Dora NodaThe vibe-coding tools market hit an estimated $4.7 billion in annualized spend in 2026, and 92% of US developers now use AI coding tools every single day. Those two numbers, on their own, would be the story in any other year. This year they're the setup: only 29% of those same developers say they trust the code the tools produce, and every dollar of that $4.7 billion is spent on the tool that writes the app — not the infrastructure that runs it.
That's the split nobody's pricing in yet. Code generation just became a commodity four different vendors are fighting to sell you at scale. The place where the real economics of "shipping an app" now live — who deploys it, who hosts it, who gets to touch it once an AI agent wrote the first draft — sits one layer downstream, and it's barely being discussed. Once the code is free to generate, "who owns the running app" stops being a footnote and becomes the actual product decision.
The Deploy-Target Map: Four Vendors Bundle It, Three Don't
Look at where the vibe-coding revenue is actually landing and a pattern falls out immediately: the fastest-growing tools all bundle deployment into the same product that generates the code, while the infrastructure layer underneath them is a separate, much older market that's only just starting to court the same agents.
| Tool | 2026 scale | Deploy pattern |
|---|---|---|
| Lovable | $400M ARR (Feb 2026), up from $100M eight months earlier; $6.6B valuation | Bundled — one click ships to Lovable's own hosting |
| Bolt.new | $40M ARR in 5 months, $700M valuation, profitable | Bundled — deploys run on Bolt/StackBlitz infrastructure |
| Vercel v0 | ~$50M ARR (Apr 2026) | Bundled — wired directly into Vercel's own deploy pipeline |
| Replit Agent | $9B valuation (Mar 2026, tripled in 6 months), 50M+ users | Bundled — built-in deploy, database, and auth in one product |
| Railway | ships an MCP server | Open — any MCP-capable agent (Claude, Cursor, Codex) can provision, deploy, and read logs |
| Render | git-native, webhook-triggered builds on push | Open — any tool that can commit to a repo can trigger a deploy, regardless of which vendor wrote the commit |
| Fly.io | flyctl/API-driven deploys | Open — agnostic to whatever generated the Dockerfile |
The bundled column is where the venture money and the user growth are. The open column is where the deploy target doesn't care which of the four bundled tools — or which fifth one launches next quarter — produced the code in the first place. That's the entire question this piece is trying to answer: which side actually needs to win for a team shipping a real app, not a demo.
What the Bundled Tax Actually Costs Once the App Is Real
A prototype built in Lovable or Bolt in an afternoon is genuinely free of this problem — the bundling is the product for a one-off internal tool or a weekend project. The tax shows up the moment traffic, team size, or uptime expectations turn the prototype into something the business depends on.
At that point, three costs compound:
No portability without a rewrite. Replit Agent's pitch is that it's "the only one with a real built-in database and auth" among the major tools — which is exactly why an app built on it is hardest to move. The database, the auth flow, and the deploy target are the same platform-specific primitives; taking the Docker image elsewhere means re-plumbing all three, not just repointing a git push.
No pricing leverage. Compute for a bundled app is billed through the vendor's own credit system, folded into a subscription tier the team never negotiated. If that vendor reprices — and every hosted platform this space has watched over the past year, from Render's egress cuts to Fly.io's new line items, has repriced at least once — there's no fallback deploy target already provisioned to absorb the switch.
The agent is capped at whatever that one API supports. An AI agent operating inside a bundled tool can only do what that tool's own control plane exposes. It can't provision a second environment, roll back a bad deploy through a different path, or read logs from a system the vendor didn't build a hook for.
None of this makes the bundled tools bad products — Lovable's revenue trajectory says the opposite. It means the tax is invisible right up until the day the app stops being disposable.
