Datadog Monitoring: A Broke Founder's Guide (2026)

The Night I Almost Signed Up for Datadog

It was 2023. I was deep into Clickly — my URL shortener, the Bitly clone I’ve confessed to over-engineering before. The whole pitch of Clickly was serving millions of redirect events for almost nothing, so naturally my brain was obsessed with one question at 1 AM: how do I know when it breaks?

A redirect service that’s down is worse than useless — every dead link is a customer’s dead link. So I did what every engineer does. I opened a new tab and typed “datadog.” Everyone talks about it. It’s the Datadog dashboard you see in every startup’s screenshots — those beautiful graphs, the flame charts, the little green dots. I wanted the green dots. I wanted to feel like a real company.

I signed up for the trial. I installed the agent. And within an hour I had the most gorgeous dashboards I’d ever seen for a product with, let me check my notes… zero users.

Then I opened the pricing page. And I closed the tab. And I went to bed.

That night is the whole reason this post exists.

---

What “Monitoring” Actually Means (Nobody Tells You This)

Here’s the thing that confused me for way too long: “monitoring” isn’t one thing. Datadog sells it like one product, but it’s really five separate jobs stapled together, and you almost never need all five. (If you want the concepts on their own — metrics vs logs vs traces, structured logging, what never to log — I wrote that up separately in monitoring and logging, explained.) Understanding the difference is what saves you money — because you’ll stop paying for four things you don’t use.

• Uptime monitoring. The dumbest and most important one. Is my site up? Ping it every minute from somewhere in the world, text me when it’s down. That’s it.
• Error tracking. When your code throws an exception in production, you want the stack trace, the user, the browser — not a customer emailing “it’s broken.” This is Sentry’s whole job. (Its free tier is plenty for a small SaaS — I broke down Sentry’s pricing and when you’d ever pay.)
• Logs. The console.logs of your server, collected somewhere you can search them. Cheap to produce, shockingly expensive to store at scale.
• Metrics. Numbers over time — CPU, memory, request rate, how many redirects Clickly served this hour. The pretty line graphs.
• APM & traces. Application Performance Monitoring. This follows a single request through every function and database call to find the slow one. It’s genuinely magic. It’s also the most expensive thing on the menu.

Datadog does all five, brilliantly. That’s the trap. You come for uptime — the free-est, simplest thing — and you leave paying for distributed tracing on a product nobody’s using yet.

---

How Datadog’s Pricing Actually Works (The Part That Bit Me)

Datadog doesn’t have a price. It has a menu, and every line is billed differently. As of 2026, here’s the shape of it — I verified these against their pricing page before writing them down, because these numbers move:

• Infrastructure monitoring: about $15 per host, per month, if you commit annually. Pay on-demand and it’s closer to $18, because on-demand runs 20–50% more than the annual commitment across the board.
• APM: roughly another $31 per host — and here’s the catch nobody mentions: you can’t buy APM alone. Every APM host needs a paired infrastructure plan. So the real cost of “I just want tracing” is more like $46 per host, per month.
• Logs: $0.10 per GB just to ingest them, and then another ~$1.70 per million events to actually index them so you can search — at a stingy 15-day retention.

Two words will hurt you here: per host. And two more: high-water mark. Datadog counts your hosts hourly and bills you on roughly the 99th-percentile count for the month. Translation: if your app autoscales to 40 containers for three hours during a traffic spike, you can get billed as if you ran 40 hosts all month. Your worst three hours set your bill.

And logs are where founders actually get murdered. A single service with debug logging accidentally left on, or a bot hammering your API, can dump terabytes in a day. You pay per GB ingested — the meter doesn’t care that the logs were garbage. People don’t set up exclusion filters until after the first surprise invoice. That’s not a bug in Datadog; it’s just what usage-based pricing does to someone who isn’t watching.

For context: a fully-instrumented Datadog setup for a modest 20-person SaaS startup realistically runs $2,000–$3,500 a month. I was a team of one with no revenue. Do the math on that.

---

The $65 Million Wake-Up Call

If you ever need to feel better about closing that pricing tab, remember this story. In 2021, one Datadog customer ran up an annual bill of $65 million. Sixty-five. Million. Dollars. For one year of observability.

For a long time it was a mystery which company it was, until Datadog mentioned an unnamed customer on an earnings call and the internet did its detective thing. The answer, as reported by The Pragmatic Engineer: it was Coinbase. During the crypto boom after their IPO, nobody was watching infra costs, and the meter just… ran. When the market cooled and they had to cut, one of the first things they did was spin up a dedicated team to move off Datadog and onto a self-hosted Grafana / Prometheus / ClickHouse stack.

Sit with that. A company paying $65M/year decided the right move was open-source tooling. Not because they were broke — because the pricing model stops making sense at scale, and it never made sense at zero. If Coinbase is fleeing to Grafana, what exactly am I, with my zero-user URL shortener, doing signing up for the same bill structure?

That’s the lesson. Datadog isn’t a bad product. It’s an incredible product, priced for a company that has already won. You haven’t won yet. Neither had I.

---

What a Broke Founder Actually Needs

Let me be blunt, the way I wish someone had been with me at 1 AM in 2023. When you have zero to a hundred users, your monitoring needs are almost embarrassingly small:

1. Tell me when the site is down. (Uptime.)
2. Tell me when my code crashes, with the stack trace. (Errors.)
3. Let me grep my server logs when I’m confused. (Logs, occasionally.)

