Dora NodaIn September 2025, two companies sold essentially the same thing — git-push frontend hosting with previews, edge delivery, and serverless functions — and the market priced them like different species. Vercel closed a $300 million Series F at a $9.3 billion valuation, led by Accel and Singapore's GIC, with BlackRock, StepStone, and Khosla Ventures piling into an oversubscribed round. Netlify, the category's other original entrant, spent the same period restructuring down to a reported roughly 180 employees and bolting AI features onto its existing platform. Same pitch, opposite trajectories — and the gap has only widened since, with Vercel's revenue run-rate hitting $340 million by February 2026 and its CEO openly signaling IPO readiness that April.
So here is the verdict up front, with the receipts below:
| Signal | Vercel | Netlify |
|---|---|---|
| Valuation / funding | $9.3B post-money, $300M Series F (Sept 2025) | No new primary round; restructuring instead |
| Revenue trajectory | $100M ARR (early 2024) → $340M run-rate (Feb 2026) | Undisclosed; slower growth, headcount cuts |
| Team direction | Hiring, IPO signals, v0 at 3.5M+ users | Successive layoffs since 2023, ~180 reported staff |
| AI bet | v0 + AI Cloud as the product; ~30% of hosted apps agent-built | Agent Runners + AI Gateway as add-ons (Oct 2025) |
| Pricing shape | 5-dimension meter + $20/seat | Credit meter + flat $20/mo Pro, unlimited seats |
The question this split answers is the one every frontend team is implicitly asking: does a frontend PaaS compete on price, on AI-agent tooling, or on neither — now that the git-push workflow itself can be self-hosted on owned hardware for about $5 a month? The evidence says price was never the moat, the agent-to-deploy loop is becoming the moat, and Netlify's late pivot proves it by reading as defense while Vercel's reads as offense. The rest of this post is the numbers behind each cell, a worked three-way bill for a typical team, and the decision framework for when to stay and when to leave.
How Vercel pulled ahead: the numbers
Start with the top line, because it is unusually clean. Vercel's ARR went from $100 million at the beginning of 2024 to a $340 million run-rate by the end of February 2026 — more than tripling in two years — with revenue up 82% and the user base doubling in the year before the Series F (Reuters, TechCrunch). Customers now include OpenAI, Anthropic, PayPal, Nike, and Walmart — enterprise logos that pay enterprise contracts, not just seat fees.
The engine underneath that growth is not cheaper bandwidth. It is agents. CEO Guillermo Rauch told the HumanX crowd in April 2026 that roughly 30% of the applications running on Vercel's platform were already generated by AI agents, framed the company's IPO readiness around that surge, and put the thesis in one line: all that agent-generated software has to go somewhere, "and we think it's going to be Vercel." The v0 vibe-coding tool passed 3.5 million users, v0 Mobile was queued for launch, and the raise was explicitly earmarked for AI Cloud services and the v0 agent — plus tuck-in acquisitions like Egypt's Stakpak in July 2026 to deepen the AI infrastructure bench.
That is what first-mover conviction spending looks like: the AI bet is the product roadmap, funded by a tripled valuation, with revenue compounding behind it. Contrast that with the other column.
Netlify's retrenchment and its AI pivot
Netlify's last few years read as the mirror image. Successive restructuring rounds starting in 2023 shrank the company toward a reported roughly 180 employees — Glassdoor's 51–200 band and the public layoff threads tell the direction even where exact figures stay private — while growth stalled against the competitor tripling above it.
Then, in October 2025, came the AI pivot: Agent Runners, which let teams point an agent at their production backlog from the Netlify dashboard, plus an AI Gateway offering model access inside the platform. On paper this matches Vercel's agent story bullet for bullet. In practice the surrounding details tell a different story — one of cost control, not conviction.
First, the pricing underneath it. Netlify moved to credit-based billing (compute around $0.07/GB-hour, bandwidth around $0.13/GB in credit terms) and in April 2026 killed per-seat pricing entirely, making Pro a flat $20 a month for unlimited team members. That is a retention discount dressed as simplification: when your rival charges $20 per seat, going flat-$20-unlimited is how a shrinking vendor stops team-size growth from becoming a churn trigger.
Second, the guardrails arrived before the momentum did. By March 2026 there was a team-wide credit cap that froze every member's agent runs at once; by July 2026 Netlify had to ship per-member AI spend limits — including a zero-credit kill switch — because runaway agent spend was a live customer problem.
Capping the cost of your flagship feature within nine months of launch is what you do when AI is a meter you resell, not a loop you own. Third, the database bundling: Netlify Database folded managed Postgres into the platform bill in 2025, joining Vercel's storage integrations and Cloudflare D1 in treating "we also run your data layer" as a retention feature — one more surface making the invoice stickier while the underlying hosting pitch stays unchanged.
None of this is irrational — it is exactly what differentiation-of-necessity looks like. When the core product (static hosting plus functions) no longer pulls decisively ahead, you bundle the adjacent layers (models, agents, Postgres) and flatten the seat tax so nobody leaves over arithmetic. Vercel spends its AI budget acquiring model-tooling startups; Netlify spends its AI budget building spend caps. Both companies now sell "agents plus hosting." Only one of them is growing because of it.
