Nobody signs off on an observability budget. You sign off on a per-host rate and a per-gigabyte rate, and then your services decide the rest. Ship a chatty debug statement into a hot code path, or let an autoscaler double your fleet overnight, and the invoice has already moved by the time anyone notices.
Datadog's own numbers show how normal large bills are: $1.12 billion in Q2 2026 revenue, growing 36 percent year over year, from roughly 4,720 accounts paying six figures a year each. This is a condensed version of a longer breakdown I published on DevToolLab, and if you want the per-platform detail with every source linked, the full article is here.
Why Switching Got Easier
Grafana Labs surveyed 1,363 engineers across 76 countries between October 2025 and January 2026, publishing the results in March 2026. Cost came out as the top tool-selection criterion for the third consecutive year at 65 percent, well clear of ease of use at 49 percent.
The more interesting figure is adoption of open standards. Prometheus investment sits at 77 percent, OpenTelemetry at 76 percent, and 65 percent of respondents are funding both. Once your services speak OTLP, the agent stops being the thing that traps you. Changing backends becomes an endpoint edit in a collector config, and every platform below will accept that traffic.
Pricing the Same Workload Five Ways
Comparison pages are written by vendors who choose the example. So pick a workload you actually run and push it through everyone's published rate card instead. Mine assumes 50 hosts, 3,000 GB of logs monthly, and 5 engineers who need full access.
// price.mjs - node price.mjs
const workload = { hosts: 50, logGBPerMonth: 3000, fullUsers: 5 }
const plans = {
// datadoghq.com/pricing, annual commitment
Datadog: ({ hosts, logGBPerMonth }) =>
hosts * 15 + hosts * 31 + logGBPerMonth * 0.1,
// newrelic.com/pricing, Standard edition, 100 GB ingest free
'New Relic': ({ logGBPerMonth, fullUsers }) =>
Math.max(0, logGBPerMonth - 100) * 0.4 + 10 + (fullUsers - 1) * 99,
// grafana.com/pricing, Pro: $19 base, 50 GB free, then process+write+retain
'Grafana Cloud': ({ logGBPerMonth }) =>
19 + Math.max(0, logGBPerMonth - 50) * (0.05 + 0.4 + 0.1),
// signoz.io/pricing, $0.30/GB logs and traces, $49 monthly minimum
SigNoz: ({ logGBPerMonth }) => Math.max(49, logGBPerMonth * 0.3),
// openobserve.ai/pricing, $0.50/GB ingested
OpenObserve: ({ logGBPerMonth }) => logGBPerMonth * 0.5,
}
const usd = (n) => '$' + n.toLocaleString('en-US', { maximumFractionDigits: 0 })
const rows = Object.entries(plans)
.map(([name, fn]) => ({ name, monthly: fn(workload) }))
.sort((a, b) => a.monthly - b.monthly)
const base = rows.find((r) => r.name === 'Datadog').monthly
for (const r of rows) {
const delta = ((r.monthly / base - 1) * 100).toFixed(0)
console.log(
r.name.padEnd(15),
usd(r.monthly).padStart(8) + '/mo',
usd(r.monthly * 12).padStart(10) + '/yr',
r.name === 'Datadog' ? 'baseline' : `${delta}% vs Datadog`,
)
}
What it prints:
SigNoz $900/mo $10,800/yr -65% vs Datadog
OpenObserve $1,500/mo $18,000/yr -42% vs Datadog
New Relic $1,566/mo $18,792/yr -40% vs Datadog
Grafana Cloud $1,642/mo $19,698/yr -37% vs Datadog
Datadog $2,600/mo $31,200/yr baseline
Two takeaways worth sitting with. First, the gap is two to three times, not the thirty times you see on landing pages. Second, look at where the Datadog money goes: APM at $31 per host is $1,550 of the $2,600, while log ingestion accounts for just $300. Teams that respond to a scary invoice by trimming log volume are attacking the smallest line on it. The expensive part is indexing, billed at $1.70 per million events, so 200 million indexed events quietly appends another $340 every month.
The Commercial Options
Grafana Cloud is the shortest migration if your team already lives in Grafana dashboards, since it is the same stack with someone else operating it. Its free tier is genuinely usable at 10,000 active series, 50 GB each of logs and traces, and 14-day retention for 3 users. Pro starts from a $19 platform fee. The tradeoff is forecasting: logs bill in three separate parts, at $0.05 per GB to process, $0.40 to write and $0.10 to retain, which is the least predictable model in this group.
New Relic trades per-host billing for per-gigabyte billing, and hands you 100 GB a month plus unlimited basic users at no cost, the most generous free allowance among the paid platforms. Data past that runs $0.40 per GB. Seats are where it bites: full platform access is $99 each beyond the first on Standard and $349 annually on Pro, so a 30-engineer team ends up paying more to let people log in than to store the telemetry.
Better Stack sells fixed bundles instead of meters, and folds uptime monitoring, on-call scheduling and status pages into the same subscription, so it can retire two other vendors alongside Datadog. Nano runs $30 monthly for 40 GB each of logs, traces and metrics, scaling through Micro at $120, Mega at $250 and Tera at $500 for 700 GB each, with roughly 16 percent off for annual billing. The ceiling is the catch: 700 GB per signal means the workload above does not fit any listed tier, and each bundle forces equal amounts of all three signal types regardless of what your traffic looks like.
