Today we launched R2 Local Uploads to improve upload performance. Here's how it works 🧵

In tests, we saw up to 75% reduction in total request duration for cross-region uploads when Local Uploads is enabled for a bucket. Try it here:

When you upload to R2, your request hits the R2 Gateway first (runs on Workers, deployed globally). Then the object data travels to the storage infrastructure of your bucket.

If your user is in North America uploading to a bucket in Europe, that data has to cross the long distance from the client to the bucket. More distance = more latency, more variability.

Local Uploads changes the flow. Instead of writing directly to the bucket's region, R2 writes object data to storage near the user first. Then we replicate it to your bucket in the background. The object is readable immediately.

If the user and bucket are already in the same region, R2 skips all this and writes directly. No overhead when you don't need it. Learn more:

Want to know if Local Uploads is right for you? Go to your R2 bucket → Metrics → Request Distribution. See where your write requests are coming from.

Enable with one commnad: npx wrangler r2 bucket local-uploads enable <BUCKET> Now in open beta, no extra cost.

it looks nice as long as the bucket is not in your region as in my test, with local upload is slower

Should it be worse with it compared to without?

Happy R2 customer here :)

This is slick and practical, the UI cues plus immediate readability make latency feel like a feature, not a bug. If Local Uploads scales, teams will rethink geo data strategies and push for more region aware UX, lowering frustration for users across regions

@irvinebroque amazing

Is the globe built with @threejs ?

cloudflare dropping features faster than my s3 bills are dropping. wait no that part's still going up

this is an awesome ship. congrats

this looks so cool! inspired by @Jilles ?

cant stop loving cloudflare

Oh it's cool

Really like how this focuses on making the user ux better, less lag for the user the better

such a good visualisation tho!

It’s basically like caching files? Amazing!

Just tried it and WOW! I saw at least 3x faster upload speeds with large files (3-8GiB). Amazing work! Thank you!!! 🙏

Looks great! will test it in my next project. Captcha going out of bounds here irks me a bit tho lol

oh no, what are you seeing with the Captcha?

Oh no no!! I just meant the captcha is a bit wider than the other stuff in the sidebar, just a tiny UI nitpick lol

This is awesome, and the demo is really nice too 🔥

Any reason to not have it on by default in all r2 buckets?

today only we launched r2 integration into so people can build apps using ai, powered by r2 storage without much hassle remove supabase storage as our provider

@kathyyliao I’m here for the animation and love R2

that map is slick

This is fire!! 🔥🔥🔥 Any news on the italian fine? 😬

Wonderful 😍

Hello , there is a mistake in Morocco, it shouldn't be devided in the map! could you please fix it ! thank you

awesome! is there a way to enable it from the dashboard?

Yes!

thanks!

Why is latency between two regions so high? Conceptually you could take the large file and split into a bunch of tiny individual streams and it shouldn’t be magnitudes slower than local, especially given you are in control of the network / protocol between hops.

Why is this something new? It should have been done this way since forever. Good job though, better late than never. PS- Congrats on having a job that's hard for agents to replace!

@CloudflareDev Looks good Anni 👍

I'm saving this post ... will go through it later

Geo replication for R2 when?

Amazing job

that's amazing, btw will there be any chances of increasing the 50MB file size limit in the future?

The only cloud platform that lets me make money. Keep up the great work cloudflare. Cheers from

Local uploads is one of those features that sounds boring until you realize it eliminates an entire class of latency complaints. The best infrastructure improvements are the ones users never notice — they just stop complaining.