ProductsSelected

Hugging Face builds its own CDN for massive AI downloads

Why we picked it

The non-obvious insight is that AI distribution inverts CDN economics: machine clients in a few cloud regions replace humans in hundreds of cities, so edge caching stops paying off. Teams serving model weights, checkpoints or datasets should re-examine whether commercial CDN contracts match their traffic shape, especially around release-day spikes. Startups building model hubs or weight-hosting services will find the capacity numbers and design rationale directly relevant to their own cost planning.

Hugging Face has been running its own content delivery network since this summer, now shipping almost 1 exabyte per month — roughly 420 GB every second — through 80,000 requests per second. The company moved away from commercial CDNs because AI workloads break their core design assumptions.

Image · Thomas Wolf
Body

Traditional CDNs were built to distribute many small web files to many human users across hundreds of edge cities. Hugging Face's traffic is the opposite: a few enormous files, a large share pulled by machines rather than people. The audience is largely GPU clusters concentrated in a handful of cloud regions, and each client wants an entire model immediately rather than a cached page. The custom CDN uses caches adapted to massive objects and can absorb release-day mega-floods when several giant datasets and trending models are pulled at once.

The shift shows how AI infrastructure is forcing a rethinking of web delivery fundamentals: caching economics flip when a 1,000 GB model serves far fewer clients, each demanding full throughput. Hugging Face also published a live traffic map of the network at huggingface-cdn-live.static.hf.space. For anyone serving large model weights or datasets, the post is a data point on why off-the-shelf CDN pricing and architecture may not fit machine-to-machine traffic.

Original · Thomas Wolf

AI downloads broke the usual CDN model Since this summer, we're running our own CDN at Hugging Face. It's shipping almost 1 exabyte per month, which means, in practice, about 420 GB every second, through 80,000 requests per second. Why did we need to build our own? Well, CDNs were designed to ship many small web files to many people. AI downloads are pretty different: it's a few enormous files, a significant part of them pulled by machines. This changes a lot of design assumptions. You're not caching small objects close to users in hundreds of cities anymore: these users are for a big part GPU clusters sitting in a handful of cloud regions. And a cached web page is useful to 1,000 visitors, while a cached 1,000 GB model is useful to far fewer clients, but each one wants all of it, immediately. So we've built our own CDN, with caches adapted to massive objects, able to handle release-day mega-floods and multiple giant datasets and trending models being pulled at once. The team made a live map of it at huggingface-cdn-live.static.hf.space/index.html Hard to stop watching it - it's quite mesmerizing

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Hugging Face, CDN, open-source models, AI infrastructure.

Source attribution: curated from Thomas Wolf (@Thom_Wolf).