Why I Left AWS After 6 Years to Build Sudden
Six years at AWS showed me where CDN optimization ends. Etienne had already built the technical seed for a different answer.
Wednesday evening. I’ll be tweaking our internal agentic tooling instead of watching the stream for Knicks - Sixers Game 2.
Few months into building this new venture. Wouldn’t trade it.
But I thought I’d be a medical doctor, by now.
Boring government job, summer time 15 years ago. I asked my childhood friend Etienne (ironically another Etienne, not my current co-founder) to lend me any book he had, anything - I had time. He happened to pass me one on entrepreneurship. I read it and it clicked. Build something from scratch.
This led me to spend the following 9 years building in startups. Mostly product, mostly zero-to-one, some AI/ML, a bunch of different industries. One of them was SweetIQ, a Montreal company that ended up getting acquired by USA Today Network, a media company. That part comes back later.
In 2020, AWS recruited me to lead a new AI/ML Solutions team. Stayed 6 years. Went deep across the cloud stack: AI/ML, compute, networking, edge. Most of that time I spent across the table from cloud customers trying to make their economics work.
That’s where the problem became obvious.
The same conversation kept coming up around a particular CDN / edge product in my portfolio. Different VPs of Engineering, different platforms, same conclusion every time:
“We’ve done everything we can to optimize our costs.”
Multi-CDN. Aggressive renegotiations. Every cache trick in the playbook. File size compression. Every couple of years, a new vendor would come through with a promise. Most went nowhere.
Doesn’t matter how good your video infra team is. Doesn’t matter how hard you negotiate. Growth and cost move in lockstep.
After enough of those conversations, I couldn’t keep building on top of the problem. I had to try to fix it differently.
Etienne Martin (opens in a new tab) came at the same problem from the other side.
He fell in love with computers young, drew on the family computer, made little animated movies. Got hired by a computer shop in his hometown to do repairs, and one day the owner told him to build the shop a new website. That’s how he got into web development. No Stack Overflow back then - change something, see what breaks, keep going.
He moved to Montreal eventually and joined SweetIQ on a software development contract. That’s where we met. We worked together for a couple of years before USA Today Network acquired the company. The parent company kept telling us how hard it was for smaller media properties to stay profitable. First time either of us heard the media-economics problem from inside the room.
After SweetIQ, Etienne went on to lead architecture for one of Canada’s biggest media companies. Content delivery - radio, news, TV, movies. He spent the pandemic in the trenches of delivering video online at scale.
He saw what I couldn’t see from the AWS side. Most of the traffic he was watching was duplicate bytes - the same segments delivered over and over to different viewers.
In 2021, Etienne had built a prototype for delivering images and web assets in a decentralized way. Multiple devices sharing chunks with each other instead of each pulling from a central system. Not a startup. Just a prototype that lived on his laptop.
When I told him I was thinking about my future, he said: “I might have something we can build on.”
Quick aside on the worldview. One of my favorite books is The End of Power by Moisés Naím. The argument is that centralized institutions are slowly losing leverage to distributed actors. So when Etienne started talking about a decentralized delivery prototype, it landed in a frame I already held. I just didn’t have the technical seed.
Most of a streaming platform’s CDN bill isn’t bandwidth delivery. It’s re-delivery.
To make it dumb simple: when 10,000 viewers watch the same segment, same bitrate at the same time, the CDN delivers that segment 10,000 times. Identical bytes, all counted and billed separately. The architecture is doing exactly what it was built to do: point-to-edge, fast and reliable. But it bills you for bytes, not unique segments.
Two academic studies, IEEE NetSession and ACM RICHTER, put the recoverable share of concurrent VOD traffic at 70-88% via peer-assisted delivery. Not a forecast. Measured.
That’s what Etienne’s prototype was already doing, same bytes delivered once, then propagated peer-to-peer. We just had to apply it to streaming-scale video instead of static web assets.
So in January 2026, we started Sudden.
A peer-assisted delivery layer that drops in alongside any CDN. No infrastructure changes. Concurrent viewers’ devices share segments over WebRTC. The CDN delivers each segment once per algorithmically selected cluster of viewers, instead of once per viewer. CDN remains as the trusted fallback.
For platforms with concurrent VOD, that’s a possible 60-80% of egress that never has to hit CDN at all.
The pricing matches the bet: we take 30% of whatever your CDN bill drops by. If it doesn’t drop, you owe nothing.
The way to actually find out your potential savings: take a look through our technical documentation (opens in a new tab). Found a fit? Contact us and we’ll run your numbers through our savings model. Happy with what you see? We support the integration within days, not weeks.
Over time, we share our monthly savings reports showing, among other things, the offload percentage. Your CFO will be able to do the math against your actual rate.
Why Etienne and I think we can pull this off:
Etienne hit the CDN redelivery problem head-on at one of Canada’s biggest media companies during the pandemic. He’d already prototyped the answer before either of us thought of it as a company.
I sat across from streaming platforms for six years watching them hit the same wall over and over. I had the customer pain catalog and the economics. I just needed the technical seed. Etienne had built it four years earlier.
Want to run the numbers on your own traffic?
Talk to our founders about your audience, delivery stack, and where peer-assisted delivery could help.
