Cloudflare R2 - S3-Compatible Storage That Doesn't Bleed You Dry With Egress Fees

Here's the deal: AWS S3 has been bleeding developers dry with egress fees for years. You upload your files for cheap, then get hit with $0.09/GB every time someone downloads them. Your cute little side project that goes viral? Congrats, you're now bankrupt.

R2 fixes this shit. It's S3-compatible storage that doesn't charge egress fees. Ever. Your boto3 code works unchanged, your existing AWS CLI tools work fine, but suddenly your bills don't make you want to cry.

How R2 Actually Works

R2 has four main pieces: the gateway that handles auth and routing, a metadata layer that keeps everything in sync using Durable Objects, a cache tier for speed, and the actual storage backend. It's spread across Cloudflare's 330+ data centers, so your files are close to users everywhere.

Performance is solid - Cloudflare claims 20-40% faster than S3, which sounds like marketing bullshit, but I tested it with a few hundred MB of images and yeah, it's noticeably faster. The strong consistency model means you get immediate read-after-write consistency across all regions.

They finally added event notifications - triggers Workers or queues when your objects change. Actually useful for real-time processing instead of polling like a caveman. The new infrequent access storage class cuts costs even further for archival data.

R2's event notification system triggers Workers or queues when objects are created, updated, or deleted, enabling real-time processing workflows.

The Real Cost Difference

Here's where it gets expensive. S3 charges $0.09/GB for egress. R2 charges nothing. Zero. Zilch.

I migrated a client's media storage from S3 to R2 last year. They were spending $1,200/month on AWS egress fees alone. After moving to R2? $18/month total. The client thought I'd made a mistake on the invoice.

Storage costs are similar - R2 is $0.015/GB/month vs S3's $0.023/GB - but that's not where the savings happen. It's the egress fees that murder your budget. Use the R2 pricing calculator to see your potential savings. Most developers using object storage for media delivery save 70-90% on their total storage costs.

The free tier includes 10GB storage monthly, plus 1 million Class A operations and 10 million Class B operations - more generous than AWS S3's free tier.

What You Need to Know Before Migrating

R2 isn't perfect. The S3 compatibility is about 90% - some edge cases with boto3 operations fail silently. Cross-region replication doesn't exist yet. Advanced S3 features like Intelligent Tiering aren't available.

But if you're serving files to users (images, videos, downloads, backups), R2 will save you a fortune. Sure, no egress fees sound great until you realize you're stuck. Want to leave Cloudflare later? You'll pay those egress fees to move your data out. But honestly, after seeing what AWS charges, that future pain seems worth it.

The migration tools work well, though plan for 2-3x longer than estimated because of rate limiting. Super Slurper handles one-time migrations up to 1TB per object (finally), while Sippy provides ongoing incremental sync from S3.

Canva uses R2 for their image storage - makes sense when you're serving millions of user uploads. The S3 compatibility means you can drop it in without rewriting your entire app. MosaicML uses R2 for AI model training data, and Prisma saved 98% on distribution costs. Real companies are using this in production.

Setting Up R2 (The Actually Important Stuff)

Getting started is pretty straightforward if you already have a Cloudflare account. You can use their web dashboard or the Wrangler CLI - I prefer CLI for anything serious.

R2 buckets can be created in specific regions (North America East, Europe West, Asia Pacific, and the new Oceania region) for compliance requirements. Public buckets let you serve files directly to browsers, private buckets need signed URLs or API token auth.

Pro tip: Start with private buckets. Public buckets are convenient but you lose control over access patterns. Use CORS configuration for browser uploads and custom domains for better SEO.

Where R2 Actually Shines

AI/ML Workloads: No egress fees means you can train models across different GPU providers without getting murdered on data transfer costs. I've seen ML teams save $10K+/month by storing datasets in R2 instead of S3. MosaicML uses R2 for LLM training data with zero vendor lock-in.

Media Delivery: The CDN integration is seamless. Store images/videos in R2, serve them through Cloudflare's edge cache. Performance is solid and costs are predictable. Use Transform Rules for dynamic image optimization and Cache Reserve for persistent caching.

Backup Storage: R2's migration tools (Super Slurper and Sippy) work well for one-time or incremental backups. Super Slurper now supports objects up to 1TB and includes Google Cloud Storage migration. Just don't expect them to be fast - I've seen 50GB transfers take 6 hours instead of the estimated 2. Plan accordingly.

Data Analytics: The R2 Data Catalog (still beta) adds Apache Iceberg support for running analytics directly on stored data. Works with Snowflake, Apache Spark, and DuckDB. Useful if you're already in the Cloudflare ecosystem.

Log Storage: Cloudflare Analytics & Logs can automatically send your logs to R2 for long-term storage and analysis. No more expensive log aggregation services eating your budget.

R2 enables true multi-cloud architecture by eliminating egress lock-in - move data freely between providers without paying transfer fees.

Real Companies Using R2

Canva uses R2 for their image storage - makes sense when you're serving millions of user uploads. The S3 compatibility means you can drop it in without rewriting your entire app.

Indeed moved their job posting assets to R2. Japan Airlines uses it for their digital transformation platform.

The [Workers Integration](https://developers.cloudflare.com/r2/api/workers/workers-api-usage/) (This Is Where It Gets Cool)

You can process files directly at the edge using Workers. Image resizing, format conversion, custom auth - all without spinning up separate compute instances. The R2/Workers combo is actually pretty slick - way better than spinning up separate Lambda functions for every damn image resize.

AutoRAG is their new AI thing that crawls your site and dumps content into R2 for search. Sounds gimmicky but the R2 integration is solid. Event notifications can trigger Workers when objects change, enabling real-time processing pipelines. Use multipart uploads from Workers for handling large file uploads efficiently.

Migration Reality Check

Moving from S3 to R2 isn't just changing your endpoint URL. Here's what actually happens:

• Simple apps: boto3 code mostly works unchanged
• Complex setups: Some operations fail silently, test everything
• Data migration: Use Super Slurper for one-time moves, Sippy for ongoing sync
• Timeline: Plan 2-3x longer than estimated due to rate limits

Most migrations I've done take 1-2 weeks for testing plus however long the data transfer takes. Had one migration where we thought we were done testing, deployed to prod, and then discovered that lifecycle policies weren't working the same way. Spent a weekend rolling back. Test your edge cases.