· 11 min read
Vercel Blob LimitsReached? Why My SiteWent Dark TwiceVercel Blob limits reached and every image returns 403? Find which Hobby meter tripped, why video and hotlinked images burn it, and the fix that held.
Twice in two months, every image on my portfolio went blank. The pages loaded, the layout held, and every image that came from Vercel Blob was gone. The project thumbnails, the blog covers and the testimonial avatars all returned 403 Your store is blocked. The Storage dashboard showed "Limits reached" on every Blob store in the account, including one that belonged to a different project.
If you are searching for Vercel Blob limits reached, the important detail is this: the two incidents tripped two different meters, and they had two different causes. Storage was never the problem. My store held 102 MB of a 1 GB allowance. This post covers how I identified which meter tripped, what spent it, and the changes that keep Blob inside the free tier now.
What "Vercel Blob limits reached" actually means on Hobby
On the Hobby plan, Blob is free inside four monthly allowances. Vercel's Blob pricing page lists them, and its wording on overage is direct: you do not pay for extra usage, but "you will not be able to access Vercel Blob if limits are exceeded", and you "have to wait until 30 days have passed before using Blob storage again."
| Meter (Hobby included) | What spends it | Incident on my site |
|---|---|---|
| Storage size: 1 GB/month | Bytes you keep in the store | 102 MB, never the problem |
| Simple Operations: first 10,000 | A blob read by URL that is a cache MISS, plus head() | 10.1k / 10k in August |
| Advanced Operations: first 2,000 | put(), copy(), list(), and browsing the store in the dashboard | Not hit |
| Blob Data Transfer: first 10 GB | Bytes downloaded or viewed | 10.7 GB / 10 GB in September |
What I learned from the two blocks, which the docs do not spell out as clearly:
- It is a hard block, not throttling. Public blob URLs return 403. Any
/_next/imageURL that was not already cached returns 502, because the optimizer cannot fetch its source. - It is account-wide. A store in a different project, well inside every limit, was blocked too. Separate stores keep your data apart, but they do not split the quota.
- The 30 days start at the overage, not on the 1st. The store banner reads "Access resumes on <date>". Read that date. Do not wait for the billing month to roll over.
- Fixing the cause does not unblock you. On Hobby, only moving to Pro lifts the block early.
- You cannot export while blocked. Content downloads return 403. If Blob holds the only copy of your uploads, you cannot migrate them away until the block lifts. I now keep local masters of every uploaded asset.
Step 1: find out which meter tripped
Open the blocked store in the dashboard. The counter shown in red is the meter you exceeded. Do this before you change any code, because the fix for operations and the fix for data transfer are different.
Then find the day it happened. On Hobby, Observability → Blob shows only the last 12 hours, so by the time you notice the block, the per-file history has already gone. Use Usage → Blob instead and open Blob Data Transfer or Simple Operations. Both show a per-day chart for 30 days. A flat baseline with one tall bar dates the culprit. After that, run git:
bash
# Everything that shipped since the day before the spike
git log --since="2026-09-28" --oneline
In my September incident, the baseline was about 0.15–0.3 GB a day. Then one day used 4.1 GB, and the next day used about 1 GB. On the spike day I had shipped a new feature, and git log named the commit.
Incident 1: 10,000 invisible operations from hotlinked images
In August the red counter read Simple Operations 10.1k / 10k. The cause was ordinary. Every <img src>, and the Open Graph and Twitter image meta tags, pointed straight at *.public.blob.vercel-storage.com.
Each request for a blob URL that misses the CDN cache counts as one Simple Operation. A portfolio gets real visitors. It also gets crawlers, link-preview bots and social scrapers from every region, and many of them arrive at a cold cache. Ten thousand misses a month is not much when every image on every page is a direct Blob read.
