Hosting cost is one of the more-opaque variables in web operations. Sticker prices are misleading; total cost includes egress, function invocations, edge requests, storage, and operational overhead. This article models true total cost across major hosting platforms (Vercel, Cloudflare, AWS, dedicated/VPS) for three common workload profiles. The numbers reveal breakeven thresholds where switching platforms produces real savings.
The workload profiles
Three workloads were modeled, representing common patterns:
- Workload A: mid-size SaaS application. 100K page views/day, 200K API requests/day, 50GB storage, 200GB monthly egress.
- Workload B: consumer content site. 1M page views/day, 100K API requests/day, 200GB storage (largely images), 2TB monthly egress.
- Workload C: enterprise application. 50K page views/day, 1M API requests/day, 500GB storage, 100GB monthly egress, longer-running compute tasks.
Vercel
Vercel's pricing structure includes function invocations, edge requests, and bandwidth. Its strength is developer experience and global edge deployment; its weakness is predictability at scale.
Workload A monthly cost (approximate): $400-800. Function invocations and edge requests fall within Pro tier limits; bandwidth approaches but doesn't exceed.
Workload B monthly cost: $1,500-3,000. Bandwidth becomes the dominant cost driver. Heavy image-serving workloads on Vercel become cost-inefficient quickly.
Workload C monthly cost: $600-1,200. API request volume is high but per-invocation cost is manageable; longer-running tasks may exceed function timeouts and require reworking.
Best fit: early-stage SaaS, marketing sites, content with modest egress. Worst fit: image-heavy sites at scale, applications with long-running compute requirements.
Cloudflare (Workers + R2 + Pages)
Cloudflare's edge-first architecture and aggressive pricing have made it the most-disruptive hosting story of recent years. Strength: extremely cheap egress (R2 has zero egress fees); weakness: vendor lock-in to specific runtime patterns.
Workload A monthly cost: $50-200. Worker invocations within free or near-free tier; R2 storage cheap; egress effectively free.
Workload B monthly cost: $100-300. The 2TB egress that costs $1,500+ on Vercel is essentially zero on R2. Storage cost dominates and is modest.
Workload C monthly cost: $200-500. Workers handle the API request volume well; longer-running tasks may exceed Worker CPU limits and require external compute.
Best fit: egress-heavy workloads, edge-deliverable applications, cost-sensitive operations. Worst fit: applications requiring long-running compute, complex stateful systems, or deep AWS-ecosystem dependencies.
AWS
AWS pricing is complex but predictable at scale. Strength: full operational flexibility, mature ecosystem, scales to extreme size; weakness: operational overhead is real, and small workloads pay disproportionately.
Workload A monthly cost: $200-500. EC2 instances or Fargate containers, RDS database, S3 storage, modest CloudFront. Operational overhead (someone managing the infrastructure) is the hidden cost.
Workload B monthly cost: $800-2,000. CloudFront egress is meaningful at 2TB; S3 storage is cheap. The operational cost of running production infrastructure on AWS competes with the platform-as-a-service options.
Workload C monthly cost: $1,000-2,500. Higher API request volume justifies AWS's flexibility; long-running tasks fit naturally in EC2/ECS without timeout constraints.
Best fit: mature engineering organizations, workloads with specific AWS service requirements, applications at large scale. Worst fit: small teams without dedicated infrastructure operations, simple applications where DX matters more than flexibility.
Dedicated / VPS (Hetzner, Linode, DigitalOcean)
Dedicated servers and VPS hosting are the cheapest option at scale and the most-operationally-demanding. Strength: extreme cost efficiency for predictable workloads; weakness: operational complexity and lack of managed services.
Workload A monthly cost: $50-150. Single Hetzner or Linode instance plus managed database. The savings are real if the team can operate the infrastructure.
Workload B monthly cost: $200-500. Larger instances or small cluster, plus CDN for global distribution. The egress savings versus PaaS options are substantial at this scale.
Workload C monthly cost: $300-800. Multiple instances for redundancy, managed database, monitoring. The cost is competitive with AWS but requires operational ownership.
Best fit: teams with operational maturity, predictable workloads, cost-sensitive deployments. Worst fit: teams without infrastructure operations capability, workloads requiring specific managed services.
The breakeven thresholds
Several patterns emerge from the cost modeling:
- Below ~$500/month total hosting cost, platform choice is dominated by DX. The cost differences are real but small relative to engineering time.
- Around 1TB monthly egress, Vercel becomes inefficient. The bandwidth pricing dominates total cost and creates strong incentive to consider Cloudflare or CDN-fronted alternatives.
- Around 50K monthly users, dedicated infrastructure becomes cost-effective if operational capability exists. The PaaS premium is most-justified at smaller scale.
- Around 500K monthly users, AWS or self-managed infrastructure becomes mandatory for cost reasons; the PaaS markup is too significant.
The hidden cost: operational overhead
The cost analysis above excludes operational time. A Cloudflare-hosted Worker that takes one engineer-hour per month to maintain has a different total cost than the same Worker that takes ten engineer-hours per month, even at the same monetary cost.
Operational overhead by platform (rough order):
- Vercel: very low operational overhead. Pay for it in the bill.
- Cloudflare: low operational overhead for fitting workloads. Higher when workload doesn't fit cleanly.
- AWS: moderate to high operational overhead. Requires someone who knows AWS.
- Dedicated/VPS: high operational overhead. Significant cost savings but requires operational expertise.
The full cost calculation includes both monetary and operational dimensions. Optimizing only on monetary cost frequently produces false savings.
The migration question
When does switching platforms make sense? The migration cost is real — typically 1-3 months of engineering time for a non-trivial application, plus ongoing risk during the transition. The savings need to be substantial enough to justify both the upfront cost and the operational risk.
A reasonable rule of thumb: migrate when the projected three-year cost savings exceed five times the migration engineering cost. Below that threshold, the savings rarely justify the disruption.
The takeaway
Hosting cost is more variable across platforms than most teams realize, and the right choice depends heavily on workload profile, team size, and operational maturity. The default of "we use [whichever platform we started on]" frequently leaves significant savings unclaimed at growth stages.
Run your actual workload through the cost calculators of two or three platforms periodically. The exercise often reveals optimization opportunities that weren't visible without the calculation.
Source notes
Cost models reflect published pricing from each platform as of January 2026. Workload profiles drawn from typical patterns documented in engineering blog publishing 2023-2025. Operational overhead estimates based on industry surveys (RightScale, Flexera) and aggregated DevOps reports.
