Grafana vs Datadog vs New Relic Cost 2026: $1,945 to $5,600 a Month

I spent a morning reading observability pricing pages; every one of them was accurate while none of them told me what I would pay. That is the honest answer to "Grafana vs Datadog vs New Relic cost": the vendors meter different things (hosts, gigabytes, series, events, seats), so a price per unit means nothing until you push one real workload through all of them. This post does that. It explains the meter each platform uses, prices one 50-node Kubernetes cluster at list price on Datadog, New Relic, Grafana Cloud, Honeycomb and SigNoz, adds self-hosted LGTM and ClickHouse's ClickStack, and lists the line items that make bills grow faster than traffic. Every price comes from the vendor's page as I read it on 24 October 2026; discounts, which everybody negotiates, are not in here.

How do Datadog, New Relic, Grafana Cloud and Honeycomb charge?

Each platform picked a different unit, and the unit decides which of your habits get expensive.

Datadog charges per thing. Infrastructure is $15 per host per month on Pro billed annually ($18 month to month), APM is $31 per APM host, containers beyond the included allotment are $1 each, custom metrics are $5 per 100, and logs are $0.10 per GB to ingest plus $1.70 per million events to index for 15 days. Hosts are counted hourly and billed on the high-water mark of the lowest 99 percent of those hours.

New Relic charges per gigabyte and per person. The first 100 GB a month is free, then $0.40 per GB ($0.60 for Data Plus). Basic users are free, core users are $49, and full platform users are $349 a month on Pro with an annual commitment. Standard is cheaper ($10 for the first full user, $99 after) but caps you at five full users.

Grafana Cloud charges per active series and per gigabyte. Pro is a $19 platform fee that includes 10,000 series, 50 GB of logs, 50 GB of traces and three users; beyond that, metrics are $6.50 per 1,000 active series, logs and traces are $0.05 per GB processed plus $0.40 per GB written, and extra users are $8 each.

Honeycomb charges per event, and every span is an event. Pro starts at $150 a month for 50M events and tops out at 750M; Enterprise starts at 10 billion events a year. Seats are unlimited.

The open-source-backed services meter the raw resource instead. SigNoz Cloud is $0.30 per GB of logs or traces and $0.10 per million metric samples, with a $49 monthly minimum and no per-user pricing. Managed ClickStack, in beta since 4 February 2026, is priced on ClickHouse Cloud infrastructure: under three cents per GB-month of storage and under a cent per GB of ingest compute, with query compute billed by what you use and not published as a per-GB figure.

Platform	Main meters	Headline list prices	Free tier	Source
Datadog	Hosts, APM hosts, containers, custom metrics, GB and indexed events	$15/host, $31/APM host, $5 per 100 custom metrics, $0.10/GB + $1.70/M events	Not listed	Pricing list
New Relic	GB ingested, user seats	$0.40/GB after 100 GB; $49 core, $349 full user (Pro)	100 GB, 1 full user	Pricing
Grafana Cloud	Active series (and data points per minute), GB, active users	$6.50/1k series; $0.45/GB logs; $8/user	10k series, 50 GB logs and traces	Pricing
Honeycomb	Events (spans, log lines)	$150 per 50M events, Pro up to 750M	20M events	Pricing
SigNoz Cloud	GB, metric samples	$0.30/GB; $0.10/M samples	Trial	Pricing
Managed ClickStack	Storage and compute	Under $0.03/GB-month storage; query compute unpublished	$300 trial credit	ClickHouse blog

What does one 50-node Kubernetes cluster cost on each platform?

Here is the cluster: 50 nodes running 1,000 containers, tracing on every node, 100,000 active metric series of which 20,000 are custom application metrics, scraped every 30 seconds (the Prometheus Operator default), 1,000 GB of logs a month at roughly 1 KB per line (about a billion events), 500 GB of traces, and ten engineers who need access. For New Relic, which bills metrics by the gigabyte, I assumed metrics and events add another 300 GB. A month is 730 hours.

