It isn’t a slice of online life — it’s the majority of it. Industry measurements (Sandvine, 2023) put video at roughly 65% of all internet traffic, and even higher on mobile networks.
Authentic Media does, with a tiny team, what used to take a roomful of infrastructure engineers: take a video off your phone, caption and encode it right in your own browser, package it for adaptive streaming, and deliver it anywhere in the world — fast. The heavy media work happens on your device; Fastly’s edge stores it and serves it. Here’s why that’s genuinely hard, and how the pieces fit together — everything below is the public story, no secret sauce.
More data, more compute, less patience. Almost every hard problem in computing shows up at once when you try to move video to millions of people.
It isn’t a slice of online life — it’s the majority of it. Industry measurements (Sandvine, 2023) put video at roughly 65% of all internet traffic, and even higher on mobile networks.
One minute of uncompressed 1080p at 60fps is about 6 gigabytes — a flood of ~3 gigabits every second, more than most home connections could carry for a single viewer. Compression isn’t optional; it’s the only reason streaming exists.
Codecs (H.264, HEVC, AV1) squeeze that firehose down to a few megabits — but the better the compression, the more compute it takes to create. Newer codecs like AV1 can take 20–100× longer to encode than older H.264.
An uploaded file usually can’t be streamed as-is. It’s re-encoded into many versions for different devices, screens, and connection speeds — one of the most compute-heavy jobs in consumer software. We do it right in the creator’s browser, so there’s no encoding farm to run.
To play smoothly on a phone on spotty cell service and on a 4K TV on fiber, each video is cut into thousands of tiny segments at many resolutions at once — the “bitrate ladder.” The player quietly switches up and down it, second by second, to match your connection.
People have little patience: a study of 23 million views (Krishnan & Sitaraman, 2012) found viewers start leaving once startup passes ~2 seconds, with each extra second driving roughly 6% more of them away.
A video is encoded once but delivered to millions at the same moment, each pulling several megabits a second. Serving that from one place would melt the network — so it must fan out across a global CDN, in real time.
With recorded video you have time to encode carefully; with live you have none — capture, compress, segment, and deliver as it happens, with no chance to redo a bad frame.
Captions are generated by speech-to-text, timed to the frame, and packaged alongside every version; here, a source on every claim is its own step on top of all the encoding.
One video sets off a chain of expensive, specialized steps — most of them run right in your browser, on the device in your hand; the edge, close to you, stores the result and serves it to the world. No distant datacenter crunching your footage.
From your phone or browser, your video is recorded or picked right where you are — the file never has to leave your device to be processed.
Speech-to-text runs in your own browser, transcribing the audio and timing captions to the frame — so every clip is captioned and watchable with the sound off, and nothing is shipped off to be transcribed.
Your browser re-encodes the video into a clean, web-ready rendition, plus a poster and a download — the heavy, expensive step, done with WebAssembly on the hardware you already have. No GPU farm, no central server.
Fastly’s edge — running our own code as WebAssembly, milliseconds from you — takes the finished files, stores them, and serves them. It stores and delivers; it doesn’t crunch your video.
The renditions are cut into small HLS segments and a playlist — the “ladder” your player climbs up and down to match your connection.
The finished video is cached across Fastly’s global network and served from a point of presence near you — close, fast, and able to absorb a crowd.
Your player auto-picks the right rung of the ladder and adapts as your bandwidth changes — no settings, no dials, no stats cluttering the video.
None of this appeared overnight. On the left, how the internet and the cloud were built. On the right, how Fastly made the edge programmable — the piece that makes the pipeline above possible for a small team.
From four wired-together computers to renting compute by the hour.
U.S.-funded researchers wired four university computers together and pioneered “packet switching” — chopping data into small pieces that travel independently. It became the basic idea behind the whole internet.
Vint Cerf and Bob Kahn published TCP/IP, a shared “language” that lets completely different networks talk to one another — making one big internetwork possible.
On Jan 1, 1983 every machine on ARPANET switched to TCP/IP on the same day — the moment many treat as the true birth of the modern internet.
