Cheapest way to host a SaaS in 2026

Hosting advice usually jumps straight to provider names, but the cheapest question is really a stage question: what does your product cost to run this month, before anyone pays for it? The short answer is that you can start at $0, and you should stay near $0 until paying users drag the bill upward.

What is the cheapest way to host a SaaS MVP?

The cheapest way to host a SaaS MVP is a free-tier application platform paired with a free-tier database — for a JavaScript stack that usually means Cloudflare Workers or Pages plus Turso, which puts a small first release at roughly $0 per month.

A typical MVP does only three things: serve a static front end, run a modest number of server functions, and store a small amount of data in Postgres or SQLite-compatible storage. All three have genuine free tiers in 2026, and the free allowances are far larger than an unlaunched product will touch. The unavoidable paid line item is your domain at roughly $10–15 per year, which is about a dollar a month. Transactional email providers, uptime checkers, and error trackers also have free starting plans that cover early traction; their limits move, so re-check the current numbers before you commit a stack to one.

The cheapest MVP hosting is boring hosting. Do not spend launch money on reserved Kubernetes capacity, an always-on GPU, a staging cluster, or a $40/month managed Postgres instance before your first user exists. None of those make an MVP more likely to succeed; they only make the runway shorter.

That free-first shape is exactly what the Startplate cost section presents: a $0 free-tier starting point running on Cloudflare with Turso as the data layer, captioned with the idea that your product decides the next number. The foundation does not arrive with a recurring infrastructure bill, so the first months of hosting cost whatever your domain costs and nothing else.

The practical rule: pick the free path while you are still learning what the product should be, and let real usage — not anxiety — trigger the first paid upgrade. The rest of this post is about where that upgrade should happen.

Can I host a SaaS for free or under $5/month?

Yes — a SaaS can run entirely on free tiers until real traffic arrives, and when free tiers pinch, the sensible next step is almost always a plan in the $5/month range rather than a $50 one.

Break the bill into buckets and each one has a free starting point. Application runtime: free allowances measured in requests, bandwidth, or build minutes. Database: managed SQLite and Postgres services with free rows, storage, or connection tiers. Email: providers whose free plans cover the first few thousand transactional messages a month. Monitoring, analytics, and error tracking: free plans that are genuinely useful at small scale. Add a domain at roughly a dollar a month and your total is $0–2.

If free limits start to bite, the under-$5 tier buys one of three things: a small VPS at roughly $4–7/month, an entry-level paid database plan around $5, or extra serverless capacity on the platform you already use. Each is a targeted fix for whichever bucket actually overflowed, not an upgrade of your entire stack.

Where “free” stops being free is overage pricing and attention. Serverless platforms can bill for egress or invocations past the allowance, some paid features only appear on the paid tier, and free-tier limits change often enough that you should set usage alerts rather than memorize the numbers. The safest free architecture is one where exceeding the allowance degrades gracefully instead of silently generating an invoice.

The FAQ answers this directly: the starter is designed to begin on free infrastructure paths, and you choose paid services only when the product needs them. The cost logic is also worth comparing against the real numbers in the SaaS boilerplate cost breakdown, because hosting is only one line in that budget.

Hetzner VPS vs Vercel vs Railway — which is cheapest at scale?

Hetzner is cheapest for steady, predictable load; Vercel (and similar serverless platforms) is cheapest at low and spiky traffic; Railway sits in between — cheap while idle, but it can pass a VPS in cost once usage becomes sustained. The winner depends on your traffic shape more than on the brand.

The three platforms price completely differently, which is why founder debates about them usually talk past each other. Serverless platforms like Vercel charge for usage: requests, bandwidth, function duration. You pay nothing small, then pay more as usage grows, and the pricing is tied to activity rather than capacity. A Hetzner VPS is a flat monthly fee for an entire machine — generous CPU, RAM, disk, and egress for a fixed price — so your cost is the same on a quiet Tuesday as on launch day. Railway bills for resources actually consumed by running services plus a platform fee, so an idle app is cheap and a memory-hungry app gets expensive fast.

