlarge lands on the same $0.19017 per vCPU-hour once you adjust for its 16 vCPU against ra3.4xlarge's 12. AWS' own pricing page now reads: "Amazon Redshift offers different node types to accommodate your workloads, and we recommend choosing RG."
›_ redshift cost, read correctly
RG on demand costs exactly what a one-year RA3 reservation costs. Read this before you sign anything.
Three exact matches, all from the same AWS offer file. This is the finding that should change a purchase decision.
A one-year No Upfront reservation on RA3 buys a 30.0% discount in exchange for a twelve month commitment. An RG node is priced at 70% of RA3. Those are the same number, and the price list proves it to five decimal places on every size where the vCPU counts line up.
| Committed RA3 price | Rate | Uncommitted RG price | Rate | What it means |
|---|---|---|---|---|
| ra3.large, 1 year No Upfront | $0.38010 | rg.large on demand | $0.3801 | Same 2 vCPU and 16 GiB. Identical hourly rate, one with a 12 month contract and one without. |
| ra3.xlplus, 1 year No Upfront | $0.76020 | rg.xlarge on demand | $0.7602 | Same 4 vCPU and 32 GiB. RG also adds the built-in data lake engine RA3 does not have. |
| ra3.16xlarge, 1 year No Upfront | $9.12800 | rg.12xlarge on demand | $9.128 | Same 48 vCPU and 384 GiB. At this size the commitment being avoided is worth about $34,000 a year. |
There is a second reason the RG column is the better side of that table. AWS' pricing page states: "Note that Spectrum is not required for Redshift RG instances since they include a built-in data lake query engine." So an RG cluster also drops the $5.00 per TB scan charge that an RA3 cluster pays every time it queries Iceberg or Parquet in S3. If you scan 20 TB a month, that is another $100 a month gone, on top of the 30%.
›_ aws redshift pricing, every node type
Every provisioned node still purchasable in us-east-1.
RA3 and DC2 remain available. DS2 does not: in the current offer file ds2.xlarge and ds2.8xlarge appear only as concurrency-scaling meters, with no on-demand SKU behind them.
| Node | vCPU | Memory | Storage | Per hour | Per node per month |
|---|---|---|---|---|---|
| ra3.large | 2 | 16 GiB | 8 TB RMS | $0.543 | $396.39 |
| ra3.xlplus | 4 | 32 GiB | 32 TB RMS | $1.086 | $792.78 |
| ra3.4xlarge | 12 | 96 GiB | 64 TB RMS | $3.26 | $2,379.80 |
| ra3.16xlarge | 48 | 384 GiB | 64 TB RMS | $13.04 | $9,519.20 |
| dc2.large | 2 | 15 GiB | 160 GB SSD | $0.25 | $182.50 |
| dc2.8xlarge | 32 | 244 GiB | 2.56 TB SSD | $4.80 | $3,504.00 |
DC2 looks cheap and mostly is not. dc2.large is $0.25 an hour against rg.large at $0.3801, but it carries 160 GB of local SSD rather than managed storage, so the moment your data outgrows the node you are resizing the cluster to buy disk. RG and RA3 separate compute from storage, which is why AWS steers new clusters there. For the same separation on other platforms, see Snowflake pricing and BigQuery pricing.
›_ redshift serverless pricing
Serverless is $0.375 per RPU-hour, and that number is all-in.
The comparison most guides get wrong: the RPU-hour bundles things the node-hour bills separately.
An RPU is a Redshift Processing Unit, and AWS defines it as 16 GB of memory. Serverless bills per second with a 60 second minimum, does not charge for warehouse start up time, and lets you set base capacity anywhere from 4 to 1,024 RPUs. Above the base it scales on its own, and you can cap it with the MaxRPU setting or cap total spend with Max RPU-hours on a daily, weekly or monthly window.
The part that matters for any honest comparison is what AWS writes next: "You do not need to pay for concurrency scaling and Redshift Spectrum separately because they are both included with Amazon Redshift Serverless." So $0.375 is an all-in rate and an RA3 node-hour is not. Anyone lining the two up without adjusting for that is comparing different kinds of number, the same trap the classic and serverless columns set on Databricks pricing.
