Hosting Platforms

In one line: For new web apps in 2026, the answer is almost always Vercel or Cloudflare. Move to AWS/GCP/Azure only when you've outgrown the platform tier or have a specific cloud-native requirement.

In plain English

"Hosting" is where your code runs. The 2026 hosting landscape splits into three tiers, roughly by how much you outsource vs. control:

• Edge platforms (Vercel, Cloudflare) — easiest. You push code; they run it globally. Best for new full-stack web apps.
• App platforms (Railway, Fly.io, Render) — middle ground. They run your containers / processes; you don't manage servers.
• Cloud providers (AWS, GCP, Azure) — most control, most operational burden. You manage the infrastructure. Standard at large companies.

Edge platforms (most popular for new apps)

Platform	Best for
Vercel	Best for Next.js. Premium pricing.
Cloudflare (Workers + Pages)	Fastest, cheapest, most global.
Netlify	Older, solid Jamstack pioneer.

App platforms

Platform	Best for
Railway	Simple, predictable pricing.
Render	Similar to Railway.
Fly.io	Globally distributed VMs, indie-developer-friendly.
Heroku	Original PaaS, declining but still used.

Cloud providers

Provider	Best for
AWS	Dominant at scale; 200+ services.
Google Cloud	Cloud Run is excellent.
Azure	Dominant in Microsoft-heavy enterprises.
DigitalOcean / Linode / Vultr / Hetzner	Simpler clouds, cheaper, fewer services.

Decision matrix

Project Type	Recommendation
Next.js indie/startup app	Vercel
Need edge globally + low cost	Cloudflare
Long-running Node/Python service	Railway / Render / Fly.io
Enterprise scale	AWS / GCP / Azure
Want one global server cheaply	Hetzner / DigitalOcean

Worked example: hosting tier per scale

Scale	Hosting
Side project, 1 user	Vercel free tier or Cloudflare Pages
Startup, 1K users	Vercel Pro or Cloudflare
Startup, 100K users	Vercel Enterprise, Railway/Fly.io, or moving to AWS
Series B, millions	AWS / GCP / Azure (probably K8s)
FAANG-scale	Custom internal platform on top of cloud primitives

Each tier solves a problem the previous one couldn't. Don't jump tiers prematurely — the operational burden grows faster than the value.

Highlight: vendor lock-in is overrated for early-stage projects

A common worry: "Vercel locks me in." It does. So does Cloudflare. So does AWS Lambda. So does Postgres-specific SQL.

For early-stage projects, ship something. Lock-in is a problem you'll have when you have customers complaining about pricing — which is a great problem to have. Most projects fail because they don't ship, not because they got locked into the wrong platform.

If you do hit scaling-cost issues later, the migration from Vercel to AWS is usually a 2–6 week project for a small team. That's a price worth paying for the year of velocity you got on the easier platform.

Common mistakes

Where people commonly trip up

• Jumping straight to AWS / GCP / Azure for a side project. You'll spend three weekends wiring VPCs and IAM roles instead of building your product. Edge platforms (Vercel, Cloudflare) cover 95% of new projects with one git push. Move to a raw cloud only when you've outgrown the platform tier or have a specific compliance/cost reason.
• Treating "serverless" as identical across platforms. Vercel functions, Cloudflare Workers, and AWS Lambda all have different runtimes, cold-start behaviors, and limits (Workers cap CPU but not wall clock; Vercel caps wall clock; Lambda has 15-min hard cap). Code that works on one can die on another — read the limits page before porting.
• Picking edge runtimes for everything. Edge runtimes are fast and globally distributed, but they're not Node — many npm packages (anything using fs, Buffer quirks, native bindings) won't run. Use edge for read-heavy, latency-sensitive routes; use a Node runtime for anything that needs the full ecosystem.
• Refusing to commit to a platform because of "lock-in." Lock-in is a problem you have when you have customers complaining about pricing — and migration from Vercel to AWS is typically a 2–6 week project. Velocity now > theoretical portability later.
• Putting your database on the opposite coast from your serverless functions. A US-East Lambda calling an EU-West Postgres adds 80–100ms of round-trip per query. Co-locate the DB with the primary compute region; use read replicas if you need global reads.

Page checkpoint

Checkpoint Quiz

Did hosting stick?

Required

What's next

→ Continue to DevOps & Infrastructure — the tools that run underneath hosting platforms (containers, K8s, IaC, secrets).