As rough estimates: a small app inside free allowances costs $0 on serverless and still costs about $4–6/month on a VPS, so VPS loses at the very bottom. Once you have steady traffic that blows past free allowances, a single $6–12/month VPS can carry a workload that costs $30–80/month on usage-based platforms, and a busy app with a real database commonly runs $40–150/month managed versus $10–30 self-hosted. Those are estimates, not quotes — but the direction is consistent: flat machines win under sustained load, usage pricing wins under spiky or uncertain load.

There is a fourth cost: your time. If operating a server costs you hours you could spend on the product, the cheaper invoice may be the worse deal. If you are comfortable with SSH, the flat VPS is usually the cheapest real production option available. Deployment details for either path live in the deployment guide, and the broader stack trade-offs are covered in the indie hacker tech stack breakdown.

Is self-hosting on a $5 VPS safe for production?

Yes — a $5 VPS can safely run a real production SaaS, because plenty of small products run perfectly well on hardware that cheap, but only if you treat patching, TLS, backups, and monitoring as part of the hosting cost instead of optional extras.

Safety on a single cheap machine has four parts. Data safety: automated nightly database dumps to off-machine storage, plus at least one restore test, because an untested backup is a rumor. Availability: a single box is a single point of failure, so accept that an outage might last minutes to hours, or add snapshots and a recovery runbook so restarts are fast. Security: SSH keys only, a firewall, unattended security updates, and rate limiting on anything public. Correctness: automatic TLS (Caddy makes this nearly zero-effort), log rotation, and enough swap or memory headroom to survive traffic spikes — running out of memory and getting killed by the OOM reaper is the most common failure mode on tiny boxes.

All-in, that setup lands around $5–10/month for the machine plus roughly $0–3 for backup storage, with the domain on top. The real risk is never the hardware; it’s the operator. If you will not patch monthly or verify restores, a managed platform is cheaper than a breach or a lost database. If you genuinely will do those things — or if you enjoy them — a cheap VPS is legitimate production infrastructure.

The alternative is not paying more forever; it’s avoiding sysadmin work entirely. The Startplate cost model is built around that: the runtime sits on an edge platform with a free path, so there is no server babysitting, and you spend money only on the parts that earn their keep (see the environment variable reference for what has to be configured either way).

How much does hosting cost for a SaaS with 10,000 users?

As a rough estimate, a SaaS with 10,000 registered users costs $0–30/month if it is a light web app on free or starter tiers, and $30–150/month once it needs a managed database, meaningful email volume, and steady serverless traffic. Registered users are a poor proxy for load; activity, requests, and data size are what you actually pay for.

Do the arithmetic in your favor: 10,000 registered users with, say, 8–15% daily activity is 800–1,500 people a day (an illustrative range, not a benchmark). If each performs a few dozen requests, that’s on the order of tens of thousands to low hundreds of thousands of requests a month — often inside cheap or free serverless allowances, but close enough that you’d be watching the dashboard. Bandwidth might be tens of gigabytes a month, your database might sit under a gigabyte for a long time, and transactional email — welcome, verification, receipts, resets, digests — can quietly become the line item that crosses a free limit first.

A plausible monthly spread, all estimates: compute $0–20, database $0–15, email $0–20, storage/CDN $0–10, monitoring and errors $0–10, backups $0–5, domain ~$1. A free-tier-optimized stack can still land near $0–5 if you stay inside allowances. A single modest VPS with self-hosted Postgres usually runs $5–20. A managed platform path for the same product commonly lands at $40–150.

The trap is paying for peak-shaped infrastructure you don’t have: an always-on cluster sized for a launch spike, or a managed database tier bought for a feature that doesn’t exist yet. The boilerplate side of the budget — a one-time license instead of a monthly platform fee — is analyzed in is a SaaS boilerplate worth it. When the bill does need to grow, Startplate’s pricing is a one-time purchase that does not add a monthly line to hosting.

The short version

Start at $0: free-tier runtime plus a free-tier database covers a real MVP, and only a domain is truly unavoidable. Under $5 a month is a normal steady state for a small product, not a compromise. At scale, flat VPS pricing wins for steady load while usage-based platforms win for spiky load — and self-hosting is safe only if you budget for backups, patching, and monitoring. Ten thousand registered users usually means single-digit to low-double-digit dollars per month for a light app, because traffic shape, not signup count, sets the invoice.