Line item, per month	Datadog (Pro, annual)	New Relic (Pro)	Grafana Cloud (Pro)	SigNoz Cloud
Hosts and containers	$1,500: 50 hosts at $15, 750 containers over the 250 included at $1	In GB	$0 with your own Prometheus	$0
Tracing, 500 GB	$1,550: 50 APM hosts at $31; inside the 7,500 GB span allotment	In GB	$205	$150
Metrics, 100k series at 30s	$750: 15,000 custom metrics over the 5,000 pooled	In GB	$1,235: 200k billable series	$876: 8.76B samples
Logs, 1,000 GB	$1,800: $100 ingest, $1,700 to index 15 days	In GB	$430	$300
Ingest		$680: 1,800 GB less 100 free
Users and platform fee	$0	$3,490: 10 full users	$75: $19 fee, 7 users at $8	$0
Total	$5,600	$4,170	$1,945	$1,326

Datadog's allotments come from its custom metrics (100 per Pro host, pooled), container (5 per Pro host) and APM (150 GB of ingested spans and 1M indexed spans per APM host) billing docs. Grafana's metrics line is explained below. Honeycomb does not get a column because it has no list price for this cluster: 500 million spans plus a billion log lines is 1.5 billion events a month, twice the Pro cap, so it is an Enterprise quote.

The shape matters more than the totals. Datadog's bill is mostly per-host fees and log indexing, New Relic's is mostly seats, and Grafana Cloud's and SigNoz's are mostly metrics. So the cheapest vendor for you depends on which of those you have a lot of. A team of three with heavy logs prices very differently from a team of forty with a quiet cluster.

Why is Datadog so expensive, and where do the surprises come from?

Datadog's list prices are not outrageous one at a time; the bill grows because each meter multiplies independently. A custom metric is a unique combination of metric name and tag values, host included, so adding a customer_id tag to one gauge can turn one metric into thousands, and histograms and distributions generate five custom metrics per combination by default. Containers beyond five per Pro host are billed hourly, so dense nodes quietly add a second per-host fee. APM is a separate per-host fee on the same nodes. And logs are cheap to ingest ($100 for our terabyte) and expensive to index ($1,700 for 15 days, $2,500 for 30).

Datadog prices ingested custom metrics at $0.10 per 100 against $5 per 100 indexed, so you can keep a tag at ingest and stop indexing it. Flex Logs stores events at $0.05 per million per month, but compute is a separate flat-rate tier and monitors do not run against Flex data, so it suits audit logs, not alerting. And skipping the annual commitment costs real money: the same cluster is about $6,340 billed month to month and about $7,195 at on-demand rates.

The best-documented Datadog bill is still Coinbase: $65 million for 2021 usage, followed by a migration team that built a Grafana, Prometheus and ClickHouse stack, double-wrote for months, and then stayed on Datadog after it restructured the contract. Customers keep paying: Datadog reported $1.12 billion of revenue in Q2 2026, up 36 percent, with about 4,720 customers above $100,000 a year.

Is New Relic cheaper than Datadog?

On data, yes, by a wide margin: $680 of ingest pays for the same telemetry that Datadog meters across five line items for $5,600. On people, it depends entirely on who needs which seat. Basic users are free and can run queries and use custom dashboards, but the curated APM and infrastructure UIs need a full platform user, and core users ($49) do not get those either. Ten full users at $349 is 84 percent of our bill. Four full users for whoever is on call plus six core users brings the total to $2,370, a split worth modelling before you sign.

New Relic also sells compute-based pricing with no user fees, measured in compute capacity units, but publishes no per-unit rate, so it is a sales conversation rather than something you can compare.

Grafana Cloud's hidden multiplier: the scrape interval

Grafana Cloud's metric price looks like a series price, but the meter is data points per minute. Pro includes 1 DPM per active series, and if your average is higher, you are billed on total DPM instead. A 30-second scrape is 2 DPM, so our 100,000 series bill as 200,000: $1,235 instead of $585. At 15 seconds it is 4 DPM and $2,535, and the total rises from $1,945 to $3,245. If you remote_write with the operator's 30-second default, as my kube-prometheus-stack guide does, this is the number to watch.

The other Grafana line items are friendlier. Usage is billed at the 95th percentile, so short spikes are forgiven. Logs are $0.05 processed plus $0.40 written per GB with 30 days of retention included, and retention beyond 30 days is charged per additional 30-day increment. The optional Kubernetes Monitoring app adds $7.20 per host and $0.50 per container a month beyond a small allowance, about $830 for this cluster. Enterprise starts at a $25,000 annual commitment, with metrics "as low as $3 per 1k series".

Is self-hosted Prometheus, Grafana and LGTM actually cheaper?

The licence is free and the storage is nearly free. Prometheus is Apache-2.0; Grafana, Loki, Mimir and Tempo are AGPL-3.0. Holding our 1.5 TB of logs and traces in S3 Standard at $0.023 per GB-month costs about $35 before compression. That number is real and misleading: the bill moves from the invoice to the nodes running ingesters and queriers and, mostly, to the people who upgrade them, size them and get paged when the compactor falls behind. I cannot give you a defensible figure for that time, and I distrust anyone who gives one without knowing your team.

