Infrastructure

Lightweight Infrastructure for Lean Online Businesses

A reference stack for a small online business running lean, with concrete choices and what they cost.

Published 2026-05-13 · By Claire Miller

A small online business running lean in 2026 needs infrastructure that is cheap, fast, observable, and replaceable. The four properties are not all forced by one stack; choosing them together is the craft. The reference stack below has served a dozen small businesses well in 2025-2026 and represents the lower bound on what the platform layer should cost when revenue is small.

The reference stack

For a small online business operating a services or agency model with up to five people, the reference stack is roughly:

Hosting. One static-site CDN: Cloudflare Pages, Netlify, or Vercel. Vendor cost: zero to low tens of dollars a month. Compute: none. Database: none. The website is a directory of files.

Domain. A primary domain, possibly a few secondary ones for marketing campaigns. Cost: $10-20 per domain per year.

Email. Google Workspace or a competitor. Cost: roughly $7 per user per month. The email system is the operational system; it stays on a managed provider because self-hosting email is a tax.

CRM. Attio, HubSpot, Pipedrive, or Folk. Cost: free to roughly $20 per user per month. The CRM is the human-side memory; managed makes sense here.

Auth. For services that need logged-in users, Clerk or WorkOS or the auth that comes with the framework being used (NextAuth, Better Auth, Lucia, custom). Cost: zero to low hundreds per month. Vendor auth avoids a category of bugs.

Storage. For files, attachments, and object storage: Cloudflare R2 or Backblaze B2 or the storage tied to the host (Vercel Blob, Netlify Large Media). Cost: pennies per gigabyte per month.

Observability. For startup logs, cron logs, and worker logs: a logging destination. Most small businesses use Logflare, Highlight, Axiom, or a simple Tailwind-Tailwind-Axiom combo. Cost: free to low tens of dollars per month.

Secrets. A managed secrets manager: Doppler, Bitwarden, or 1Password for Teams. Cost: small monthly fee. Managing secrets in env files is the first thing small businesses get wrong.

Email transactional. Postmark, Resend, or SendGrid for transactional email (receipts, notifications). Cost: pennies per email.

Backups. A scheduled snapshot of any stateful system: the database, the storage bucket, the email logs. Most small businesses use a script that runs weekly and stores snapshots in R2 or B2. Cost: nearly zero.

That is the stack. Total vendor cost: roughly $50 to $200 per month for a small business.

What does not need a vendor

Several pieces of operational infrastructure can be run locally without a vendor:

• Cron jobs. The host's scheduler or a small VM running cron.
• Status pages. A static status page generated from a JSON file.
• Internal tools. A small Next.js or Astro admin app, deployed to the same host.
• Scripts and CI. The host's build pipeline plus a couple of cron jobs.

The discipline is to defer the vendor until the local version's maintenance becomes meaningful. Most local automations remain cheaper than the vendor alternative for years.

What always needs a vendor

Four categories of infrastructure should always be on a managed vendor, regardless of the business's size:

Email. Self-hosting email is its own full-time job. Use a vendor.

Backups of state. Use a managed object store. Self-hosted backups often stop running.

Domain and DNS. Use a managed DNS provider (Cloudflare, DNSimple, Google Domains successors). Self-hosting DNS is rarely worth the security implications.

Auth. For any service with logged-in users, use a vendor. Self-rolled auth is a security liability.

The non-functional properties

The reference stack earns four non-functional properties:

Speed. Static sites on a CDN load in under a second globally. Email vendors deliver in seconds. APIs respond in tens of milliseconds. The stack is fast by construction.

Cost. The stack's marginal cost is roughly a few cents per customer per month. As the business grows, the cost grows slowly until the business hits a transition point (significant database, significant media needs) that justifies a more capable stack.

Observability. The stack's logs are piped to one place and one search. When something breaks, the operator finds the broken thing in minutes.

Replaceability. Every piece of the stack is replaceable on its own. The CRM can be swapped; the host can be swapped; the email vendor can be swapped. The reference stack's design point is that no vendor becomes so load-bearing that switching becomes a major project.

What the stack cannot do

The reference stack cannot run a marketplace. It cannot run a high-throughput data pipeline. It cannot run a real-time collaboration tool. It cannot run a SaaS application with hundreds of thousands of users.

For the businesses the stack serves, none of those are requirements. For the businesses that have one of those requirements, the stack is not the right starting point.

What to do this quarter

For a small online business starting from zero in 2026, the practical project is to build the stack above with the providers that fit the team's existing comfort. Most decisions are reversible. The discipline is to keep the stack replaceable so that when a vendor changes pricing or shuts down a feature, the business can move.

The lightweight-infrastructure discipline is: cheap, fast, observable, replaceable. The reference stack delivers all four for under $200/month, and that is the right starting point for a lean online business in 2026.