CloudWatch vs Datadog for startups: what you actually need
Both tools work. The question is what 'work' costs you, and whether you're paying for features you'll use in the next 12 months.
What CloudWatch actually is
CloudWatch is AWS's built-in monitoring service. It collects metrics from every AWS service automatically, stores logs, fires alarms, and lets you query logs with Logs Insights. It's tightly integrated by default — no agent to install, no SDK to add.
Side-by-side comparison
| Feature | CloudWatch | Datadog |
|---|---|---|
| Zero-config AWS metrics | ✓ Every AWS service out of the box | ○ Requires agent or integration setup |
| Log storage + querying | ✓ Logs Insights included | ✓ More powerful with ML/pattern detection |
| Threshold-based alarms | ✓ On any metric, SNS notifications | ✓ Plus anomaly-based and forecast alerts |
| Anomaly detection | ✗ Fixed thresholds only | ✓ Baseline-aware alerting |
| Distributed tracing / APM | ✗ Not available | ✓ Full request tracing across services and DBs |
| Cross-service correlation | ✗ Manual — compare timestamps across tabs | ✓ Linked automatically in the same view |
| Dashboard UX | ○ Functional but slow to build | ✓ Fast and intuitive |
| Uptime synthetics | ✗ Not built-in | ✓ Multi-region monitoring |
| Service dependency maps | ✗ Not available | ✓ Visual graphs of service-to-service calls |
| Log → trace correlation | ✗ Not available | ✓ Jump from a log line to its trace |
| AWS low-level metrics | ✓ Native, most detailed | ○ Some require CloudWatch as a fallback |
| Lambda cold-start overhead | ✓ No added latency | ○ Datadog layer adds a small cold-start penalty |
| Agent required | ✓ No agent — AWS emits metrics automatically | ✗ Agent required on every host |
| Cost (small startup) | ~$60–100/month | ~$250–350/month (5 hosts + APM) |
What Datadog actually is
Datadog is a purpose-built observability platform. It requires an agent running on your infrastructure (or Lambda layers, ECS Fargate agents, etc.) to collect metrics. It has integrations for hundreds of services and a much richer UI than CloudWatch.
Cost comparison
| Line item | CloudWatch | Datadog (Pro) |
|---|---|---|
| Base AWS metrics | Free (standard resolution) | ~$15–23/host/month |
| APM / distributed tracing | Not available | ~$31/host/month additional |
| Log ingestion (20 GB/month) | ~$10 ($0.50/GB) | ~$20 ($0.10/GB + storage) |
| Log storage | $0.03/GB/month | Included in plan up to retention limit |
| Alarms (30) | ~$3/month ($0.10 each, first 10 free) | Included |
| Container / host metrics | $0.35/instance/month (Container Insights) | Included in host plan |
| Custom metrics | $0.30/metric/month | Included up to plan limit |
| Uptime synthetics | Not available | ~$5 per 10,000 API test runs |
| Estimated total — 5 hosts | ~$60–100/month | ~$250–350/month |
| Estimated total — 20 hosts | ~$100–160/month | ~$700–1,000/month |
What you actually need at each stage
| Stage | Tool | Est. cost | What to set up |
|---|---|---|---|
| 1–5 engineers, pre-PMF | CloudWatch only | $30–80/month | Alarms on 3–5 key metrics (CPU, memory, error rate, latency), Container Insights for ECS, Logs Insights for manual investigation |
| 5–15 engineers, post-PMF | CloudWatch alarms + Datadog APM | $200–400/month | Keep CloudWatch for alarms; add Datadog only if multi-service latency issues can't be diagnosed from logs alone |
| 15+ engineers, SLA commitments | Datadog full suite — or evaluate Honeycomb / Grafana Cloud | $500–1,000+/month | Choose based on specific gaps (tracing depth, cost model, team familiarity), not demos |
The feature neither tool gives you out of the box
Both CloudWatch and Datadog tell you what is wrong. Neither automatically tells you why — that correlation step (metrics + logs + recent deploys + dependency health) is where engineers spend most incident investigation time.
A team with a consistent 5-step investigation process using CloudWatch will outperform a team with Datadog but no process. The tool choice matters, but it matters less than the discipline to actually investigate systematically.
The recommendation
| Team size | Start here | Add this when… |
|---|---|---|
| Under 5 engineers | CloudWatch only — invest in your investigation process, not tooling | You're consistently unable to diagnose latency issues from logs alone |
| 5–15 engineers | CloudWatch alarms + Datadog APM (~$200–400/month) | Multiple services are calling each other and tracing would save material debug time |
| 15+ engineers | Evaluate Datadog, Honeycomb, or Grafana Cloud on specific gaps | You have SLA reporting requirements or cross-team dashboard sharing needs |
The trap to avoid: buying Datadog because the demo looks good, not because you have a specific observability problem it solves that CloudWatch doesn't. Start with the problem, then find the tool.
Frequently asked questions
Should a startup with fewer than 5 engineers use CloudWatch or Datadog?
CloudWatch only. At 1–5 engineers pre-PMF, the $250–350/month Datadog cost rarely solves a problem you actually have. CloudWatch covers threshold-based alarms, log storage, Logs Insights queries, and zero-config AWS metrics. The exception: if multi-service latency debugging from logs alone consistently costs more than 30 minutes per incident — then Datadog APM pays for itself.
How much does Datadog cost for a small AWS startup?
Approximately $250–350/month for 5 hosts with APM, compared to $60–100/month for CloudWatch alone. Datadog bills on multiple dimensions: hosts ($15–23/host/month), APM hosts ($31/host/month additional), log ingestion ($0.10/GB), and synthetics. Bills often grow 20–30% beyond initial estimates as you scale — budget a buffer before committing.
What does CloudWatch not do that Datadog does?
CloudWatch lacks: distributed tracing / APM, anomaly detection (fixed thresholds only), cross-service correlation in a single view, service dependency maps, log-to-trace correlation, and uptime synthetics. If you need to debug why a specific user request is slow across 3 microservices, CloudWatch requires manual correlation across tabs; Datadog shows it in one view.
Is AWS CloudWatch free?
CloudWatch has a free tier: 10 alarm metrics, 1 million API requests, 5 GB of log data ingestion, and 3 GB of log archive per month. Beyond the free tier, costs are approximately $0.10/alarm/month, $0.50/GB log ingestion, and $0.35/instance/month for Container Insights. A realistic small startup with Container Insights and 30 alarms pays $60–100/month.
When should I switch from CloudWatch to Datadog?
Switch when you have a specific problem CloudWatch doesn't solve: multiple services calling each other where log-based debugging takes more than 30 minutes per incident (add Datadog APM), or a need for anomaly-based alerts that adjust to traffic patterns. Don't switch because the Datadog demo looks good — start with the problem, then find the tool.
Related reading
Still debugging incidents manually?
ConvOps does this automatically — root cause in under 60 seconds, delivered to WhatsApp or Slack.