What an Open Deploy Target Buys Instead
Railway's MCP server is the concrete version of the alternative: it lets Claude, Cursor, or any MCP-capable agent create a project, set environment variables, trigger a deploy, and read the resulting logs — the same actions a human would take through a dashboard, exposed as tools any agent can call. Render's webhook-triggered builds are the older, plainer version of the same idea: push a commit, a build runs, a service updates. Neither cares whether the commit came from a human, from Lovable's export button, from Cursor, or from a tool that doesn't exist yet.
That's the structural difference. git push and a documented deploy API are decades-old, boring interfaces that predate every vibe-coding vendor on the bundled list and will outlive most of them. An agent operating against that interface isn't locked to the tool that generated the first commit — it can keep working the app long after the codegen session that created it is over, using whichever model or tool is best for the next change.
Walk through what that actually looks like in practice: an agent finishes a change, calls a deploy tool exposed over MCP (or hits a webhook URL directly), the platform pulls the commit, runs the build, and the agent's next tool call reads back the deploy logs to confirm the service came up healthy — all without a human opening a dashboard in between. Swap Cursor for Claude Code, or Railway for Render, and nothing else in that sequence has to change. That's what "open" means in practice: the interface is the constant, not the vendor on either end of it.
This is the gap bex.co is built to close on the self-hosted side of that same open pattern: push a repo, get a running HTTPS service on machines you own, through a Render-compatible API that treats an AI agent as a first-class caller rather than a human clicking through a dashboard translated into API calls after the fact.
The Honest Caveat: Bundled Is Winning the Revenue Race Today
None of the above should read as "bundled is losing." By revenue, it's winning decisively. Lovable quadrupled ARR in eight months. Replit is guiding toward $1 billion in 2026 revenue on a user base that tripled its valuation in six months. For a non-technical builder or a PM who wants a working page today, one-click bundled deployment isn't a compromise — it's the entire value proposition, and portability is a problem they were never going to hit.
What the caveat doesn't change is where the fragility sits. Most vibe-coded apps never leave the tool's own preview environment, which is exactly why the lock-in cost stays hidden — it only bites the minority of apps that graduate from prototype to production system with real traffic, a team bigger than one person, and an on-call rotation. That graduation rate is small today. It's also the only part of this market where deploy-layer economics, not codegen-layer subscription revenue, decide who keeps the customer.
That graduation rate is also the variable the whole argument hinges on, and it's moving in one direction. The same analysts putting today's market at $4.7 billion project it reaching $12.3 billion by 2027 at roughly 38% annual growth — and growth at that rate means more teams crossing from "prototype in a tool" to "production app with paying users" every quarter, not fewer. Each one of those crossings is a fresh decision about which side of the deploy-target map it lands on.
Codegen Commoditized, Deploy Didn't
Ninety-two percent daily adoption is close to a ceiling — there isn't much room left for "will developers use AI to write code" to become a more interesting question. What's still wide open is which layer captures the durable relationship with a team once the novelty of generation wears off. A codegen tool gets used in bursts, during the session where the code gets written. A deploy target gets used every day after that, for as long as the app stays up.
That asymmetry is why "the code got easier to write" makes "who owns the running app" a more contested question, not a less important one. The four vendors above are still the fastest-growing part of this market. But the open deploy targets underneath them are the layer that doesn't get replaced when the fifth vibe-coding tool launches next quarter — it just gets another webhook pointed at it.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, and let any AI agent drive deploys through a Render-compatible API instead of being boxed into one vendor's control plane. Star the repo on GitHub or deploy your first app today.
Sources
- State of Vibe Coding 2026: Market Size, Adoption & Trends
- Vibe Coding in 2026: $4.7B Market, 84% More Apps, and Apple Is Worried
- Vibe coding statistics 2026: Adoption, productivity, and security data
- The 2026 AI Coding Platform Wars: Replit vs Windsurf vs Bolt.new vs Lovable
- Vercel v0 in 2026: AI App Builders Hit Production
- The Best Platforms to Deploy AI Apps in 2026
- What is a PaaS? A Developer's Guide for 2026.
