Cloudflare Workers vs Vercel Edge vs AWS Lambda: pick edge functions 2026
Cloudflare Workers, Vercel Edge, or AWS Lambda — which edge functions platform actually fits small Indonesian dev teams in 2026.
Your user in Serpong just hit your API. The request bounced off a server in Virginia, waited 180ms for a cold container to boot, and finally returned JSON. You paid for that. Twice — in latency and in money.
Edge functions exist to fix exactly this. But “edge functions” in 2026 means three very different products: Cloudflare Workers, Vercel Edge Functions, and AWS Lambda (including Lambda@Edge and the newer Lambda Function URLs). Picking the wrong one for your team’s context is the kind of decision that quietly costs you six months of friction.
What actually differs between these three
The category label “serverless edge compute” hides real architecture differences.
Cloudflare Workers runs on the V8 isolate model — no containers, no cold starts. Every request starts in microseconds. The trade-off: you get a constrained runtime. No Node.js fs, no arbitrary npm packages that use native bindings, memory capped at 128MB per request. The platform uses its own KV, D1 (SQLite at the edge), and R2 storage — all good products, but they are a Cloudflare-specific stack.
Vercel Edge Functions is essentially a curated layer on top of Cloudflare Workers, scoped to Vercel’s deployment model. If you use Next.js middleware or Astro SSR on Vercel, you are already using Edge Functions without thinking about it. They inherit Workers’ cold-start-free speed but add Vercel’s routing, environment variable management, and their own billing layer on top.
AWS Lambda runs full container-based isolation. You get a real Node.js (or Python, Go, Java, etc.) runtime. Native modules work. Execution time up to 15 minutes. The flip side: cold starts. In ap-southeast-1, a typical Node.js Lambda cold start with a medium-sized package sits at 400–900ms. With Lambda SnapStart (Java) or provisioned concurrency (any runtime) you can eliminate this — but provisioned concurrency costs money even when idle.
Comparison table
| Dimension | Cloudflare Workers | Vercel Edge Functions | AWS Lambda |
|---|---|---|---|
| Cold start | None (V8 isolate) | None (Cloudflare underneath) | 400–900ms (mitigated with provisioned concurrency) |
| Max execution time | 30s CPU, 1min wall | 30s (same as Workers) | 15 minutes |
| Runtime | V8 / Web APIs only | V8 / Web APIs only | Full Node/Python/Go/Java/etc. |
| Free tier | 100k req/day, generous | Hobby plan, Vercel-tied | 1M req/month (+ API GW cost) |
| Pricing at scale | USD 0.30 / million req | Included in Vercel plan limits | USD 0.20 / million + API GW |
| Jakarta/SG latency | Excellent (PoP in JKT) | Excellent (same network) | Good (ap-southeast-1) |
| Lock-in | Medium (Cloudflare APIs) | High (Vercel platform) | Medium (AWS APIs) |
| Native module support | No | No | Yes |
| DX for Indonesian team | Excellent | Excellent if on Vercel | Steeper setup curve |
When Cloudflare Workers is the right pick
If your team is in the BSD City / Tangerang area building a SaaS product and you want to minimize infrastructure overhead, Workers is the clearest starting point. The free tier handles real traffic. Deployment with Wrangler CLI takes under five minutes. There is no cold start to explain to your team or to your client.
The constraint to internalize early: you cannot use most native npm packages. sharp for image processing, bcrypt (the C++ binding version), anything touching the filesystem — all out. Use bcryptjs (pure JS) instead of bcrypt. Use Cloudflare Images or an R2-based transform for image processing. The workarounds exist, but you need to know about them before you commit to the platform.
For teams running a lightweight API — auth tokens, rate limiting, routing, CRUD against D1 or an external Postgres via Hyperdrive — Workers is genuinely excellent. Response times under 40ms from Jakarta are not marketing copy; they are what you actually see in production.
When AWS Lambda makes more sense
Lambda’s advantage is the full runtime. A payment processing job that calls a vendor SDK with native bindings, processes a PDF, and writes to RDS — Lambda handles all of that. Workers cannot.
The DX cost is real, though. Setting up Lambda with API Gateway properly (routes, CORS, authorizers, IAM roles, VPC config if you need RDS access) takes a day minimum for someone who hasn’t done it before. SST v3 and the AWS CDK have improved this, but the abstraction is thinner than it looks until you hit an edge case at 2am.
Cold starts in ap-southeast-1 remain the concrete operational concern. A Node.js Lambda with a bundled SDK that touches 50MB on first boot will make users in South Jakarta visibly wait. Provisioned concurrency solves this but adds a fixed monthly cost. Budget roughly USD 10–20/month per function kept warm — acceptable for a production API, not ideal for a side project with uncertain traffic patterns.
Vercel Edge Functions: the non-choice
Vercel Edge Functions are not really a standalone decision. If you deploy on Vercel and use Next.js middleware, you are using them. They are the right tool for geolocation-based redirects, JWT validation at the edge, A/B test cookie injection — tasks that run before the framework render.
Do not architect a backend around Vercel Edge Functions as a standalone compute layer. The moment you need something outside Vercel’s ecosystem — a different framework, a self-hosted deploy, more control over routing — Edge Functions disappear from the equation. It is platform glue, not infrastructure.
Verdict
Pick Cloudflare Workers if:
- Your app is a standard CRUD API or middleware layer with no native module requirements
- You want zero cold start without paying for provisioned concurrency
- Your team is small and you want fast iteration without AWS configuration overhead
- Cost predictability matters — the pricing model is simple and the free tier is real
Pick AWS Lambda if:
- You need native Node modules, long-running jobs, or background processing over 30 seconds
- You are already invested in the AWS service mesh (RDS, SQS, S3 with IAM roles, etc.)
- Your org has AWS credits (common for Indonesian startups on AWS Activate)
- Provisioned concurrency cost is acceptable to eliminate cold starts
Use Vercel Edge Functions if:
- You are already deploying on Vercel and need middleware logic (geolocation, auth, rewrites)
- You do not need them as a standalone product — just as part of the Vercel workflow
The cleanest combination for a small Indonesian SaaS team with no dedicated DevOps: Cloudflare Workers for the API layer, a Fly.io or Railway container for anything requiring a full runtime, and AWS only when you need a specific service (SES for email, S3 for storage). Resist the pull to standardize everything on AWS prematurely — the operational overhead is real, and your senior dev will be the one absorbing it at 11pm.
Ditulis oleh Asti Larasati