This is part two of a five part series on the practical AI and web stack for a small or medium business. Part one covered where your AI should run. This part covers where your website and apps should live, and how to choose without painting yourself into a corner you have to pay a developer to get out of.
The good news is that for most small businesses, hosting is now nearly free and genuinely fast. The trap is picking a platform for a feature you love today and discovering later that the same feature is the thing keeping you from leaving. So the real question is not "which host is best," it is "which host fits this specific job, and what does it cost me to change my mind."
Prices below were rechecked against all three vendors on 6 August 2026. If you want the fuller cost breakdown, including exactly what each platform meters and which line item generates a surprise bill, that is what hosting a new business site actually costs in 2026. This post is the decision framework and the lock-in discussion.
The three platforms, and what each is actually for
Cloudflare is the cheap, fast default for a static site or a small app. You can serve your pages straight off its global network for free, and requests to those static files are not billed at all. A dynamic app runs on Workers, with a free tier of 100,000 requests a day at 10 milliseconds of CPU per invocation, and a paid tier starting at $5 a month that includes 10 million requests and 30 million CPU milliseconds, then $0.30 per additional million requests. Note what is being metered there: your code executing, not your pages being served. Its object storage, R2, charges nothing for bandwidth out, which matters if you serve a lot of images or downloads. Cloudflare now tells you to start new projects on Workers. For a brochure site, a menu, a simple booking form, or an internal tool, this is hard to beat on price and speed. The full tour is in the Cloudflare developer platform.
Vercel is the premium experience for teams building on the Next.js framework. You get a preview link for every change, fast rebuilds, and polished tooling. The Hobby tier is free and covers 1 million edge requests, 100 GB of data transfer, and 1 million function invocations a month. Pro is $20 per user per month and includes a $20 monthly usage credit, lifting those ceilings to 10 million edge requests and 1 TB of transfer. If your business has a developer who lives in Next.js and ships a content-heavy, frequently-changing site, that experience is worth paying for. Just note that the seat count multiplies: three people who need to deploy is $60 before any usage.
Netlify is the simplest path for a marketing site. Connect a repository, set a build command, and you get deploy previews and niceties like Forms, which capture submissions for free with no backend code. Since its April 2026 pricing change it runs on a credit model rather than discrete resource limits, and there are now four tiers: Free at $0 for 300 credits a month, Personal at $9 for 1,000 credits, Pro at $20 for 3,000 credits with three or more concurrent builds and unlimited team members, and Enterprise on custom terms.
What burns credits is worth knowing before you pick a tier, because the unit is not what people expect: a production deploy costs 15 credits regardless of how small the change was, bandwidth costs 20 credits per GB, compute is 10 credits per GB-hour, and web requests are 2 credits per 10,000. At the Pro pack rate a credit is roughly two thirds of a cent.
Do that arithmetic and the free tier's shape becomes clear. 300 credits is about twenty production deploys if you spend them on nothing else. A site that changes weekly sits comfortably inside it; a site under active daily development will exhaust it on deploys alone, which is precisely the gap the $9 Personal tier fills. For a small team that wants to think about infrastructure as little as possible, Pro is a comfortable home, and unlimited team members is the reason it wins the many-client-sites case.
Matching the platform to the job
Here is how it shakes out for the kinds of businesses I work with, using the actual shape of the work.
- A solo realtor or a single-location service business with a brochure site. This is mostly static pages plus a contact form. Cloudflare for near-zero cost and speed, or Netlify if you want Forms to handle submissions with zero code. Either is the right call. Do not reach for Vercel here; you would be paying for a framework you do not need.
- A business with a real app: bookings, a customer portal, a quote tool. Now you have dynamic logic and maybe a database. Cloudflare Workers handles this cheaply if you or your developer are comfortable writing the backend directly. If the app is built on Next.js and changes constantly, Vercel's developer experience earns its $20 a seat.
- An agency or operator running many small client sites. Netlify's flat $20-a-month-with-unlimited-seats plan and simple per-site deploys fit the many-small-sites pattern well, and Cloudflare's free tier is excellent for the static ones. This is also where keeping a repeatable, portable setup pays off, because you are doing the same thing dozens of times.
The lock-in nobody quotes you
Every platform creates lock-in through the features you will most want to use. The skill is knowing where, and keeping it out of the parts that are expensive to rebuild.
