Dora NodaYour side project costs $0 a month to host right up until the moment it works. Not until it scales, not until it goes viral — until it earns its first dollar.
That is the deal inside Vercel's Hobby tier: free forever, but licensed for personal, non-commercial use only. The day you add a price tag, a paywall, or a sponsor button, the terms say you belong on Pro at $20 per seat per month. The upgrade bill arrives with the first dollar of revenue, not the first traffic spike.
This is worth spelling out because almost every hosting comparison gets the trigger wrong. People model the Hobby-to-Pro jump as a usage event — you outgrow 100GB of bandwidth, so you upgrade. But for most commercial side projects, the trigger is a licensing event. Your traffic can be flat at a few thousand visits a month and you still owe the seat fee, because the thing that changed is what the app earns, not what it consumes.
So let's do the math the trigger actually implies: a worked first-revenue-year comparison of the Hobby-to-Pro seat jump versus a flat self-hosted box that never asks what your app earns.
The first-revenue-year math
Assume a typical just-monetized side project: modest traffic, comfortably inside every included-usage meter on either side. No overages anywhere — this is purely the cost of the license trigger firing.
Vercel Pro is $20 per seat per month with no annual discount, and each seat carries a $20 usage credit that covers small-project overages but doesn't reduce the seat fee itself. The self-hosted side is a Hetzner CX22 — 2 vCPUs, 4GB of RAM, 40GB of disk, 20TB of included traffic — at roughly €4.50 a month, or about $60 for the year.
| Team | Vercel Pro, year 1 | Hetzner CX22, year 1 | Delta |
|---|---|---|---|
| Solo founder (1 seat) | $240 | ~$60 | ~$180 |
| Two co-founders (2 seats) | $480 | ~$60 | ~$420 |
| Small team (5 seats) | $1,200 | ~$60 | ~$1,140 |
Three things stand out. First, the solo-founder row is the entire story for most readers: the moment of first revenue converts a $0 hosting line into $240 a year, a 4x multiple over the flat box, with zero change in traffic.
Second, the Hetzner column doesn't move. A flat box prices machines, not headcount — the fifth teammate costs the same as the first, which is nothing extra. Third, the delta scales with the team, not the app. A five-person team pays twenty times the flat box before a single overage meter fires, purely because five humans can deploy.
To be fair to Vercel, that $20-a-seat fee is renting something real — zero-ops deploys, preview environments, and a global edge network — and we'll come back to when it's worth it. But notice what it is not: it is not capacity. You are not buying headroom for growth. You are buying permission to earn.
Seats, not servers, drive the bill
The seat ladder deserves a closer look because it behaves so differently from infrastructure pricing. On Pro, every deploying teammate is another $20 a month. Viewer seats are free, which softens the blow for stakeholders who only look at dashboards — but anyone who pushes code, including the designer fixing copy and the contractor shipping one feature, is a paid seat. And there is no annual prepay discount on Pro, so the year-one numbers above are the floor, not a starting offer you can negotiate down with commitment.
Compare that to how a flat box scales with people: it doesn't. Ten contributors deploying through a git-push pipeline on a $5 VPS cost exactly what one contributor costs. The box meters CPU, RAM, disk, and traffic — inputs to running software. It has no concept of who typed git push.
This produces a simple rule of thumb: Vercel's entry fee scales with your team's size; a VPS's entry fee scales with your app's appetite. For a solo founder with a hungry app, Vercel can be the cheaper side. For a five-person team with a light app — the modal just-monetized side project, a Next.js site with a Postgres database doing thousands of visits a month — the seat ladder is pure overhead over capacity you could rent for a twentieth of the price.
One more wrinkle: enforcement of the Hobby non-commercial clause is famously soft. Vercel emails and asks you to upgrade rather than pulling the site, and the trigger is usually commercial-shaped usage patterns rather than an audit of your Stripe account.
That softness tempts founders to stay on Hobby "just until revenue is real." Treat that as what it is — borrowed time on a tier whose terms you've outgrown, not a pricing strategy. The honest comparison prices the compliant path.
The second cliff: metered usage after the seat jump
The seat fee is only the first cliff. Once you're on Pro, the second one is metered usage — and its shape is what turns a $20 subscription into a four-figure surprise.
Start with bandwidth, the most legible meter. Pro includes 1TB of fast data transfer a month; the Hetzner CX22 includes 20TB. That is a 20-to-1 ratio on the same line item, before overage rates even enter the picture.