The arithmetic that decides it is utilization. At base 8 RPU, Serverless costs $3.00 an hour while it is awake. A two node rg.large cluster costs $0.7602 an hour and never sleeps, which is $554.95 a month. Serverless reaches that same $554.95 after about 185 hours, roughly six hours a day. Query less than that and Serverless is cheaper. Query more and the provisioned cluster is, by a widening margin: at 24 hours a day, base 8 RPU costs $2,190 a month, almost exactly four times the RG cluster.
›_ redshift reserved instances
The discount ladder, and why All Upfront is usually the wrong button.
Percentages derived from the AWS Price List API by amortizing each upfront quantity across the term. Identical on RG and RA3.
| Term and payment option | Provisioned nodes | Serverless RPUs | How to read it |
|---|---|---|---|
| 1 year, No Upfront | 30.0% | 20.0% | No cash down. On provisioned this is exactly the rate an RG node already charges on demand. |
| 1 year, Partial Upfront | 33.0% | not offered | Three points better than No Upfront in exchange for roughly half a year of spend up front. |
| 1 year, All Upfront | 34.0% | 24.0% | Four points better than No Upfront. Implied return on the prepaid cash is about 11.5% a year. |
| 3 year, No Upfront | 56.5% | 45.0% | The best discount available without writing a cheque. Usually the right default. |
| 3 year, Partial Upfront | 60.5% | not offered | Four points better than 3 year No Upfront. |
| 3 year, All Upfront | 62.5% | 50.0% | Six points better than 3 year No Upfront. Implied return about 10.4% a year on provisioned, 6.5% on serverless. |
Two things fall out of that table that a procurement conversation should turn on. The first is that prepaying is a financing decision, not a discount decision. A three year All Upfront ra3.large costs $5,348 today; the No Upfront equivalent costs $172.46 a month for 36 months, $6,208.65 in total. Solve that as an annuity and the prepayment is earning you about 10.4% a year on the cash. That is a real return, but it is not obviously better than what the same cash does inside most US businesses, and it is a long way from the "save up to 75%" framing that reservations usually get sold with.
The second is that committing to Serverless earns you materially less than committing to nodes. Three year All Upfront gets 62.5% off a provisioned node and 50.0% off an RPU, and the implied return on the serverless prepayment drops to about 6.5% a year. Serverless Reservations do have one structural advantage: they sit at the AWS payer account level and can be shared across multiple accounts, and any usage above the reserved RPU level simply bills at on-demand rates rather than failing.
›_ the meters nobody adds up
Compute is the headline. These five meters are where the invoice actually diverges from the forecast.
All us-east-1, read 22 August 2026.
| Meter | Rate | Unit | What to watch |
|---|---|---|---|
| Redshift Managed Storage (RMS) | $0.024 per GB-month | about $24.00 per TB-month | Identical on every RG node, every RA3 node and Serverless. Billed on data stored, independent of how many compute nodes you run. |
| Redshift Serverless compute | $0.375 per RPU-hour | 1 RPU = 16 GB memory | Per-second billing with a 60 second minimum. Base capacity is settable from 4 to 1,024 RPUs. No charge for warehouse start up time. |
| Redshift Spectrum / data scan | $5.00 per TB scanned | bytes scanned in S3 | Not charged on Serverless, which bundles it, and not charged on RG, which has a built-in data lake query engine. RA3 and DC2 pay it. |
| Concurrency scaling | per node-second | 1 free hour per cluster per day | ra3.large $0.00015, ra3.4xlarge $0.0009, ra3.16xlarge $0.00362, rg.large $0.00011, rg.12xlarge $0.00253, dc2.large $0.00007 per node-second. |
| Snapshot storage | $0.023 / $0.022 / $0.021 per GB-month | tiered downward | Charged only above the free allocation, and only on unique data blocks across all active snapshots rather than on nominal snapshot size. |
| Cross-region data sharing | $0.02 per GB | transferred | Applies to cross-region data sharing and to cross-region snapshot copies. |
Concurrency scaling deserves a sentence of its own, because it is widely assumed to be expensive and is not. Multiply the per-node-second rates by 3,600 and you get $0.5400 an hour for ra3.large against a $0.543 on-demand node, and $13.032 against $13.04 for ra3.16xlarge. Within half a percent. Scaled capacity is priced like ordinary capacity; the benefit is that it is metered per second and that every cluster earns up to one free hour a day, which AWS says covers 97% of customers. rg.large is the lone exception at 1.042x, an artefact of AWS rounding $0.0001056 up to $0.00011.
›_ redshift pricing calculator
What common warehouse shapes actually cost per month.
