Vercel, explained: the platform where websites ship from Git

The short answer

Vercel is a cloud platform for building, deploying, and serving websites and web applications - with a specific superpower: it is the company that created and maintains Next.js, the framework most of its customers use. Push code to Git, and Vercel builds it, deploys it to a global CDN, and gives every pull request its own preview URL.

It is also the company behind v0, the AI app generator, and it has repositioned itself as an "AI Cloud" - which is what a September 2025 funding round of $300 million at a $9.3 billion valuation was buying. This site, for full disclosure, deploys on it.

From ZEIT to Vercel

Guillermo Rauch founded the company in 2015 as ZEIT, with a single tool called `now` that deployed a folder to a URL in one command. The next year he released Next.js, and the two became a matched pair - framework and host designed together. The company renamed itself Vercel in April 2020.

The funding arc mirrors the product's reach: seed through a Series C at a $1.1 billion valuation in mid-2021, Series D at $2.5 billion later that year, then the big step in September 2025 - a Series F of $300 million at a $9.3 billion post-money valuation led by Accel and Singapore's GIC, explicitly to scale the AI Cloud and v0. What was once a hosting startup became one of the most valuable developer-tool companies in the world.

Along the way it assembled the surrounding ecosystem deliberately: it acquired Turborepo - the monorepo build system - along with its creator Jared Palmer in 2021, built the Rust-based Turbopack bundler to replace webpack inside Next.js, and over the years hired or absorbed the teams behind much of the open-source stack it runs on, from the SWR data-fetching library to the maintainers of shadcn/ui and the Nuxt framework team. The pattern is consistent: where the frontend toolchain had a critical piece, Vercel tended to end up stewarding it.

The core insight: deployment as a Git workflow

Vercel's founding insight was not a technology but a workflow. Before it, deploying a site meant servers, build scripts, FTP, and prayer. Vercel's bet was that deployment should be a consequence of version control: every branch and pull request gets a built, live URL automatically. Code review becomes site review.

That loop - push, build, preview, merge, promote - is now so standard it is hard to remember it was novel. It is the reason frontend teams describe Vercel as an accelerant rather than a host: the platform removed an entire category of friction between writing code and showing it.

What the platform actually does

Under the workflow is a real infrastructure stack. The build system compiles your app - Next.js natively, plus frameworks from Astro to Nuxt - and the output is split across a global CDN for static assets, serverless functions for on-demand compute, and an edge network that runs code geographically close to the visitor.

Newer pieces extend that base: Fluid Compute pools serverless invocations more efficiently (reusing warm instances across concurrent requests instead of the one-function-per-request model), Cron jobs handle scheduled work, skew protection keeps a user's session pinned to one deploy, and instant rollback lets any previous deployment go live again. A Marketplace of integrations wires in databases, storage, and services from partners including Supabase and Stripe - so the platform is the deployment surface on top of a stack of specialized providers rather than a do-everything cloud.

v0: the AI builder that crossed over

v0 is Vercel's answer to the question everyone in the vibe coding world was circling: what if describing a UI just produced it? Launched in September 2023 as a private alpha - with a hundred-thousand-person waitlist in three weeks - it started as "generative UI": describe an interface, get three options of real React + Tailwind + shadcn/ui code, copy it into your project.

The arc since is the interesting part. From code generator it became an app builder that plans, connects databases, and deploys, and in August 2025 Vercel relaunched it at v0.app explicitly for non-developers - the pitch widened from "a tool for engineers" to "a builder for anyone with an idea," complete with an iOS app and one-click production deploys to the platform. Reported usage is in the millions. For Vercel it is the top of the funnel: a prompt becomes an app, the app deploys to Vercel, and the deploy lands on Next.js - each layer of the stack feeding the next.

The AI stack underneath: SDK, Gateway, agents

Below v0 is the part of the company most non-customers never see: a deliberately complete stack for building AI-powered software. The AI SDK is the open-source TypeScript toolkit that became the standard way to call models - unified provider API, streaming-first, structured output, tool calling, framework support across React/Next/Vue/Svelte/Node. It is the reason "build an AI feature" in the JS ecosystem usually means Vercel's library.

AI Gateway sits at the network layer: one managed gateway to 200+ models across providers, with centralized credentials, failover between providers, spend tracking, zero data retention, and - the notable policy - zero markup on token prices, usable from any infrastructure, not just Vercel. Around it: a Workflow SDK for long-running durable agents, Sandbox for running untrusted or agent-generated code in isolated microVMs, Eve (an open-source agent framework), a Vercel Agent for code review, a Chat SDK for building conversational agents, Connect for handing agents short-lived OAuth tokens to external systems - the MCP wiring, essentially - and Passport for identity. Bring Your Own Cloud even lets enterprise customers run the whole thing inside their own AWS tenant.

Read together it is not a feature list but a thesis: if software is going to be built and run by agents, Vercel wants to own every layer that proposition needs.

Storage and the data layer

The data story has quietly changed. Early on Vercel offered its own Postgres and KV products - both now deprecated, replaced by the Marketplace, which provisions partner databases (Neon, Upstash, Supabase, and others) and injects the credentials straight into your project as environment variables. What remains first-party is the edge-native layer: Blob for storing and serving files, and Global Config (formerly Edge Config) for sub-millisecond reads of things like feature flags and redirects replicated across the CDN.

