Vercel's Four Repricings Since 2024: What the Pattern Predicts for Your Next Bill Shock — and What a Fixed-Price Box Actually Fixes

If you budgeted a Vercel bill in early 2024 and haven't re-run the math since, your model is four pricing models out of date. Not four price changes — four pricing models. In eighteen months, the meters you're billed on were split, replaced, re-split, and finally converted into a credit system. Each change arrived on short, non-negotiable notice, and each one moved every tenant onto the new model whether their budget was sized for it or not.

Here's the timeline, because it's the whole argument:

The gaps between events: roughly ten months, then four, then three. Read honestly, that's a re-model every three to ten months, trending faster. If you're doing annual budgeting on a platform with that cadence, the question isn't whether your line item is wrong by Q3 — it's by how much, and in which dimension.

This post walks through the four events, reads the pattern, takes the strongest counter-argument seriously (spoiler: Hetzner raised prices too, and the difference is more interesting than "flat rate good"), and puts the same app on both worlds with real numbers.

Four Models in Eighteen Months​

April 2024: Two Meters Become Seven​

Before April 2024, Vercel billed two big things: bandwidth and function usage. The "improved infrastructure pricing" update, effective June 25, 2024, split those into granular meters — Edge Requests at $2 per million, Fast Data Transfer at $0.15/GB, ISR reads and writes as their own line items, and per-region rates for data transfer where one global price existed before.

Vercel's framing was optimization: granular meters let you tune each dimension. That's true. It's also true that a budget sized on "bandwidth plus functions" had no line for edge requests or ISR writes, and teams discovered those numbers only when the first re-modeled invoice arrived.

The failure mode isn't hypothetical. The most common bill-shock story of this era is mundane: a $20/month Pro plan that became $700, then $1,100, from nothing more exotic than a traffic spike hitting $0.15/GB overage past the included terabyte. The catastrophic version is the developer who woke up to a $23,000 bill after a DDoS, because attack traffic metered at the same bandwidth rate as customer traffic.

February 2025: Fluid Compute Changes What a Function Costs​

Fluid Compute changed the execution model: instead of one isolated instance per request, a single function instance serves many concurrent requests. Billing was still time-based at launch, but the meaning of the meter changed — concurrency shifts how duration and memory accumulate, so the same traffic produced a different bill even before any rate changed.

Winners: apps with I/O-heavy, concurrent workloads. Losers: anyone whose cost model assumed the old per-invocation isolation math. Either way, the spreadsheet from 2024 stopped describing reality.

June 2025: Active CPU — a Repricing Down Is Still a Re-Model​

In June 2025, Vercel introduced Active CPU pricing: you pay for CPU-active milliseconds ($0.128/hr), provisioned memory ($0.0106/GB-hr), and invocations — and I/O wait time costs nothing. For I/O-heavy workloads like AI inference gateways and database-bound APIs, bills dropped up to 90%.

This is the repricing that's easy to wave away because it was good news. But for budgeting purposes, direction doesn't matter — dimension does. A team that had just instrumented its cost model around duration-based billing now needed to understand active-CPU-versus-wall-clock profiles per route to predict its own invoice. The meter changed again, four months after the last change.

September 2025: Allocations Become Credits​

The Pro plan restructure (announced September 9, 2025; automatic rollout from October) replaced fixed per-product allocations across 20+ products with a $20 flexible monthly credit, plus $150+ dedicated to Fast Data Transfer and $20+ for Edge Requests. Viewer seats became free; spend alerts became default.

Vercel's own numbers: over 100,000 teams saw bills decrease or hold, while about 7% of teams saw increases. Seven percent of a six-figure customer base is a lot of surprised finance departments — and every team, increased or not, got migrated to a model with different failure modes than the one they'd sized their budget against. Three months after the previous re-model.

What the Pattern Predicts​

Four billing models in eighteen months is not a company that occasionally adjusts prices. It's a company for which the pricing model is an actively developed product surface — tuned, like any product, on a quarterly-ish release cycle.

That's not an accusation; it's structural. Three forces make frequent re-modeling near-inevitable for a growth-stage usage-based platform:

• Margin tuning. Usage-based pricing is a knob the business must turn as its own infrastructure costs shift. When AI workloads changed what compute costs underneath, Active CPU pricing followed.
• Competitive repositioning. Each re-model re-segments which customers are cheap and which are profitable. Granular meters in 2024 and credits in 2025 are both answers to "which workloads do we want?"
• Revenue pressure. A platform priced for land-grab growth eventually has to be priced for margin, and that transition doesn't happen in one step. It happens in four. Or more.

The budgeting consequence is the point most cost comparisons miss: on a frequently-repricing platform, you are not buying a price — you are subscribing to a pricing team's roadmap. A budget sized under model N isn't just inaccurate under model N+1; it can be unsized, because the dimensions changed. You can't forecast an ISR-write line item from a 2023 bill that didn't meter ISR writes. "Compare today's sticker price" undersells the risk precisely because the sticker keeps being reprinted in a different unit.

Predicting from the cadence — ten months, four, three — the safe planning assumption is: the model you're billed on today will not be the model you're billed on within a year, and possibly within a quarter.