- Cloudflare's deepest lock is its stateful pieces: Durable Objects, Workflows, and the bindings model. Your plain code is portable, but if your core logic lives inside those, moving is a re-architecture, not a copy.
- Vercel's lock was historically Next-specific features. That has eased a lot: Next.js now ships a stable adapter that other hosts use, so leaving Vercel is a known procedure rather than a rewrite.
- Netlify's lock is its conveniences: Forms, the redirects and headers configuration files, and the Blobs storage API. Pleasant to use, but they do not transfer to another host as-is.
None of this means avoid these platforms. It means do not let one vendor's proprietary feature own the part of your business that would be painful to rebuild.
A better way to keep your options open
A few habits do almost all of the work, and none of them slow you down today.
- Keep your content portable. Write pages as Markdown or plain HTML in a folder you own, not trapped inside one platform's proprietary page builder. Moving hosts then means re-pointing a deploy, not re-typing your site.
- Keep your real data in a database you can move. Your own Postgres on a provider like Neon or Supabase, plus S3-compatible storage, travels with you. Use a host's proprietary database for what it is uniquely good at, not as the system of record for your business.
- Own the domain at a neutral registrar. Your domain is your address. Keep it somewhere independent of your host so switching hosts never holds your traffic hostage.
- Write down how it is built. Even a one-page note on the build command, the environment variables, and where the data lives turns "we can never touch this, the person who set it up is gone" into a Tuesday afternoon.
That is the whole premise of the Digital Empire book The $100 Network: own the pieces that are yours, rent the pieces that are commodities, and never let the plumbing trap you. When I start a new Cloudflare build I do not hand-write the config; the Single Site Gen tool generates a production-ready setup with the security headers and bot posture already correct. If you want the deeper architecture discussion, including when to go all-in on one platform versus splitting across several, that is the frameworkless-versus-lock-in post.
The honest summary
Static brochure site: Cloudflare or Netlify, near free. Real app on Next.js with a developer: Vercel is worth it. Many small client sites: Netlify plus Cloudflare. Whatever you pick, keep your content, your data, and your domain portable, and you can always change your mind for the cost of an afternoon instead of a rebuild. Next in this series, the same thinking applied to taking money.
The series
- Previous: Part 1, local AI vs cloud APIs
- Part 2 (this post): Picking a web platform without lock-in
- Next: Part 3, Square and a portable storefront for taking money
Related reading
- What hosting a new business site actually costs in 2026, the current numbers for all three plus what each platform meters
- Observe, evaluate, launch, the whole launch sequence this hosting decision sits inside
- The Cloudflare developer platform, the full tour of what you can build
- Going frameworkless on Cloudflare vs vendor lock-in, the deeper architecture decision
- Netlify WAF vs Cloudflare double-CDN, what happens when platforms overlap
- Single Site Gen, generate a production-ready build prompt
Fact-check notes and sources
Hosting prices and limits change; treat these as approximate mid-2026 figures and confirm on each vendor's page before relying on them.
Prices rechecked against each vendor's live pricing page on 6 August 2026.
- Cloudflare Workers (Free 100,000 requests/day at 10 ms CPU per invocation; Paid from $5/month including 10 million requests and 30 million CPU milliseconds, then $0.30 per additional million requests; "Requests to static assets are free and unlimited"): Workers pricing. R2 zero egress: R2 pricing. "Start with Workers, not Pages": best practices.
- Vercel (Hobby free with 1M edge requests, 100 GB transfer, 1M function invocations; Pro $20 per user per month with a $20 included usage credit, 10M edge requests, 1 TB transfer): Vercel pricing.
- Netlify (credit model: Free $0 / 300 credits, Personal $9 / 1,000 credits, Pro $20 / 3,000 credits with unlimited team members and 3+ concurrent builds, Enterprise custom; deploy 15 credits, bandwidth 20 credits per GB, compute 10 credits per GB-hour, 2 credits per 10,000 web requests; Forms free): Netlify pricing and the April 2026 changelog for the move to credits.
- Next.js cross-platform adapter (eased Vercel lock-in): Next.js across platforms.
This post is informational, not a paid endorsement, and no affiliation with Cloudflare, Vercel, or Netlify is implied. Prices and limits were verified on 6 August 2026 and change often; verify before building on them.