The fix is to put a cache that I control in front of Blob. In Next.js, /_next/image already is that cache. I added a small helper and used it on every Blob URL in the site:
ts
const REMOTE = /^https?:\/\//;
export function imgProxy(src: string, width: 640 | 1080 | 1200 = 1080, quality = 75): string {
if (!src || src.startsWith('/_next/image') || src.startsWith('data:')) return src;
if (!REMOTE.test(src) && !src.startsWith('/api/')) return src;
return `/_next/image?url=${encodeURIComponent(src)}&w=${width}&q=${quality}`;
}
The helper wraps blog cards (640), article covers (1200), markdown body images (1080), testimonial avatars, <video poster> and the absolute OG image URL. Then I made the optimizer's cache last, and I made Blob's own cache headers long:
js
// next.config.mjs
images: {
minimumCacheTTL: 2678400, // 31 days
remotePatterns: [
{ protocol: 'https', hostname: '*.public.blob.vercel-storage.com' },
],
},
ts
// upload script
await put(name, buffer, {
access: 'public',
contentType,
allowOverwrite: true,
cacheControlMaxAge: 31536000, // 1 year; the SDK default is one month
});
There is a trade-off. /_next/image has its own Hobby meter. Vercel's Image Optimization limits include 5,000 transformations a month, billed on each cache MISS or STALE. That is why the helper accepts only three widths and why the TTL is long. A small set of fixed widths, each cached for a month, gives a bounded number of transformations. Free-form widths would multiply them. After this deploy, the operations count dropped.
Is your marketing site's media breaking and nobody can say why? This kind of failure looks like "the images are broken" to a founder and like "it works on my machine" to whoever built the site. Tell me what your stack is and I'll find the meter that is leaking.
Incident 2: one video feature, 4 GB of egress in a day
The second block came at the end of September. The red counter read Blob Data Transfer 10.7 GB / 10 GB. Operations were at 7.3k / 10k, which was close but not the cause.
The spike day was the day I shipped a video "loupe" on the /work archive: a hover lens that plays a short loop of each project. The component built its sources like this:
ts
function reelSources(project: WorkProject): IReelSource[] {
const loop: IReelSource[] = [];
if (project.loopVideoWebm) loop.push({ src: project.loopVideoWebm, type: 'video/webm' });
if (project.loopVideo) loop.push({ src: project.loopVideo, type: 'video/mp4' });
if (loop.length > 0 || !project.showreelVideo) return loop;
// The fallback that cost a month of Blob:
const type = /\.webm(\?|$)/i.test(project.showreelVideo) ? 'video/webm' : 'video/mp4';
return [{ src: project.showreelVideo, type }];
}
When a project had no short loop, the component fell back to the full showreel. Those files are 3–15 MB each, and there were 13 of them, 71 MB in total. The <video> elements used preload="auto", which allows the browser to download the whole file before anyone hovers. As a result, a single view of the /work page could pull most of the showreel library from Blob.
The second multiplier was my own development work. The local dev server, every Vercel preview deployment and every Playwright verification run read from the same production store. A busy day of building and checking the feature spent real quota at the same rate as real visitors.
The fix removes the fallback. If no short loop exists, the component plays no video and shows the poster image.
ts
function reelSources(project: WorkProject): IReelSource[] {
const loop: IReelSource[] = [];
if (project.loopVideoWebm) loop.push({ src: project.loopVideoWebm, type: 'video/webm' });
if (project.loopVideo) loop.push({ src: project.loopVideo, type: 'video/mp4' });
return loop;
}
The broader rule I follow now: on Hobby, do not stream video from Blob. Encode short loops of 6–10 seconds and ship them from public/, or host video on a CDN that does not charge for egress. Cloudflare R2, for example, does not charge egress fees. A full showreel belongs on a page where the visitor clicks play, not in a preloaded hover effect.
The baseline leak: canvas and WebGL code that skipped the proxy
The spike explained the block, but it did not explain the 0.15–0.3 GB/day baseline. The <img> tags had used the proxy since August. The remaining leak was in code that never renders an <img> at all.
My homepage samples each project thumbnail into a canvas and turns it into a tone grid for a WebGL LED effect. The loader looks like this:
ts
const img = new Image();
img.decoding = 'async';
img.crossOrigin = 'anonymous'; // needed to read pixels from a cross-origin host
img.src = src; // was the raw Blob URL
await img.decode();
new Image(), THREE.TextureLoader, drei's useTexture, and <Image unoptimized> all fetch the URL you give them. None of them goes through the optimizer unless you send it there. Each one was a full-resolution read from Blob, on every page view that missed the cache.
The fix is the same helper at a smaller size:
ts
loadImageEl(imgProxy(project.src, 640)).then((img) => {
if (!img) return;
const tone = sampleImageTone(img);
// ...
