Our First Investment In Edge AI — Deeplite
My number one investing thesis for 2021 is around AI at the edge. As machine learning proliferates, there are so many small devices that can benefit from ML functionality. I expect AI to be in cameras, wearables, tvs, remote controls, vehicles, phones, headsets, computer peripherals, toys, games, clothing, glasses, appliances, and the list goes on and on.
The problem is — machine learning models are large and getting larger. More data, bigger models, more compute. That is the trend. But not all of these edge tools can be constantly cloud connected. Sometimes they can’t be for latency reasons. A few dozen extra milliseconds for a vehicle traveling 65 mph might be the difference between stopping safely and crashing. Sometimes they can’t be connected for privacy reasons. A service that authenticates you locally for access to sensitive data would probably rather keep that data local, on the device.
Most commonly though, it just doesn’t make economic sense to stay cloud connected. A $79 video camera that has automatic face detection may cost $5 — $25/year to stay cloud connected. If you are deploying hundreds of them, that adds up. Why not add a $1 microcontroller and put the ML model on that and run it locally to avoid the cloud charges? Well, because in most cases the microcontroller can’t run the ML model. The model is too large.
Today I’m excited to announce an investment we’ve made that is fixing that problem — Deeplite. The core technology behind Deeplite is an AI model that can optimize neural networks based on parameters you choose. Deeplite can make sure you hit a size or performance target, with minimum loss of accuracy. The technology is impressive, and the company has a great customer set and a long pipeline. And this is a market that is only going to grow as more ML models move to the edge.
I’m a bit embarrassed to admit that we passed on Deeplite the first time because yours truly thought the market was too small. Only after digging in a second time did I realize that the edge AI market is going to be much much much bigger than most investors currently realize. Luckily when Deeplite came around for this seed round, the timing was perfect and this time we understood the market much better.
If you are building a device that might benefit from AI at the edge, please give Deeplite a look. And if you are an early stage company building for the edge, please reach out if you are considering a venture round.