On-demand compute at 730 hours plus Redshift Managed Storage at $0.024 per GB-month. Reservations take 30.0% to 62.5% off the compute column.
| Warehouse shape | Compute | Storage | Total per month | Who this fits |
|---|---|---|---|---|
| 2 x rg.large, 1 TB stored | $554.95 | $24.00 | $578.95 | A first production warehouse. Smallest sensible RG cluster. |
| 2 x ra3.large, 1 TB stored | $792.78 | $24.00 | $816.78 | The same shape on the previous generation, $237.83 a month more for the same 4 vCPU. |
| 4 x rg.xlarge, 10 TB stored | $2,219.78 | $240.00 | $2,459.78 | A mid-size analytics team with real concurrency. |
| 2 x rg.4xlarge, 50 TB stored | $4,442.30 | $1,200.00 | $5,642.30 | Heavier workloads on larger data. |
| Serverless, base 8 RPU, 8 h a day | $730.00 | $24.00 | $754.00 | Bursty use. 8 RPU x $0.375 x 8 h x 30.4 days, with Spectrum and concurrency scaling included. |
| Serverless, base 8 RPU, 24 h a day | $2,190.00 | $24.00 | $2,214.00 | The warning case. Always-on serverless at 8 RPU costs four times a 2 x rg.large cluster. |
Row two is the cost of not migrating. The same 4 vCPU of capacity costs $237.83 a month more on RA3 than on RG, and nothing else about the cluster changes: AWS states RG maintains feature parity with RA3, and existing clusters can move with snapshot and restore, elastic resize or classic resize. Over a year that is $2,853.96 for a two node warehouse, and it scales linearly with the cluster.
›_ redshift free trial
The Redshift free trial changed, and most guides still describe the old one.
Serverless: $300 credit, 90 days
This is the current default. If you have never used Redshift Serverless, AWS grants a $300 credit with a 90 day expiration that covers both compute and storage. How fast it burns depends on your base capacity: at 8 RPU the credit funds 100 hours of active querying, at 128 RPU it funds about six. Set the base low before you start the clock.
Provisioned: two months, 750 hours, conditional
The familiar two month trial still exists but it is now a fallback. AWS offers it only "in regions where Amazon Redshift Serverless is not yet available". When it ends, or when usage passes 750 hours in a month, you either shut the cluster down or start paying on-demand rates. Any guide presenting this as the standard Redshift trial is out of date.
›_ honest read
Where Redshift pricing is genuinely good, and where it bites.
We build an AI data analyst, so weigh this accordingly. Every figure above is checkable against the AWS Price List API.
Nobody pays for a seat
Redshift bills capacity, not people. Give two hundred people read access and the cluster costs what it costs. That is the opposite of the seat model in Tableau pricing or Power BI pricing, where every assigned licence bills whether it is opened or not.
The rate card is unusually honest
Every number on this page came out of a public API with a publication date on it. Compare that with Looker pricing, where Google publishes no dollar figure for any platform edition, or Tableau, which stopped publishing role prices altogether.
Four meters, four ways to be surprised
Node hours, managed storage, scanned bytes and scaled seconds all bill independently, and only compute responds to a reservation. A team that reserves nodes and then triples its Spectrum scanning has not saved anything. RG removes one of those four meters, which is a quieter reason to move than the 30%.
Serverless can be the expensive option
Serverless is marketed as the cost-saving choice and it is, for intermittent workloads. Left running against a base capacity set too high, it is the most expensive way to buy Redshift. Base defaults matter more than the rate does, and MaxRPU plus a Max RPU-hours limit are the only real guardrails.
Worth saying plainly, because this page discusses Amazon Q generative SQL: Agentsql does not connect to Redshift. We connect read-only to Postgres, MySQL, Snowflake and BigQuery, turn a plain-English question into SQL, run it, and show the SQL every time so an analyst can check the logic. If your warehouse is Redshift, Amazon Q in Query Editor v2 is the built-in option to evaluate, and the rates above are what the warehouse underneath it costs. If your data already sits in one of our four engines, ours is a flat monthly price with no RPU meter attached, and you can read it on our pricing page.
Still comparing warehouses? The usage-metered side of the market is covered in BigQuery pricing, Snowflake pricing and Databricks pricing, the seat-based side in QuickSight pricing and Grafana pricing, and all of them side by side in the BI tools comparison.