});
This change has two benefits. First, /_next/image is on my own origin, so the canvas is not tainted and pixel sampling still works. Second, a 640px image is enough for tone sampling, so each fetch is a fraction of the original file size.
Audit your own site in five minutes
Search for every place where code loads media, and confirm that each source is either proxied or local:
bash
grep -rn "loadImageEl\|new Image()\|TextureLoader\|useTexture\|<video\|unoptimized" src
For each match, check the following:
- Images and textures: Is the
srcwrapped in/_next/image, or is it a file inpublic/? A rawblob.vercel-storage.comURL is a leak. - Video: Does the source come from Blob? On Hobby, move it. Is there any fallback to a longer file? Remove it.
preload: Usepreload="none"or"metadata"for anything below the fold or behind a hover.- Meta tags: Do the OG and Twitter image URLs point at your domain, or straight at Blob?
- Environments: Do the dev server, previews and test runs read the production store? If they do, that usage counts against the same quota.
After the September fix, I scrolled / and /work at 1920×1080 with the network panel recording. The result was zero direct requests to blob.vercel-storage.com and zero <video> elements with a source. Before the fix, the same pass showed all 13 showreels referenced on /work. The target I am watching after the block lifts is Blob Data Transfer at or below about 0.1 GB a day, with no spike days. I will report whether that holds once there is a month of data.
The bigger lesson: never hotlink a metered origin
Both incidents share one cause. Every service that charges per read or per byte, including Blob, S3 and any other storage bucket, needs a cache you control between it and the browser. Video should not come from that service at all on a free plan. This also applies outside Vercel. The same pattern made my Neon database hit its limit on a low-traffic site, and it made an uncached Next.js fetch keep hitting the CMS. Low traffic does not prevent a large bill when every request goes to the origin.
One more note for anyone who runs a business site on Vercel. Vercel's fair-use terms restrict Hobby to non-commercial personal use. My portfolio sits on Hobby, so I accept the hard block as the cost. A client's revenue-earning site should be on Pro, with spend management configured, so that an overage costs money instead of taking images offline for a month.
FAQ
How long does a Vercel Blob block last on Hobby?
Thirty days from the day you went over, not until the next billing month. The store page banner shows the exact "Access resumes on" date. On Hobby, upgrading to Pro is the only way to lift it early.
Can I download my files while the store is blocked?
In my experience, no. Content downloads return 403. Users on the Vercel community forum report that list() and head() still respond, but the file bytes do not. Keep local copies of everything you upload.
Does creating a second Blob store reset the limit?
No. On Hobby the allowances are per account. When my store was blocked, a store in an unrelated project was blocked at the same time.
Does routing images through next/image just move the cost?
Partly, yes. Image Optimization has its own Hobby allowance of 5,000 transformations a month. A small fixed set of widths and a long minimumCacheTTL keep that number bounded, and repeat views come from cache instead of Blob.
Is Vercel Blob a bad choice for video?
It is a bad choice on Hobby, where 10 GB of transfer covers fewer than 700 full plays of a 15 MB reel. On Pro it is a cost question. For hover loops and background video, short files in public/ or an egress-free CDN are the safer default on any plan.
Need a site where the media pipeline does not break?
I build animated, media-heavy Next.js sites with WebGL, video and large image sets, and I plan their delivery as carefully as their motion. That means proxied images, bounded transformations, video kept off metered storage, and one dashboard you can read. Have a look at recent work, then tell me about your project.
04 — Studio note
For agencies, studios and brands
Need this levelof execution?
05 — Related notes
Same tags · same craft
Continue the thread.
- 01Reduce Vercel Edge Requests: 97% Used at 30 Views a DayStudio note / 094 · 18 Nov 26How to reduce Vercel Edge Requests when the free 1M is nearly gone: on this site 97% came from my own Link prefetch, not bots. Fix with before/after numbers.11 min
- 02Vercel Deploy Not Updating Content (Cache, Not Code)Engineering note / 081 · 11 Sep 26Vercel deploy not updating content? The build shipped fine — the cache outlived it. How to prove it in 30 seconds and the fix your developer owed you.12 min
- 03Website Not Loading in China? What It Actually IsEngineering note / 087 · 23 Jul 26A client's site "wouldn't load" in mainland China. It wasn't the Great Firewall. Here's how I actually diagnosed a website not loading in China — and the fix.11 min