Where Should You Host Your App? Hosting Providers Compared

Episode Summary

Executive Summary: The episode is a broad comparison of web hosting options, explaining the technical differences between serverless, server-full, and edge hosting while evaluating major providers like Vercel, Render, Cloudflare, Netlify, Fly.io, Railway, Heroku, DigitalOcean, Linode/Akamai, Begin, Flight Control, and Deno Deploy. The hosts focus on tradeoffs in pricing, scaling, CI/CD, cron jobs, deployment previews, and developer experience, while sharing personal usage stories and recommendations.

Main Topics: Hosting models: serverless, server-full, and edge: The hosts define the core infrastructure types: serverless functions that spin up on demand, server-full long-running servers/VMs, and edge functions that run closer to users with limited APIs and low-latency execution. CDNs, CI/CD, cron jobs, and scaling: They explain adjacent infrastructure concepts that modern hosting providers bundle in: CDNs for asset delivery, Git-based continuous deployment, scheduled cron tasks, preview environments, and auto-scaling behavior under traffic spikes. Major cloud providers as the foundation: The episode frames AWS, Azure, Google Cloud, and other large infrastructure vendors as the base layer under many smaller hosting products, with boutique providers often wrapping or simplifying these systems. Provider-by-provider comparisons: Each major platform is assessed on workflow, pricing, reliability, and technical fit, including Flight Control, Render, Vercel, Begin, Heroku, DigitalOcean, Linode/Akamai, Netlify, Fly.io, Railway, Cloudflare, and Deno Deploy. Pricing and billing models: A major discussion covers the differences between wall time billing, CPU-time billing, and box-based pricing, plus bandwidth costs and the way team-seat pricing affects SaaS-style hosting bills. Developer experience and support quality: The hosts repeatedly compare onboarding, deploy previews, logs, UI clarity, and support responsiveness, arguing that ease of use and good feedback loops often matter more than raw infrastructure features. Personal hosting preferences and war stories: Both hosts share real-world examples of production incidents, migrations, and favorite setups, including Render memory failures, Vercel/Netlify satisfaction, Cloudflare Worker limits, and a strong preference for automation and low-friction deploys.

Key Arguments: Serverless is ideal for bursty or infrequent workloads because code only runs when needed, reducing idle cost and automatically scaling to demand. Server-full hosting remains valuable when you want full Node.js behavior, long-running processes, or more control over the runtime and infrastructure. Edge hosting is attractive for latency-sensitive work and request interception, but limited APIs and stricter execution budgets make it unsuitable for some apps. Modern hosting is no longer just about servers; CDNs, previews, Git integration, cron jobs, and deploy tooling are now core product features. Wall-time billing rewards waiting-heavy functions less, while CPU-time billing can be much cheaper for apps that spend lots of time awaiting network responses. Some providers excel at certain app types: Render for Node apps, Vercel for Next.js/SvelteKit, Netlify for static/Jamstack workflows, Cloudflare for edge-heavy apps, and Fly.io for globally distributed machines. User experience, support speed, and deployment feedback can be decisive factors because debugging hosted apps is slow and painful compared to local development. The hosts favor providers that reduce operational overhead, but they acknowledge that cheaper or more flexible options often require more self-management.

Data Points: Hosting providers named: 12 - Providers discussed include Flight Control, Render, Vercel, Begin, Heroku, DigitalOcean, Linode, Netlify, Fly.io, Railway, Cloudflare, and Deno Deploy. Big cloud vendors listed: 5+ - The hosts cite AWS, Azure, Google Cloud Platform, IBM, Alibaba, Oracle, Salesforce, Rackspace, and Vultr as foundational infrastructure vendors. Render app Node version incident: Node 18 vs Node 20 - Wes’s Render app broke after a redeploy/move because package.json allowed any version greater than or equal to 18, so it ran against Node 20 unexpectedly. Render outage timing: 3-minute crash loop - His app spiked memory and crashed repeatedly every few minutes while Render tried to rebuild/move the service. Render failover setup: 2 instances - The production app avoided downtime because two instances were running and failing at different times. Vercel base plan execution: 1,000 gigabyte-hours - The hosts mention Vercel includes a large amount of execution on the base plan. Vercel team pricing: $20 per seat - They note Vercel charges by seat, which can become expensive for larger teams. Cloudflare/Deno worker execution limit: 10 ms CPU time on free plan - They describe the free plan CPU budget as extremely tight for actual processing. Cloudflare paid plan CPU limit: 50 ms CPU time - They mention the paid plan increases CPU time limits but still requires very fast code. Cloudflare setup time: 200 ms - Workers get a brief wall-time budget for initialization/imports before request execution. Begin quick runtime concept: Arc - Begin is described as built on the Arc open standard. Cloudflare storage product: R2 - Mentioned as Cloudflare’s S3-style object storage offering. Cloudflare database product: D1 - Discussed as Cloudflare’s SQLite-based relational database product. Railway free credit: $5 - Railway is noted to give new users five dollars free. Flight Control pricing threshold: $50 or $100 AMS spend - Flight Control charges based on usage of AWS-managed services (AMS) rather than hosting itself. Heroku acquisition price: $212 million - The hosts cite Salesforce’s purchase of Heroku. DigitalOcean initial droplet price: $5/month - They recall DigitalOcean’s famous low-cost SSD VPS offering. Akamai/Linode region count: 16 actual server locations plus ~20 gateways - The hosts estimate Fly-like global placement for Akamai/Linode after discussing its edge strategy. Railway security incident year: 2021 - Wes references a Railway incident where secrets may have been exposed.

Pivotal Quotes: "“You are only paying for the time that serverless function is running.”" — Wes Boss: Explaining wall-time billing for serverless functions. "“It’s a tripping hazard where people are like, all right, let’s do this.”" — Wes Boss: Describing inconsistent local vs deployment behavior on Deno Deploy and similar platforms. "“I know lots of major, major companies that will use it.”" — Scott Delinski: Discussing Akamai/Linode as an enterprise infrastructure choice after the Linode rebrand.

Implications: Listeners should choose hosting based on app shape, team size, and operational tolerance—not brand loyalty. The market favors specialized platforms, but hidden costs, API limits, and runtime quirks can matter as much as performance.