Dora NodaOne year ago this month, Netlify told new customers the seat was dead. On September 4, 2025, every new account landed on credit-based billing: one pool of credits per team, drained by bandwidth, compute, requests, and deploys alike. Twelve months, two repricings, and one seat-count reversal later, we can finally answer the question that matters: what does the credit model actually do to a team's month-to-month bill versus the seat pricing it replaced?
The short answer: the bill got cheaper for big teams, more expensive for small ones, and less predictable for everyone. Here is the math.
One year of cloud rent: three dates that reshaped the bill
The credit era did not arrive all at once. Three beats define the first year:
- September 4, 2025 — credit-based plans launch for all new accounts. The Free plan gets 300 credits a month behind a hard limit; paid plans get monthly allotments plus auto-recharge. Pre-existing teams keep legacy seat plans, grandfathered, no action required.
- April 14, 2026 — Netlify removes per-seat charges on Pro entirely (unlimited seats at a flat $20/month base) and simultaneously raises the credit price of bandwidth, compute, and web requests. The seat cross-subsidy moves: headcount stops driving the bill, consumption drives all of it.
- July 14, 2026 — Pro stops being a single 3,000-credit tier and becomes a ladder: 3,000 credits at $20, then 5,000 at $33, 10,000 at $63, 15,000 at $95, and 20,000 at $126, with unused-credit rollover on the 5,000-and-up tiers — an explicit concession to teams with seasonal surges.
That arc — launch metered billing, then remove seats, then add tiers with rollover — is the industry's standard walk toward consumption pricing. The question is what it costs at each step.
What 3,000 credits actually buy
Netlify's published credit rates (current as of this writing) are:
| Meter | Unit | Credits |
|---|---|---|
| Production deploy | 1 deploy | 15 |
| Bandwidth (web + database) | 1 GB | 20 |
| Web requests | 10,000 requests | 2 |
| Compute (functions, preview servers, agents, DB) | 1 GB-hour | 10 |
| AI inference (Agent Runners, AI Gateway) | $1 of model spend | 180 |
Preview and branch deploys, failed deploys, and rollbacks are free; only published production deploys draw 15 credits each. Run out of credits and the meter stops politely: the Free plan hard-pauses production deploys and agent runs, while paid plans auto-recharge (1,500 credits for $10 on Pro) or let you buy credit packs.
Now translate the $20 base tier's 3,000 credits into capacity. Spent purely on bandwidth, the whole month's allotment buys 150 GB. Under legacy Pro at $19 per member, the bundled bandwidth allowance was 400 GB. That single comparison captures the trade: the base tier looks $1 more expensive than one legacy seat while carrying less than half the bandwidth — and that is before a single function invocation or deploy draws from the same pool.
The worked comparison: same team, old bill vs new bill
Take a typical small-team workload — a marketing site plus an app: 200 GB of bandwidth, 2 million requests, 60 production deploys, and 100 GB-hours of function compute in a month. Under today's rates that is 4,000 + 400 + 900 + 1,000 = 6,300 credits.
Pricing that workload old versus new:
| Team | Legacy seat bill | Credit-model bill |
|---|---|---|
| Solo dev, typical month | $19 (400 GB allowance covers it) | ~$50–63 (base $20 + recharges, or the 10k tier at $63) |
| 5-person team, typical month | $95 (5 × $19) | ~$50–63 (same usage, seats now free) |
| Any team, quiet month (~3,100 credits) | Same as above (seats don't vary) | $20 (base tier covers it) |
| Any team, 2x spike month (~12,600 credits) | Same as above, absent overage | ~$90–95 (base + recharges, or the 15k tier) |
Two things stand out. First, the cross-subsidy flip: the solo dev pays roughly triple the legacy bill for identical usage, while the five-person team pays less than before — Netlify moved the cost from headcount to consumption, and small teams absorbed it. Second, the variance: the same team's bill swings from $20 in a quiet month to ~$90 in a spike month, where the legacy bill would not have moved at all. Predictable seats became a meter that follows your traffic.
