When millions of people press play at the same moment, for a World Cup final or an Olympic race, say, it’s not uncommon for a stream to stall. For years, the industry’s answer to this conundrum has been to lean on traditional content delivery networks (CDNs) and deliver more bandwidth. But when you really think it through, the fundamental problem with the internet today isn’t capacity, it’s the routing protocol. It’s called the Border Gateway Protocol, or BGP, and it really is the cause of internet congestion. Imagine if your GPS sent every driver in the city down the same highway because it was the shortest route. Would anyone be surprised by the traffic jam?

At one level, BGP makes sense. It sends data along the route with the fewest “hops”, based on the logic that every router a packet passes through adds to congestion. The fallacy is that when every router follows the same rule, effectively all the traffic ends up on the same “shortest” route. So, the rule designed to prevent congestion actually adds to it.

I saw this during the 2016 Olympics in Rio de Janeiro. During the major events, Europe was effectively blacked out on the internet. Almost all of the data headed from Brazil to Europe was routed through the island of St. Croix, and it completely congested the routers there. Keep in mind that Brazil, at the time, relied on U.S. undersea cables for almost all of its communications with Europe. The first direct link between the two only opened in 2021.

I realized fairly early on that this was a transportation problem. Moving data packets is no different from moving any other goods, except that it is tremendously more complex. When Amazon delivers a package, the roads don’t move around. The internet looks more like the Milky Way, with routers as stars and links running between each of them. Everybody assumed that applying transportation optimization at this scale would be NP-complete, in other words, mathematically unsolvable.

The breakthrough came in two parts. First, the industry forecast net capacity, i.e. capacity less demand, and from a statistical perspective, a net difference is one of the worst things you can try to forecast. Second, capacity is somewhat fixed. These are physical devices, and we can measure exactly how much capacity they have. That tremendously simplifies the calculation, because all you are then forecasting is demand. And demand is fairly easy to forecast. Even a rough estimate works pretty well.

That is where AI comes in (and only there). AI is built for probabilistic problems, whereas transportation optimization is a deterministic one. In other words, there is one right answer. So we use AI to forecast where demand will be, and transportation optimization to calculate the best route for each individual data packet.

The result is a tremendous increase in what the internet can handle. If you think about cable, it has a specifically allocated number of strands. It can handle very high resolution, but each step up in resolution uses more strands and leaves room for fewer channels. The internet, on the other hand, can carry a virtually infinite number of streams, but under BGP, often at very low resolution. Better routing removes that trade-off: unlimited streams at the highest resolution, all the way through 4K and 8K.

And quality is what keeps subscribers. How many people have canceled a streaming service because of how the picture looked, rather than what was on it? At System73, the company I founded to solve this problem, we saw the difference during the Copa América final: 95% of viewers received the highest-quality stream their device could handle, and the broadcaster cut its operating costs by half.

Nobody would put up with a GPS that sends the whole city down one highway. We already know how much capacity the internet has. So why do we keep sending everyone down the same road?