Jason LeeVercel

Item 58 of 59

The free tier is a loan against the day the project works

5 min 1,171 words

The Vercel logo, a black triangle beside the Vercel wordmark, centered on a pale gray background

At 9:41 on a Thursday morning in March my analytics graph went vertical, and I got to watch the interesting part happen in real time. A two-year-old essay about status meetings had been linked by a newsletter with more readers than my entire site, possibly more than my entire life. Traffic multiplied something like forty times within the hour. The hosting did exactly what the hosting was designed to do. Around eleven, the free tier's cap arrived, and the site stopped serving. Not an invoice. A pause. My best day as a writer, and the domain returned a polite screen explaining that the project had reached its limit. I make this sound more dramatic than it felt; I was eating lunch. But I remember thinking that a machine had made a decision about my work without consulting me, and the decision was technically correct.

Because that pause is the argument. The free tier is a loan against the day the project becomes load-bearing, and the day the project becomes load-bearing is the day you can least afford the renegotiation. The cap is not a flaw in the offer. The cap is the offer.

The caps are honest, which is what makes them persuasive

Start with what Vercel gets right, because the critique doesn't land otherwise. The Hobby plan is free for personal, non-commercial projects and ships hard caps: 100 gigabytes of data transfer a month, a million function invocations, a million edge requests. Hit one and the project pauses until the month resets. Nobody gets billed by surprise. Anyone who has run infrastructure knows how unusual that is. Most clouds would have run the meter in silence and mailed the invoice as a discovery. The pause is a kind of honesty, and the honesty is what makes the upgrade conversation feel reasonable rather than extortionate.

Then you look at the shape of the paid tier. Pro costs $20 per developer seat monthly and includes a $20 usage credit, per Vercel's pricing page. Beyond a terabyte of bandwidth, data transfer bills at $0.15 per gigabyte. Edge requests beyond ten million bill at $2 per million. Functions bill by the CPU-hour. The company runs at least seven separate meters, and each one is a sentence in a FAQ you will read only after the sentence matters.

The genre this produces has a name by now: the bill-shock post. The documented cases in the wild, per a July 2026 roundup of community incidents, include a $1,477 invoice, about $1,267 of it bandwidth, after a crawler pulled 8.4 terabytes of media from one site, a bill near $23,000 rung up by a distributed denial-of-service attack, and about $3,200 of pure attack traffic on another. The stories function like scar exhibits. They circulate among people deciding where to host, and they perform a service for the vendor, whether or not anyone intends it: terror is excellent marketing for spend management, and spend management is a paid feature. Consider the incentive. The company profits twice from the same meter, once when you exceed it and once when you buy the thing that watches it.

What the same traffic costs three ways, checked against posted pricing in October 2026:

Path Monthly cost Your best day
Vercel Hobby $0 project pauses at the cap
Vercel Pro $20 a seat, then meters the meters run
Six-dollar server, static site $6 nothing happens

The framework is the funnel

Now the part that makes Vercel structurally interesting rather than merely expensive. Next.js, the most popular React framework, was created by Vercel's team in 2016, and its development is led by a core team working full-time at Vercel, per the project's own governance page. The license is MIT and the copyright sits with the company. Guillermo Rauch, the CEO, described the arrangement to The Register plainly enough in 2021. React core team members have crossed over to Vercel payroll over the years, and the company announced at least one such hire in a blog post of its own. None of this is hidden. All of it is legal. None of it is neutral.

The complaint consensus among developers tracks the structure. Features that are trivial on Vercel are awkward to self-host. The documentation increasingly assumes the framework's server-centric path. The framework's most natural deployment target is the company that pays its core team. There is a real counterargument here, and it deserves its paragraph: the React team at Meta originated server components and went looking for an external sponsor to prove them out, so part of the causality runs the other way. Both things are true. The React team had the vision, and the company that funded the vision owns the funnel it flows through. I spent two years in SaaS product management, and I sat in the room where a usage meter was designed. The question on the whiteboard was never how to be fair. It was where, exactly, generosity should stop.

The outrage has survivorship bias, and the caps are kinder than the alternatives

The strongest version of the defense, argued properly this time. For every horror invoice there are thousands of hobby sites that never approach a cap and pay $0 forever, and the horror stories are a selected sample. Nobody writes the post about the year nothing went wrong. Egress genuinely costs money, and the competitor advertising zero-egress pricing is subsidized by its other businesses, not more honest about physics. Twenty dollars a seat is small against the engineering hours a zero-config deploy with per-request preview environments saves a small team. Spend management lets you set your own ceiling, which means the crawler story ends at your limit instead of your card's. And the pause, I'll say it again, beats the surprise invoice that most of the industry's actual infrastructure would have sent. The framework critique at its strongest stops short of accusing anyone of malice and lands on something more ordinary: roadmaps follow payrolls, which is worth watching rather than condemning.

I should state my own position in the ledger, since it's not clean either. I took two years of value from that free tier and never once asked what subsidized it until the graph went vertical. My preference for a boring server partly serves my self-image as a person who owns the boring parts, and I'd sound less principled if the migration had taken a weekend instead of three evenings.

Owning the boring parts is cheaper than renting the exciting ones

It depends on whether the project stays yours. If it stays small and personal, Hobby is free, honest, and better than most of the industry at admitting its limits. The variable worth pricing is the day the project stops being personal; that day, the seat fee is the cheap part and the meters are the bill. This site runs on a six-dollar server and a twelve-dollar domain now. The graph has gone vertical since, nothing happened, and the essays stayed up while it happened. A free tier is a loan. The boring server is a purchase.