Vercel Flat Rate CDN: who saves and who pays more

On 8 September 2026 Vercel made Flat Rate CDN generally available: instead of paying per gigabyte and per request, a Pro team picks a monthly capacity tier. It is on by default for new Pro teams, and existing teams can opt in. Most write-ups lead with “50 TB for $20”. That part is real. But the tier you land in is set by CDN requests, not bandwidth, and that is where people will guess wrong.

The short answer

• Small sites (under ~1M CDN requests a month) pay nothing extra either way. Flat Rate only changes things for you if you have a spike.
• Sites with real traffic and normal-weight pages save. How much depends on how much you would have paid for data transfer, and on whether your visitors are outside the US, where requests cost more.
• Request-heavy sites with light pages, especially just past a tier edge, can pay more than on demand. A tier costs exactly what its request allowance would cost on demand in US regions. The saving only comes from bandwidth.
• Your function bill does not change. Active CPU, provisioned memory, Fast Origin Transfer, ISR and image optimization are still metered.

The tiers

Prices are from Vercel’s docs, checked 2026-09-29. The last column translates requests into page views at our assumption of 20 CDN requests per page view (see how to measure yours below).

How usage is judged: if a whole billing cycle goes over your tier, Vercel moves you up one tier from the next cycle. A one-day spike doesn’t. Above the $300 tier you drop back to on-demand pricing. If you change or switch off Flat Rate yourself, that also takes effect at the next cycle. Fair-use rules exclude bulk downloads, media-heavy delivery and using the CDN as file hosting. Those projects can be moved to Vercel’s Flex CDN network or onto another plan.

What the tier covers and what it doesn’t

The $20 monthly usage credit that comes with Pro still applies to metered resources. On on-demand pricing it gets used up by CDN usage first. On Flat Rate it is left for your functions.

The detail most write-ups miss

On demand, CDN requests cost $2.00 per million in US regions, $2.60 in Frankfurt and up to $3.20 in the most expensive region. Now look at the tier caps:

Every paid tier costs exactly what its full request allowance would cost on demand at US rates. So a US-heavy site that uses its whole allowance only saves on data transfer, and one that uses a fraction of it pays for requests it never makes. The real discounts are the bandwidth (up to 50 TB that would cost $0.15–0.35/GB) and cheaper requests for non-US traffic.

Four traffic levels, priced both ways

These are our standard scenarios, the same ones behind every cost page. Transfer comes from visits × pages × page weight, and requests assume 20 per page view. The on-demand columns show the CDN part of the bill only, before the $20 credit.

How to read it:

• Hobby: fits the included tier. On demand it would also be covered by the $20 credit. No difference.
• Startup: at 20 requests per view this site passes 1M requests and lands on the $20 tier, while its on-demand CDN cost ($17.10 US) would have fit inside the $20 credit. It is roughly a wash, and slightly worse on Flat Rate if your functions would have used the credit. The upside is protection from spikes.
• E-commerce and high traffic: Flat Rate is cheaper, by $57.30 and $39.75 a month at US rates, because the transfer (782 GB and 1,465 GB) is now included. The saving is smaller than “50 TB for $20” suggests: at this request volume most of the tier price is paying for requests.

Requests per page view change everything

The same scenarios with a lean page (10 requests), our middle case (20) and a heavy page with many images, fonts and prefetches (40). Each cell shows Flat Rate / on-demand US.

Green: Flat Rate is cheaper. Red: on demand is cheaper, before counting the $20 credit and the value of spike protection.

The tier-edge trap

Tiers are steps. Take a content site doing 11M requests and only 200 GB a month: lots of small pages, good compression. It is just over the 10M cap, so after one full cycle it moves to the $100 tier. On demand, the same month would cost $52.00 at US rates. That is $48 a month for a buffer of 39M requests you don’t use.

What to do if that’s you:

• Cut requests. Vercel’s own advice: look for images that re-mount, polling, and SWR/React Query refetch-on-focus. Check prefetching too, since every prefetched link is a request.
• Exclude projects that aren’t critical (docs, staging-like marketing pages) from capacity. They move to the Flex CDN network. This needs Vercel DNS.
• Switch back to on-demand in Billing. It applies from the end of the current cycle, so decide before the cycle rolls over.

Measure your own number in five minutes

1. In the Vercel dashboard open Usage → Edge Requests for the last full month, or run vercel usage, which breaks usage down by day.
2. Take page views for the same month from your analytics.
3. Edge Requests ÷ page views = your requests per view. Multiply your expected page views by it, then match the result against the tier table above. Do the same with Fast Data Transfer in GB.
4. If you are on on-demand pricing today, compare that to your last invoice’s CDN lines (Fast Data Transfer + Edge Requests), not the whole bill.

Does this change the case for leaving Vercel?

Partly. Before September, bandwidth was the most common reason a Vercel bill blew up, and Flat Rate mostly fixes that for normal web apps. Here is the same Next.js app on a few alternatives, from our live dataset (as of 2026-10-05):

The Vercel row is Pro ($20) plus the Flat Rate tier the scenario lands in at 20 requests per view. Our cost pages don’t price requests yet, so they show Vercel lower on request-heavy sites. No row includes function compute or extra seats. Hetzner has no CDN, previews or image optimization built in. Cloudflare Workers Paid ($5) includes 10M requests and free static asset requests, but its 128MB Worker memory is below our server-rendered Next.js profile, so it only fits static or lean edge builds.

Where leaving still makes sense: when function compute or seats drive the bill (a long-running server on a VPS or PaaS costs the same at 10 or 10,000 requests a minute), when the traffic is bulk media that fair use excludes, or when you would sit above the $300 tier. For everything else, turning on Flat Rate is a lot less work than a migration. Check it before you plan one.

Methodology: Vercel rates are official list prices from vercel.com/docs, checked 2026-09-29, in USD and excluding VAT. On-demand columns show the CDN part only (Fast Data Transfer + CDN requests) at the named region’s rate. Real bills mix regions. Requests per page view is an assumption, not a measurement, so change it to yours. 1 TB is counted as 1,000 GB. None of the providers in this article pay us a commission.