Paul Mockapetris invented the Domain Name System — the internet’s phonebook — so you could type “example.com” instead of memorizing numbers.
At CERN, Tim Berners-Lee proposed the Web — linking documents with clickable hyperlinks — put the first website online, and gave the software away for free.
Mosaic showed pictures alongside text and made the web point-and-click, pulling millions of ordinary people online and kicking off the web boom.
When one popular site could be flooded by a sudden crowd and crash, Akamai invented the content delivery network — copying files onto servers around the world so you load them from somewhere nearby.
AWS began renting storage and compute by the hour, so companies could stop buying servers and rent exactly what they needed, when they needed it.
Netflix started streaming and the iPhone put the full internet in your pocket — together they sent traffic, and especially video, soaring.
Browser makers shipped WebAssembly: a way to run programs at near-native speed inside a tightly sealed sandbox — soon a key building block for running code safely almost anywhere.
Platforms began using WebAssembly to run your code not in one distant datacenter but at hundreds of locations near users — “edge computing” — so apps respond in a blink worldwide.
A CDN you could program — then a place to run real code, microseconds from the viewer.
Artur Bergman (ex-CTO of Wikia) starts Fastly to build a faster, programmable kind of CDN — a network you could configure yourself, in real time.
Built on the open-source Varnish cache and its config language, Fastly could push custom logic to the edge in seconds and wipe stale content network-wide in about 150 milliseconds — cache aggressively, yet stay live.
Customers got live streaming logs and real-time visibility into the edge, instead of the delayed reports legacy CDNs offered.
Fastly went public, and unveiled Compute@Edge — running your own code at the edge with WebAssembly, claiming a ~35-microsecond startup (Fastly’s figure), effectively no “cold start.”
The WebAssembly edge-compute runtime reached general availability, and Fastly added a web application firewall and attack protection alongside delivery.
A brief outage publicly revealed how much of the web routes through Fastly — momentarily affecting sites and services people use every day.
Today it’s one edge cloud: content delivery, edge compute, and security — running customer code milliseconds from users on every continent.
of edge network capacity (Fastly’s public figures) — enough headroom to absorb huge spikes and attacks without slowing down.
“points of presence” on six continents, placed close to internet users almost everywhere in the world.
to start your WebAssembly code (Fastly’s claim) — millionths of a second, so there’s no “cold-start” lag for the viewer.
to purge cached content across the entire global network — the speed of caching with the freshness of live data.
Fastly’s philosophy: a smaller number of very powerful locations, so each caches far more and serves faster on average.
Until recently, building this whole pipeline — re-encode every upload, auto-generate captions, package an adaptive ladder, and deliver it worldwide — meant standing up racks of servers in many regions and stitching together a CDN: realistically, only big companies with dedicated infrastructure teams could do it. Two things changed. WebAssembly got fast enough to re-encode video and run speech-to-text right inside the creator’s browser, so the heaviest work no longer needs a server farm at all. And tiny programs compiled to WebAssembly now run within microseconds at hundreds of locations around the globe, right next to the viewer, on a network with hundreds of terabits per second of capacity — a secure, sandboxed runtime fast and cheap enough to run per-request, sitting on a planet-scale network. Together that’s what lets one person caption, encode, and globally deliver video that used to need a small army and a roomful of hardware.
There’s a “Report a problem” button on every page. When you use it, your note — and, if you choose, a screenshot you take — is sent to a separate reports service we use to track issues. We read every report, reproduce it, fix it on the live edge, add a test so it can’t come back, and mark the report resolved. No ad trackers, no behavioral profiling, no following you around — a report is something you deliberately send, not something collected in the background. If you leave an email we can reply; if you don’t, the report is anonymous.
Sources for the figures above: internet-traffic share — Sandvine Global Internet Phenomena (2023); startup-delay abandonment — Krishnan & Sitaraman, ACM IMC (2012); codec encode cost — public codec comparisons (AV1 vs. H.264); Fastly network, instant purge, and Compute@Edge figures — Fastly’s own public materials and SEC filings. Scale numbers are Fastly’s public claims and change over time. Raw-video and compression figures are illustrative, derived from typical streaming bitrates.