The volatility is the product
Spiky workloads are where the single-pool design bites. A viral month pushing 2 TB of bandwidth burns 40,000 credits on bandwidth alone — about $267 in auto-recharges at 1,500 credits per $10. To be fair, the legacy overage math was not kinder: 1.6 TB beyond the 400 GB allowance at $20 per 100 GB is $320 before the base plan. The credit model did not invent the viral-month invoice; Netlify's infamous $104k static-site bill dates to the pre-credits era, when automatic bandwidth overage already had teeth.
What changed is the shape of the risk. Under seats-plus-overages, only bandwidth could surprise you. Under credits, every meter shares one pool with one exhaustion behavior: a deploy-heavy release week, a function retry storm, and a traffic spike all draw down the same balance, and an empty pool pauses production deploys until credits arrive. There is no isolation between "the site got popular" and "CI misfired fifty times."
Netlify knows the shared pool is dangerous, because it built guardrails that admit it. In March 2026 the company shipped a team-wide credit cap covering Agent Runners and AI Gateway spend; in July 2026 it added per-member ceilings, including setting a member to zero to cut off their agent access entirely. That is a kill switch for the exact failure the model creates: AI agent runs — billed at 180 credits per $1 of underlying model spend, a ~20% markup at recharge rates — draining the same credits that keep production sites deploying. When the vendor ships a per-human circuit breaker for its own billing unit, believe the meter is the risk surface.
Everyone metered at once
Netlify is an instance, not an outlier. Within days of its September 2025 launch, Vercel replaced per-product allowances with $20-per-seat Pro plus a flexible usage credit (each seat brings $20 of spendable credit; bandwidth overage sits at $0.40/GB). In April 2026, Render inverted from the other direction: flat $25 Pro and $499 Scale workspace plans with unlimited members, announced April 23 and force-migrated from $19–29 per-seat legacy plans on August 1 — while keeping compute itself usage-metered.
Three vendors, three routes to the same destination: headcount decoupled from price, consumption metered in fine units, and the bill following workload shape month to month. The Hacker News thread greeting Netlify's announcement said it plainly — "subscriptions aren't enough, we're now moving towards credits" — and a year later that reads less like a complaint than a sector forecast. Even the Free tier changed character: Netlify's forums now carry confused threads about sites paused on "operational credits," a failure mode seat plans never had.
The flat alternative
Contrast all of this with the cost curve of machines you own. Hetzner dedicated and auction servers run roughly €46–70 a month with unlimited 1 Gbit traffic; whether the box serves 150 GB or 2 TB, runs 60 deploys or 600, the invoice is the same number. Builds, function invocations, edge requests, and preview environments all cost exactly $0 at the margin — the meters that make up the entire credit bill simply do not exist.
That flatness is the whole argument. The credit model charges you most in exactly the months you can least afford a surprise — the launch, the viral spike, the incident-week deploy storm. Owned hardware charges you the same in all of them. The trade is real operational work (OS, TLS, deploys, patching) versus a bill that can triple between quiet and spike months with no warning until the recharge lands. Teams should price that trade honestly: take the ~$50–95/month the worked example above spends on credits, multiply by twelve, and ask what a flat box plus a day of automation buys instead.
What year two will decide
Netlify spent year one sanding the model's edges — unlimited seats, tiered allotments, rollover, AI kill switches. Each fix concedes the same point: pure single-pool metering is too volatile to sell unmodified. Year two's question is whether the platform keeps bolting predictability back on (more caps, more rollover, per-meter budgets) or whether teams conclude the predictability was the product all along, and go find it in flat-rate infrastructure.
Watch the grandfathered legacy plans. As long as pre-September-2025 teams can stay on seats, Netlify is running a live A/B test between the two philosophies on its own customer base. The day those plans are retired, the experiment ends — and every team still renting will be on the meter, like it or not.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, with a bill that stays flat when your traffic doesn't. Star the repo on GitHub or deploy your first app today.