Axiom is an event store rather than a monitoring suite, designed around retaining everything instead of sampling. Its free Personal tier is remarkable at 500 GB of loading per month with 30-day retention, and Axiom Cloud is $25 monthly including a terabyte. It also declines to publish a flat overage rate, describing discounts as automatic and sub-linear, which makes a heavy month genuinely hard to forecast. That is an awkward gap for a product being evaluated by people fleeing unpredictable bills.
The Open Source Options
SigNoz comes closest to a like-for-like swap. It is OpenTelemetry-native, runs on ClickHouse, and handles traces, metrics and logs in a single app. Licensing is open core rather than fully permissive: MIT for the bulk of it, with the ee/ and cmd/enterprise/ trees carved out separately. Version v0.141.1 landed on September 9, 2026, with roughly 32,000 GitHub stars. Cloud is $0.30 per GB for logs and traces at 15-day retention, minimum $49 monthly. Self-hosting hands you a ClickHouse cluster to operate, which at three terabytes a month is a real job for a real person.
OpenObserve is written in Rust and parks telemetry in object storage such as S3 rather than local disks, which is the architectural reason its storage costs undercut index-based systems. Cloud is $0.50 per GB ingested plus $0.01 per GB queried, with no per-host or per-seat charges at all, and self-hosting is free up to 50 GB daily with SSO and RBAC included. It is AGPL-3.0, so read the license carefully if you intend to offer a modified build as a service. Latest stable is v0.92.2 from August 17, 2026, with a v1.0.0 release candidate line running since late August.
VictoriaLogs is a component, not a platform, and that is the point. Single-node and cluster builds are both plain Apache 2.0 with no enterprise carve-out, making it the most permissive option here. The project documents up to 30x lower RAM and 15x lower disk usage than Elasticsearch and Grafana Loki, and it accepts data from OpenTelemetry, Datadog agents, Vector, Filebeat, Fluentd, Fluent Bit, Grafana Alloy and syslog. What you do not get is a trace UI, incident management or on-call tooling. You build the surrounding stack yourself. VictoriaLogs v1.52.0 shipped July 16, 2026.
The original post on DevToolLab gives each of these its own section with the free tier, the published rates and the specific thing it fails at.
Side by Side
| Platform | Model | Free tier | Entry price | Self-host | License |
|---|---|---|---|---|---|
| Datadog | Per host + per GB | 14-day trial | $15/host/month | No | Proprietary |
| Grafana Cloud | Per series + per GB | 50 GB logs, 3 users | $19/month + usage | Yes | AGPL-3.0 |
| New Relic | Per GB + per seat | 100 GB/month | $0.40/GB | No | Proprietary |
| Better Stack | Bundled tiers | 3 GB/month | $30/month | No | Proprietary |
| Axiom | Load + compute + storage | 500 GB/month | $25/month | No | Proprietary |
| SigNoz | Per GB | Community edition | $49/month minimum | Yes | MIT core |
| OpenObserve | Per GB | 50 GB/day self-hosted | $0.50/GB | Yes | AGPL-3.0 |
| VictoriaLogs | Self-hosted only | Unlimited | $0 license | Yes | Apache 2.0 |
How to Avoid Migrating Twice
- Start from your invoice, not a pricing page. Feed last month's real host count, ingest volume and seat count into the script above. Workload shape moves the ranking more than workload size does.
- Find out which line item is actually large. Cheaper log storage is worthless if 60 percent of your spend is per-host APM. Datadog breaks the bill out by product already, so you do not have to guess.
- Dual-ship before you commit. An OTLP exporter can target more than one endpoint, so run the candidate in parallel for a full billing cycle and compare two real invoices.
- Match retention to how you actually investigate. Defaults here cluster at 15 and 30 days. If your postmortems routinely look back a quarter, price the extended retention before comparing anything.
- Treat the license and the ops time as costs. Apache 2.0, MIT-with-carve-out and AGPL-3.0 commit you to different things, and saving $1,700 a month stops being a saving if it eats more than an engineer-week.
Which One Fits
If your team already builds Grafana dashboards, Grafana Cloud is the least disruptive move available, and the free logs allowance may cover a small production footprint on its own.
If hosts rather than data drive your bill, New Relic is where that 40 percent saving comes from, as long as you keep full platform seats to five or fewer.
If you are also paying separately for uptime checks and on-call rotation, Better Stack at $30 collapses three subscriptions into one, which beats shaving cents off a per-GB rate.
If you want out of seat and host billing altogether, SigNoz is the most complete open source replacement and has the lowest genuine entry point of any full platform listed.
If object storage is already part of your infrastructure, take VictoriaLogs for logs alone or OpenObserve for the full signal set in one binary. Apache 2.0 versus AGPL-3.0 is the deciding factor, not the feature grid.
Two Tools That Help During a Migration
Moving backends means proving your pipeline still produces the fields you depend on. The Grok Pattern Tester lets you validate %{SYNTAX:SEMANTIC} patterns against real log lines before a pipeline reaches production, and the Syslog Parser decodes RFC 5424 and RFC 3164 messages so you can confirm what a collector is genuinely forwarding rather than what you assume it is.
Wrapping Up
The thing that shifted in 2026 was not pricing. It was that 76 percent of surveyed teams now emit OpenTelemetry, which demotes the backend from a rewrite to a configuration change. Measured honestly, the saving on one workload lands between 37 and 65 percent, not the order of magnitude the marketing implies.
So before the renewal, work out which line is growing. Growing host count is a billing-model problem and you should shop. Growing indexed events is a retention-policy problem, and you might fix it without changing vendors at all.
Top comments (0)