The honest read is that Vercel tried being a database company and pivoted to being the deployment surface that brokers best-in-class ones - which is why its stack composes so naturally with dedicated backends like Supabase.

Security and observability

The platform bundles a real security surface: a web application firewall and DDoS protection, bot detection (BotID), deployment protection and password-protected previews, and a Security dashboard surfacing threats and events across a team. On the other side of running software is visibility - the Observability suite pulls in function logs, web vitals, and analytics, so the metrics that matter (including the Core Web Vitals our PageSpeed guide covers) are measured where the traffic actually is.

The feature that sells it: preview deployments

Ask a team why they are on Vercel and the first answer is usually previews. Every pull request builds and deploys its own URL - the designer, the PM, and the client all click the same link and look at the real site, not screenshots. For an agency like ours it changed how work gets reviewed: approvals happen on the actual page, not a description of it.

The preview itself has grown into a collaboration surface too - a toolbar lets reviewers drop comments directly on the live preview, flag issues, and even toggle feature flags, so feedback lands on the thing itself rather than in a screenshot thread. Around that sit the small compounding features: automatic HTTPS, instant rollbacks to any previous deploy, analytics, and the image and font optimizations the framework assumes. None is individually revolutionary; together they are why the platform's churn is low.

The strategy: own every layer

Read together, Vercel's products are less a catalog than a stack with intent: the framework (Next.js, Turbopack, Turborepo), the deployment platform, the AI generation surface (v0), the AI infrastructure layer (SDK, Gateway, agents), and the data and security surfaces around the edges. Each layer feeds the next - a prompt becomes a v0 app, the app deploys on Next.js to the platform, and the platform sells the compute and the observability.

That is what the $9.3 billion Series F priced: not hosting margins, but a vertically integrated bet on how software gets built and run in the agent era. It is the cleanest corporate expression of the same shift that produced the vibe coding wave.

Pricing and the honest tradeoff

Vercel has a genuine free tier - Hobby covers personal projects - then Pro at around $20 per member per month, and Enterprise above that. The catch is the second axis: beyond included limits, usage is metered. Bandwidth, function invocations, image optimizations, and edge requests all have per-unit pricing, and a traffic spike or a traffic-heavy app can produce a bill that scales faster than the revenue it serves.

This is the legitimate criticism, not a hidden one: Vercel is priced like a premium managed service, and at high scale a raw cloud or a CDN-plus-origin setup is cheaper. The trade is the same one managed platforms always offer - you pay for not operating it yourself - but it is worth going in with eyes open, and heavy-traffic apps should model it rather than assume it.

The lock-in question, honestly

Because Next.js is open source, the lock-in is softer than it looks: the same codebase builds to a standalone output that runs on any Node host, a container, or via OpenNext on Cloudflare or AWS. What you would lose leaving Vercel is not the framework but the platform conveniences - previews, edge network, one-click rollbacks, the dashboards.

The real dependency is subtler: Vercel steers the framework's roadmap, and features land on the platform first and most smoothly. For most teams that alignment is a feature; for teams that want a host-neutral framework, it is a fair thing to weigh.

The alternatives

The closest rivals cover different points on the same map. Netlify pioneered the same Git-based deployment model and remains the strongest general alternative. Cloudflare Pages plus Workers offers similar reach at a much cheaper usage price for teams who can do more of the plumbing. Railway and Fly.io serve teams that want closer-to-the-metal containers. AWS Amplify and the cloud giants offer hosting as one line item in a larger cloud bill. And plain self-hosting - a VPS with a container - is always the sovereign option.

Where Vercel is genuinely strong

The platform's sweet spot is well-defined: marketing sites and web apps where speed of iteration and performance matter more than marginal infrastructure cost. Preview-driven teams, agencies shipping many client sites, startups that want to deploy daily without a platform team - these get real leverage. The edge network is genuinely fast, the Next.js integration is native, and the operational surface is close to zero.

Enterprise buyers get SOC 2, SSO, audit logs, and support tiers - the things that turn a developer tool into a vendor a company can sign.

Where it is weaker

The weaknesses follow from the model. Usage pricing penalizes success at the high-traffic end; a media-heavy site or a viral moment can outgrow the value of the convenience. The platform is for the frontend and the request layer - anything long-running, stateful, or compute-heavy still belongs elsewhere. And because the stack is managed, deep infrastructure customization hits a ceiling that a self-hosted stack does not have.

Why we deploy on it

This site runs on Vercel for the unglamorous reasons: every branch gets a preview URL that clients can open, the global edge network serves the statically generated pages fast enough that our Lighthouse scores pass without tuning, and deploying is a `git push`. The Next.js integration is native because the same company maintains both.

For a services business shipping client sites, the operational savings are the feature - no platform to run means more of the budget goes to the site itself. And because the deliverable is an open-source Next.js codebase, the client is never trapped by our choice. If you want the details on how a project actually moves through this pipeline, book a call or see the website development page.