›_ frequently asked
Amazon Redshift pricing questions, answered.
How much does Amazon Redshift cost?
A small production Redshift cluster starts at about $555 a month for compute: two rg.large nodes at $0.3801 an hour each, running continuously. Storage adds $0.024 per GB-month, so 1 TB is $24. Redshift Serverless instead bills $0.375 per RPU-hour with a 60 second minimum, which suits workloads that do not run all day. Rates are us-east-1, read 22 August 2026.
Is Amazon Redshift free?
No, but there is a free trial. If you have never used Redshift Serverless, AWS gives you a $300 credit that expires after 90 days and covers both compute and storage. In regions where Serverless is not yet available you can instead run a provisioned cluster free for two months, up to 750 hours a month. After that you pay standard rates or shut the cluster down.
What is an RPU in Redshift?
An RPU, or Redshift Processing Unit, is the capacity unit Redshift Serverless bills. Each RPU provides 16 GB of memory and costs $0.375 an hour in us-east-1. You set a base capacity anywhere from 4 to 1,024 RPUs, and Redshift scales above it automatically. Billing is per second with a 60 second minimum charge, and start up time is not billed.
What is the difference between RG and RA3 nodes in Redshift?
RG nodes are Graviton-based, launched 12 May 2026, and are priced at exactly 70% of the RA3 rate for the same vCPU count. They also include a built-in data lake query engine, so Spectrum scan charges do not apply. AWS now says on its own pricing page that it recommends choosing RG. RA3 remains available and feature-comparable, but costs more for the same compute.
How much do Redshift reserved instances save?
On provisioned nodes the AWS Price List gives 30.0% for one year No Upfront, 33.0% Partial Upfront and 34.0% All Upfront, then 56.5%, 60.5% and 62.5% for the three year equivalents. The percentages are identical on RG and RA3. Serverless Reservations save less: 20.0% and 24.0% for one year, 45.0% and 50.0% for three.
Is Redshift Serverless cheaper than a provisioned cluster?
Only if the warehouse is genuinely idle much of the time. Serverless at base 8 RPU costs $3.00 an hour while active, so it matches a two node rg.large cluster at about 185 hours a month, roughly six hours a day. Query more than that and provisioned wins. Serverless does bundle Spectrum and concurrency scaling, which provisioned RA3 bills separately.
How much does Redshift storage cost?
Redshift Managed Storage is $0.024 per GB-month in us-east-1, which works out to about $24.00 per TB-month. The rate is identical across every RG node, every RA3 node and Serverless, and it is billed on the data you store rather than on the number of compute nodes you run. Snapshot storage beyond the free allocation is billed separately at $0.023 per GB-month.
Is concurrency scaling free in Redshift?
Partly. Each cluster earns up to one hour of free concurrency scaling credits per day, which AWS states is sufficient for 97% of customers. Beyond that you pay a per-node-second rate. Multiplied out to an hour, that rate lands within about half a percent of the base node rate, so scaled capacity is priced like ordinary capacity, just metered per second.
How much is Redshift Spectrum?
$5.00 per terabyte of data scanned in S3. It applies to RA3 and DC2 clusters. It does not apply to Redshift Serverless, which bundles Spectrum into the RPU-hour, and it does not apply to RG nodes, because AWS states Spectrum is not required for RG instances since they include a built-in data lake query engine.
Can you pause Redshift to stop billing?
Yes, on provisioned clusters using on-demand pricing. AWS supports a pause and resume feature that suspends on-demand compute billing while a cluster is not in use. Managed storage keeps billing while paused, because the data is still there. Reserved instances continue to bill regardless, since you have committed to the term.
Is it worth paying All Upfront for a Redshift reservation?
Usually not. All Upfront beats No Upfront by only four percentage points over one year and six over three. Treating the prepayment as a loan, a three year All Upfront ra3.large costs $5,348 today instead of $172.46 a month, which solves to about 10.4% a year. If your business earns more than that on its cash, take No Upfront.
Is Redshift cheaper than Snowflake?
For steady always-on SQL workloads, usually yes, because a reserved RG node is cheap per hour and Redshift storage runs about $24.00 per TB-month against Snowflake at $23.00 per TB-month with far pricier compute. For spiky workloads Snowflake per-second credits can win. The honest answer depends on utilization, which is why both pages here publish the arithmetic.
›_ pricing and comparisons
›_ connect your database