The pricing math: one team, three bills, three months
Assertions about moats are cheap, so here is the concrete artifact: a worked monthly total for a typical small team — a marketing site plus one SSR app, five developers, preview deploys on every PR — across Vercel, Netlify, and a flat Hetzner box. Assumptions stated so you can check the arithmetic: Vercel Pro at $20/seat with ~1 TB included bandwidth, overage bandwidth at $0.15/GB, invocations at $0.60 per million, active-CPU compute at $0.128/hour; Netlify Pro at flat $20/mo unlimited seats with bandwidth around $0.13/GB in credit terms; Hetzner CX22 (2 vCPUs, 4 GB RAM, 40 GB disk, 20 TB included traffic) at roughly $5/mo. Compute burn is estimated at $30–60/mo on either metered platform for this shape.
| Month | Vercel (5 seats) | Netlify (Pro) | Hetzner CX22 |
|---|---|---|---|
| Quiet: 20 GB egress, ~200k invocations | ~$100–120 (seats dominate; usage inside allowances) | ~$20–30 (flat fee + pocket-change credits) | ~$5, awake all month |
| Typical: 400 GB egress, ~3M invocations | ~$140–170 ($100 seats + ~$40–70 usage) | ~$70–100 ($20 + ~$50 bandwidth + compute credits) | ~$5, egress included |
| Spike: 2 TB egress (launch month, 5x traffic) | ~$290–320 ($150+ overage past 1 TB included) | ~$220–280 (overage with no included pool to absorb it) | ~$5, still inside 20 TB |
Three honest readings fall out of this table. First, Netlify's flat $20 Pro genuinely undercuts Vercel for small teams in normal months — dropping the seat tax was the right defensive move, and the per-seat model is Vercel's most attackable line item at roughly $100 before a single request is served. Second, the spike row is where both meters bite and the flat box wins by 40–60x: neither platform's overage structure forgives a launch month, and no workspace fee softens it. Third — and this is the row the flat-hardware triumphalists skip — the quiet month belongs to the meters: a side project sipping 20 GB costs nearly nothing on Netlify's credits and fits Vercel's Hobby allowances, while the box charges its $5 whether anyone visits or not. Scale-to-zero still wins at idle; it just stops winning the moment real traffic or real seats show up.
The sensitivity range is the point, not any single cell. Below ~10% utilization or inside free allowances, metered platforms are cheaper than owned hardware. Past a few hundred GB of egress or a handful of seats, the ranking inverts hard — and it inverts on the exact variables (traffic success, team growth) you least want to be taxed on.
Where self-hosting actually competes
Which returns us to the title question. The Vercel–Netlify split proves the git-push workflow is not the moat: Netlify sells the same deploy-from-git experience, charges less for it in most of the table above, and is still losing. If price were the competition, the cheaper, flat-fee vendor would be winning. It is not, because Vercel stopped selling hosting and started selling the place agent-generated software lands — v0 generates it, AI Cloud and Sandbox run it, the hosting is the exhaust.
That reframes where self-hosting competes. The deploy-from-git workflow is thoroughly commoditized: Coolify, Dokploy, CapRover, and open platforms replicate previews, custom domains, and TLS on a $5 box. No team needs to pay a per-seat meter for the workflow itself anymore. What is not commoditized is the agent-to-deploy loop — sandboxed execution with usage metering (Vercel Sandbox bills active-CPU hours precisely so idle agent time is free), model gateway access, and review-to-production automation — plus enterprise distribution into accounts like the OpenAI and Anthropic logos on Vercel's customer slide. A self-hosted PaaS competes exactly there or not at all: push a repo, get a running HTTPS service, and give the team's agents a sandboxed, metered place to build and deploy without a per-seat tollbooth.
The decision framework writes itself from the table:
- Stay on a frontend PaaS when you are idle-light or seat-light — side projects inside free allowances, tiny teams where $20 flat beats operating a box, launch windows where someone else's incident response is worth the overage risk.
- Leave when the meter taxes your growth variables: egress climbing past hundreds of GB, seats multiplying a $20-per-head fee, or agent sandbox volume crossing the breakeven where flat hardware with no per-second meter wins by an order of magnitude.
- Either way, revisit when the invoice's most interesting line is the meter rather than the product — that is the migration stories' common rhyme, and it now applies to AI credits just as it once applied to bandwidth.
The compressing middle
Step back and the pattern is bigger than two vendors. Frontend hosting margins are compressing toward zero — Cloudflare Pages ships unlimited bandwidth, Netlify flattened seats to $20, free tiers cover every side project — while all the investment flows one layer up, into whoever owns the agent that writes the app and the sandbox that runs it. Vercel's $9.3 billion is a bet that the deploy target for agent-generated software is worth more than the hosting underneath it. Netlify's restructuring-plus-bundling is the same realization arriving as defense.
For a team choosing infrastructure in 2026, that is clarifying. Don't pick a frontend platform for its git-push workflow — that part is free everywhere, including on hardware you own. Pick it, or leave it, based on who owns your agent loop and what they charge per unit of it. And if the answer is "a per-seat fee plus five usage dimensions plus a credit cap with a kill switch," price the flat box. It is still $5, in every row of the table.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, with flat hardware economics instead of per-seat or per-second meters. Star the repo on GitHub or deploy your first app today.
