By now, many consumers know that uploading a photo, video, or other file to “the cloud” on a service like Google or Facebook often means storing it in a data center. But the computer servers inside data centers don’t just store information—they are constantly processing complex algorithms and using energy to glean new insights for the companies that store it.
A Seattle-based startup, Xnor.ai, is attempting to build a business on the belief that it could also be possible to run machine learning algorithms round-the-clock on everyday devices like smartphones and cameras. That may help businesses crunch more data and use less energy than they could by primarily running the algorithms on powerful computer servers housed in data centers.
On Wednesday, Xnor.ai unveiled a small, solar-powered computer chip that can run machine learning algorithms at power levels low enough that the chip doesn’t require a battery and can run off one solar cell. The chip, which is about the size of a quarter, is equipped with a camera that can detect and distinguish between visual objects—people, for example.
Xnor.ai operates under a business model that involves licensing its proprietary software to customers in exchange for monthly or annual fees, says Ali Farhadi, who co-founded the startup in 2016.
He says the solar-powered chip is not available for purchase, and instead the company is staging a “showcase” to get current and potential future customers of Xnor.ai thinking about ways to incorporate the technology into devices they sell.
Machine learning is part of the fast-growing artificial intelligence software sector. It involves programming algorithms to analyze large data sets and, over time, identify patterns—thus making the algorithm “smarter.”
By outfitting basic electronic devices with chips like the one Xnor.ai has developed and turning the them into artificial intelligence workhorses, more data could get crunched, making the chips’ underlying algorithms more sophisticated, Farhadi says
Another potential benefit of having smaller and cheaper computing devices perform more machine learning work is energy savings. Powering a data center tends to require lots of electricity because servers get hot from processing information all day. Xnor.ai says its chip can run state-of-the-art machine learning models using a cheap, off-the-shelf solar cell as a power source.
Xnor.ai says its chip can transmit images and data recorded by the built-in camera to other devices without an Internet connection. The information is instead beamed over “low data rate wireless communication protocols,” similar to how some recently introduced smart devices send and receive data, the company says.
Xnor.ai’s technology allows computer processors to analyze data in a way that consumes relatively low amounts of power. The startup developed a method for performing the types of calculations many machine learning models make, but which requires simpler inputs than those used for such algorithmic calculations previously, as detailed in this Xconomy report from 2017.
Ambarella (NASDAQ: [[ticker:AMBA]]), a device- and computer chip-maker based in Silicon Valley, is one of Xnor.ai’s customers.
Farhadi says Xnor.ai plans to “watch the market very carefully for feedback” on Wednesday’s announcement, and “see what the next move is” for the startup.
“We might be able to sell this as a chip, or sell the IP for it, or sell an end product,” he says. “We haven’t decided on that quite yet.”
If Xnor.ai does elect to mass-produce and sell the ships, they’ll likely sell for less than $1 each, says Sophie Lebrecht, the company’s senior vice president of strategy and operations.
Doing so would take Xnor.ai, which was incubated within the Allen Institute for Artificial Intelligence and has raised $14.6 million across two financing rounds, beyond the software-as-a-service realm it’s operated in up to this point.
“We built algorithms that are powerful enough to extend to whatever [devices are] out there,” Farhadi says. “This level of power efficiency and walking to the edge of possibilities … we have to offer to our potential future customers a combination of our own hardware and our own software together.”