Past the included terabyte, Vercel bills bandwidth by the gigabyte — and bandwidth is the meter most coupled to success. A launch-day spike, a front-page post, an image-heavy page that gets shared: all of these are good news for the business and bad news for the invoice, on exactly the month you least want a surprise.
Compute is the sneakier meter. Every API route, every server-rendered page, every middleware execution counts, and the billable unit is duration — how long your functions run, not just how often they're called.
The canonical cautionary tale came in July 2025, when a Vercel customer reported on X that moving roughly 20 client-side requests to the server side on a single page took their bill from $300 a month to $3,550, with 99% of it from serverless function duration. Nothing about their traffic changed; the architecture did. Slow-returning calls — AI API round-trips are the classic case — turn each invocation into a long, metered burn.
None of this is hidden. Vercel publishes the meters and the $20-per-seat usage credit absorbs small excursions. The point is structural: on metered serverless, cost is a function of architecture and attention, and both degrade exactly when you're busiest.
A flat box inverts that relationship. Its cost is fixed whether your page makes 2 requests or 20, whether functions take 50 milliseconds or 50 seconds. The failure mode of success on a VPS is a slow page you can see coming in your own graphs — not an invoice you discover after the fact.
The counter-example: free tiers that allow commercial use
Here is the part that reframes the whole discussion: the revenue trigger is a business choice, not an industry law. Cloudflare Pages' free tier — unlimited bandwidth, 500 builds a month — permits commercial use. No clause converts your first dollar into a $20 seat.
Netlify's free tier likewise allows commercial projects. The "free until incorporation, not until scale" trigger is Vercel-specific packaging, and knowing that changes the migration math.
It means a just-monetized static or lightly-dynamic site has three doors, not two. Door one: pay the Vercel seat ladder for the platform you're already on. Door two: move the frontend to a free tier without the commercial tripwire and keep everything else. Door three: take the whole workload to a flat box and fix the cost line permanently.
Doors two and three both exist because competitors decided the first dollar is not the moment to start charging — which tells you the Hobby clause is doing pricing work, not cost-recovery work. Vercel's costs don't change when you charge money; only your willingness to pay does.
That is also why the honest self-host comparison is against Pro, not Hobby. Hobby is a development tier wearing a production tier's clothes. The moment your project is a business, your Vercel price is the seat ladder plus the meters — and that is the number the flat box beats by 4x to 20x in year one.
When staying on Vercel still wins
None of the above means leaving is always right. The seat fee rents genuine value, and for some teams it's cheap at the price. Stay on Vercel when:
- Deploys are your scarcest resource. If nobody on the team wants to own a VPS — patching, TLS renewal, deploy pipelines, 3 a.m. restarts — $20 a seat is less than one hour of anyone's time. The ops you don't do is the feature you're buying.
- Preview environments are load-bearing. Per-branch preview URLs with zero setup are Vercel's most underrated feature. Reproducing them self-hosted is real engineering, not a weekend script.
- Your traffic is spiky and global. The edge network and instant autoscaling earn their keep when traffic arrives in bursts from everywhere. A single-region VPS can't match that shape without overprovisioning.
Leave — or never arrive — when the reverse holds: someone on the team is comfortable owning a box, traffic is predictable, and the app's shape is a boring web service plus a database. That describes an enormous fraction of just-monetized side projects. A single CX22-class machine runs the app, Postgres, and Redis in Docker Compose with room to spare, behind Caddy for automatic TLS, for less than one Vercel seat costs in three months.
The cleanest mental model: Vercel charges for the team and the tail; the box charges for the machine. If your costs live in headcount and traffic spikes, rent the platform. If your costs live in steady-state compute, own the machine.
The bottom line
The Hobby tier's non-commercial clause makes Vercel's pricing page read backwards for founders. The question is never "when will I outgrow free" — it's "when will I charge money," and the answer is usually an afternoon's work with Stripe, long before any usage meter matters. Price the compliant path from day one: $240 a seat-year against a ~$60 flat box, with bandwidth at a 20-to-1 included-traffic disadvantage and compute meters that punish exactly the slow, AI-flavored requests modern apps are built from.
That doesn't make Vercel a bad deal. It makes it a deal with a specific shape — metered, seat-scaled, ops-free — that fits spiky teams better than steady ones. Know which team you are before the first dollar lands, and the invoice will never surprise you.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Star the repo on GitHub or deploy your first app today.