What I can say is where the work is. Running the metrics half yourself is mature: kube-prometheus-stack is one Helm release, and the hard parts are cardinality and CRD upgrades, not the install. Running logs and traces at a terabyte a month is a larger commitment. In between sit the open-core services: SigNoz is MIT-licensed outside its enterprise directories, ClickStack pairs Apache-2.0 ClickHouse with the MIT-licensed HyperDX UI, and both let you self-host the same software you would otherwise rent. Grafana Labs' 2026 observability survey of 1,363 practitioners found cost was the top tool selection criterion for the third year running, at 65 percent, and that half of respondents now use SaaS in some form.

Whatever you choose, measure the inputs before you take a sales call. These queries work against any Prometheus, including the one from the companion guide:

# Measure what every vendor will price, against Prometheus v3.15 (port-forwarded to :9090)
PROM=http://localhost:9090
q() { curl -s "$PROM/api/v1/query" --data-urlencode "query=$1" | jq -r '.data.result[] | .value[1]'; }

# Active series: Grafana Cloud's unit, and roughly Datadog's custom-metric count for app metrics
q 'prometheus_tsdb_head_series'

# Samples per month (730 h): SigNoz's unit, and what drives New Relic's metric GB
q 'sum(rate(prometheus_tsdb_head_samples_appended_total[1h])) * 3600 * 730'

# Average data points per minute per series: above 1 means Grafana Cloud bills on DPM
q 'sum(rate(prometheus_tsdb_head_samples_appended_total[1h])) * 60 / sum(prometheus_tsdb_head_series)'

# The ten metric names with the most series: your custom-metric bill in waiting
curl -s "$PROM/api/v1/status/tsdb" | jq -r '.data.seriesCountByMetricName[] | "\(.value)\t\(.name)"'

# Log GB per month, if Fluent Bit ships your logs and Prometheus scrapes it
q 'sum(rate(fluentbit_output_proc_bytes_total[1d])) * 86400 * 30 / 1e9'

How do you cut an observability bill without switching vendors?

Most of the savings are pipeline work, which is why this is a platform engineering problem before it is a procurement one. Drop the labels nobody queries before they leave the cluster; the high-cardinality guide shows how to find them. Scrape infrastructure every 60 seconds. Index fewer logs, and route debug lines to cheap storage or nowhere. Sample traces at the tail, keeping errors and slow requests; Honeycomb does not count events dropped before ingestion. Give dashboard viewers free seats. And if you want a better contract, double-write to an alternative for a quarter, the way Coinbase did, because a credible exit is what moves a renewal. The longer version of the pipeline argument is in my post on OpenTelemetry and observability costs.

FAQ

Is Datadog more expensive than New Relic?

For the 50-node cluster in this post, at list price, yes: $5,600 a month against $4,170 with ten full New Relic users, or $2,370 with four full and six core users. Datadog's cost scales with hosts, containers and indexed logs; New Relic's scales with gigabytes and seats, so a large team with little data can reverse the result.

How much does Grafana Cloud cost per month?

The Pro plan starts at a $19 platform fee that includes 10,000 active series, 50 GB of logs, 50 GB of traces and three users. Beyond that, metrics are $6.50 per 1,000 series at one data point per minute, logs are $0.45 per GB with 30 days of retention, and users are $8 each. The cluster in this post comes to $1,945 a month at a 30-second scrape.

What is the cheapest observability platform for Kubernetes?

At list price for this cluster, SigNoz Cloud ($1,326) and Grafana Cloud ($1,945) are far below Datadog and New Relic, and self-hosted Prometheus with Grafana has no licence cost at all. The cheapest option in practice is the one whose meter matches your workload, after you have cut cardinality and log indexing, plus the engineering time you are willing to spend running it.

Does Honeycomb charge per user?

No. Honeycomb's plans include unlimited seats and charge by event volume, where each span counts as one event. The free tier covers 20 million events a month, Pro starts at $150 for 50 million and stops at 750 million, and above that you need an Enterprise contract.

This quarter, if I were choosing for a team of this size, I would run the five queries above against production, fill in the table with real numbers instead of mine, and get two quotes with a double-write plan written down. I would move infrastructure scrapes to 60 seconds and stop indexing debug logs before signing anything, because those two changes cut every vendor's bill, including the self-hosted one, and they are cheaper to make before the contract than after.

Related: Observability vs Monitoring: What's the Difference in 2026?, How to Fix Prometheus High Cardinality: Find and Drop Series (2026) and Observability in 2026: OpenTelemetry Graduated, the Collector Is Still v0.162, eBPF Instrumentation Is v0.14, and Datadog Bills $1.12B a Quarter.