That’s it. You do not need distributed tracing. You do not need custom metric dashboards with 40 widgets. You do not need APM following a request across seventeen microservices, because you don’t have seventeen microservices — you have one Railway box and a Vercel deploy. Every hour you spend tuning a Datadog dashboard is an hour you didn’t spend on the only thing that matters, which is getting anyone to use the thing at all.

The uncomfortable truth: pretty dashboards are procrastination in a lab coat. They feel like real founder work. They are not. Watching a graph of a metric nobody’s generating is just a very expensive screensaver.

---

My $0 Monitoring Stack

So here’s what I actually do instead. Every piece has a real free tier — not a 14-day-then-we-bill-you trial, an actually-free tier. I verified all of these limits before publishing, because free tiers shrink over time. (Want a fuller shopping list, including paid tools that are still cheaper than Datadog? See the Datadog alternatives compared.)

Job	Tool	Free tier (2026)
Uptime	Better Stack	10 monitors, 3-min checks, hosted status page — free forever
Uptime (more monitors)	UptimeRobot	50 monitors, 5-min checks (personal use)
Errors	Sentry	Generous free tier, enough for a small SaaS
Errors (if you already use it)	PostHog	Error tracking bundled with your analytics
Logs + metrics	Grafana Cloud	10k metric series, 50 GB logs, 50 GB traces, 14-day retention — free, no card
All-in-one	New Relic	100 GB/month ingest, 1 full user — perpetual free tier

Let me walk through how I actually wire it.

Uptime: start here, it’s free and it’s 90% of the value

Better Stack gives you 10 monitors, checks every 3 minutes, phone/text alerts, and a genuinely nice hosted status page on the free plan. UptimeRobot is the old faithful — 50 monitors on the free tier, though it’s technically personal-use only. Either one tells you the single most important thing: is my product up right now? If you set up nothing else, set up this. It took me ten minutes and it’s caught more real problems than any dashboard ever did.

Errors: Sentry, or PostHog if you’re already there

Sentry’s free tier is enough for a bootstrapper, full stop. It catches the exception, shows you the stack trace and the user, and shuts up until the next real problem. If you’re already running PostHog for analytics — and if you’ve read my $0 analytics stack post, you probably are — it does error tracking too. Why pay for two tools when one covers it? That’s the whole Broken Engineer religion in one sentence.

Logs & metrics: Grafana Cloud’s free tier is absurdly generous

This is the part that makes the Datadog pricing feel almost rude. Grafana Cloud’s free tier hands you 10,000 active metric series, 50 GB of logs, and 50 GB of traces every month, with 14-day retention, three users, and no credit card required. That is more headroom than most bootstrapped SaaS products will touch for a year. Datadog wants $0.10/GB just to ingest logs; Grafana gives you 50 GB free. Same graphs. Same flame charts. One of them just isn’t trying to bill you on your worst three hours. (Grafana not quite your speed? I rounded up the alternatives worth knowing.)

New Relic is the other big free lunch — 100 GB of ingest a month and one full platform user, perpetually free. If you want everything under one login and don’t want to assemble pieces, that’s a real option.

If you love self-hosting (know the risk first)

If you’re the kind of engineer who enjoys running your own infra, SigNoz is the open-source Datadog alternative worth knowing — built on OpenTelemetry, it does logs, traces, and metrics in one app, and you can self-host it for the price of a small VPS. Their cloud plan starts at $49/month, but the community edition is free.

I’ll say the same thing I say about self-hosting your own email server: it’s wonderful, it’s educational, and it’s a nightmare factory if your actual goal is to make money. Self-host your observability if it’s a hobby. If it’s a business, let the free tiers do the work and get back to shipping.

---

When Datadog Actually Makes Sense

I want to be fair, because I respect the tool. There is a version of you for whom Datadog is exactly right: you’re past product-market fit, you’re a team of 10+, you’re running real infrastructure with real revenue, and an outage costs you more per hour than Datadog costs per month. At that point the single-pane-of-glass, the correlation across metrics/logs/traces, the maturity of the product — it’s worth every dollar. Companies pay Datadog a fortune because it’s good, not because they’re foolish.

But that’s a future-you problem. The mistake isn’t using Datadog. The mistake is using Datadog’s pricing model to watch a product that hasn’t earned a rupee. You’ll know when you’ve outgrown the free stack — your bill will still be $0 and you’ll be annoyed by a limit, which means you finally have enough traffic to hit one. That’s a fantastic problem. Have it first.

---

The Bottom Line

I never did pay Datadog. Clickly never got the flame charts. And you know what? Not once did I sit there wishing I’d spent $2,000 a month watching graphs of traffic I didn’t have. What I actually needed was a text message when the site went down and a stack trace when it crashed — and both of those are free.

Monitoring should be a smoke detector, not a mission-control room. A smoke detector costs almost nothing, runs quietly in the background, and screams exactly when it matters. That’s the whole job. The day you genuinely outgrow the free tiers is the day you should be celebrating, not the day you started. If you’re still adding up every line of your monthly burn — I broke down what it really costs to run a SaaS, and observability doesn’t have to be on that list at all. And whatever you do, on Vercel or anywhere else, set a billing cap before something with a meter runs away from you.

This is the Broken Engineer Guide — I over-engineer everything, fail at business, and hand you the receipts so you don’t repeat my 1 AM mistakes. Go watch your uptime for free, then go build something worth monitoring.