"But Hetzner Raised Prices Too" — Yes, and the Difference Is the Whole Point​

The strongest objection to "buy a fixed-price box instead" landed in 2026, and it deserves the honest version, not a strawman: Hetzner — the canonical cheap-dedicated-box vendor — raised prices significantly this year. An April 1, 2026 adjustment, then a bigger one effective June 15, 2026: some shared-vCPU cloud instances up 140–210%, dedicated-vCPU instances more than doubling, and dedicated servers like the AX42 moving from roughly €47 to €57/month for new orders, driven by a DRAM and NVMe cost shock from AI demand.

So no — a fixed-price box is not immune to price movement, and pretending otherwise would falsify this whole article. What it's immune to is a different, worse thing. Three distinctions carry the argument:

1. Existing contracts were grandfathered. Hetzner's June 2026 increase applies to new orders and cloud rescales — a server you already rent keeps its contracted price unless you change its configuration. Vercel's re-models migrated every existing tenant onto the new model. One vendor raised the price of the next box; the other changed what your current bill measures.

2. A price-level change is not a price-model change. When Hetzner raises a flat monthly rate, updating your budget is one multiplication: €47 becomes €57, done in the time it takes to read the email. When Vercel re-models, you have to re-instrument your application — profile active-CPU versus I/O-wait per route, estimate edge requests and ISR writes — just to know what you'll pay. The cost of the change isn't only the delta; it's the modeling work the new dimensions demand.

3. Cadence. Hetzner's 2026 increases were its first broad repricing in years, driven by a documented hardware-market shock. Vercel re-modeled four times in eighteen months as routine product iteration. One is a shock you absorb once a decade; the other is a subscription to uncertainty.

"Fixed-price," properly stated, means: fixed for the contract you hold, in a unit that doesn't change. That claim survived 2026. The hosted-platform equivalent — "your bill's dimensions are stable" — did not survive any single year since 2023.

The Same App, Both Worlds​

Take a representative production app: an SSR web app on a small team (3 developer seats), moderate compute (~100 active-CPU hours and ~5M edge requests a month), and the variable that actually drives the result — egress between 500 GB and 1.5 TB a month.

Vercel, current model (Pro): $20/seat × 3 = $60, plus usage. Using current meter rates — Active CPU at $0.128/hr ($13), Edge Requests at $2/M ($10, partially inside included credits), and Fast Data Transfer at $0.15/GB past the included allocation:

The spread between the low and high egress cases is roughly 2x — and that's under today's model, with no traffic spike, no DDoS, and no repricing event between now and your next invoice. The $23,000 DDoS bill is this same table with the egress row set to "unbounded."

Hetzner AX42 running a self-hosted PaaS: €57.30/month at 2026 new-order prices (€46.52 if you rented it before the adjustment — grandfathering in action), with unlimited traffic on a 1 Gbps port:

The egress sensitivity — the variable that produced every bill-shock story in this article — is literally zero. The number your finance team writes down in January is the number on the December invoice.

Now the honest fine print, because a flat box is not a free lunch: that €57.30 does not include your operations time (patching, monitoring, backups — realistically hours per month that a hosted platform absorbs), and there's no built-in CDN or DDoS absorption layer, so the $23,000-attack scenario becomes an availability problem for you instead of a billing problem. You'd front it with a CDN or filtering layer regardless. What you're buying with the flat rate isn't zero cost — it's zero variance in the dimensions a pricing team controls.

Budgeting Rules for a Repricing World​

Whatever you host on, three rules fall out of the pattern:

1. Assume a re-model every 3–10 months, and re-run your bill within a week of each announcement. Not "read the blog post" — actually recompute your invoice under the new dimensions. The teams in the 7% found out from the invoice; the announcement was available a month earlier.
2. Cap spend everywhere the platform allows. Vercel now ships spend limits enabled by default with a $200 default cap for new teams — set yours deliberately, not at the default. An outage at your cap is recoverable; a $23,000 invoice is a negotiation.
3. Keep egress portable. Every repricing era in this timeline landed hardest on bandwidth. Serving heavy assets from a flat-rate box or object store — even while the app stays hosted — removes the single biggest variance term from your bill.

And know which side of the trade you're on. If your app lives comfortably inside a hosted platform's included tiers and you have zero ops capacity, the hosted platform is still the right call — repricing risk on a $0–20 bill is noise. The owned-box route wins when the usage-based bill has become a three-digit line item with two-digit variance, when egress is your dominant cost, or when you're running a fleet where per-service platform pricing multiplies. At that point, a fixed-rate machine plus a self-hosted PaaS layer gives you the deploy experience without the pricing-roadmap subscription — and the only repricing you're exposed to is the one that comes with your next hardware contract, not your current one.

Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, at whatever flat rate your hardware costs. Star the repo on GitHub or deploy your first app today.

Sources:

• Vercel: Improved infrastructure pricing
• Vercel: Introducing Active CPU pricing for Fluid compute
• Vercel changelog: Lower pricing with Active CPU pricing
• Vercel: A more flexible Pro plan for modern teams
• Flexprice: Breaking down Vercel's 2025 pricing plans, quotas and hidden costs
• UsageBox: The $23,000 Vercel bill
• Deploybase: Vercel bill shock — $1,100 bandwidth bills
• Hetzner: Statement on price adjustment as of April 1st 2026
• Hetzner Docs: Price Adjustment 15 June 2026
• WZ-IT: Hetzner price increase June 2026 — CPX and CCX up to +176%
• Hetzner: New AX42 announcement