Google is staking its claim in the next big thing after cloud computing with a new line of AI-powered hardware for developers
- Google introduced Coral, a line of hardware to help hackers build and experiment with AI-powered gadgetry.
- It's similar in principle to popular minicomputers like the Raspberry Pi, but with some Google special sauce - it uses a custom Google processor, customized for AI, and is designed to run the Google-created TensorFlow AI software.
- This could help Google spread the word of the already-popular TensorFlow, while also staking its claim in edge computing.
- Edge computing refers to the concept of putting more intelligence on a device, rather than in the cloud. Indeed, some believe that edge computing could be a larger market than cloud computing.
Google has quietly launched Google Coral, a line of relatively cheap hardware aimed at helping developers experiment with building gadgetry powered by artificial intelligence.
On its website, Google Coral has product listings for a $150 motherboard, a $75 USB device to bring AI to existing systems, and a $25 camera that slots into the board. The listings were first spotted by the Verge.
"Coral offers a complete local AI toolkit that makes it easy to grow your ideas from prototype to production," writes Google in a blog post announcing Coral.
In theory, it's more than a little bit like the Raspberry Pi, the pioneering $35 minicomputer, which is mega-popular among hackers as an easy and cheap way to build experimental hardware and other oddities.
In practice, the Coral lineup appears to come with lots of Google special sauce.
The processor on the Google Coral developer board is an Edge TPU, a chip specifically designed by the search giant to bring AI to low-powered devices like cameras and home appliances. It's also designed to run TensorFlow Lite, a version of Google's very popular open source AI framework designed, again, for low-powered devices.
It's important to note that these devices aren't actually much good at training AI algorithms - as the Verge notes, you'll need much more powerful hardware for that. Rather, these are good for putting those algorithms to work, and helping gather the real-world data to refine them.
And this may be the real significance of Google Coral, as the company looks to stake its claim in so-called edge computing, the market that many industry insiders believe could be bigger than the cloud.
The big idea behind edge computing is to bring more intelligence to devices like phones, TVs, appliances, factory robots, and even self-driving cars and other vehicles. While the cloud brings unprecedented levels of supercomputing power to anything with an internet connection, there's a serious latency problem; you don't want your self-driving car waiting to get a response from the server while it figures out whether to stop at a traffic light.
The solution, then, is to give the device (or car, or robot) enough computing power to make decisions on its own. The massive processing power of the cloud can help formulate, analyze, improve, and generally fine-tune the algorithm, while the device itself has enough AI to run the algorithm quickly and accurately.
Hack away and spread the gospel of TensorFlow
Cloud players like Microsoft Azure and Amazon Web Services both already have their own plays for edge computing, while legacy companies like Intel and Hewlett Packard Enterprise see the opportunity to gain ground after largely losing out in cloud computing. Indeed, Intel offers its own cheap AI hardware to developers.
For Google's part, TensorFlow and its Lite variant - open source projects that are free to use - have basically become the standard software for powering artificial intelligence, with the Facebook-created PyTorch as its primary competition. In mid-2018, Microsoft even bought a startup powered by the Google-created TensorFlow.
On Wednesday, Google also announced that TensorFlow had been downloaded 41 million times as of November, and that TensorFlow Lite is running on 2 billion phones and other mobile devices. Google itself uses TensorFlow Lite to run the Google Assistant, Google Photos, and even Google Search on phones.
Which is a very long way to come back around to Google Coral. By reaching out to developers with tools that make it relatively cheap and easy to hack away at new hardware projects, it could very well spread the gospel of TensorFlow and the Edge TPU.
That's good for Google in the long haul, because while TensorFlow might be free software, Google Cloud offers developers plenty of services for powering these devices on the backend. Indeed, Google says in its blog entry that Coral is made to integrate nicely with Google Cloud's internet of things (IoT) backend services.
That, in turn, only stands to boost Google Cloud's reputation as the best place to run TensorFlow apps, which could help it build its credibility in both AI and edge computing - a plus as it pushes against the leading Amazon Web Services and second-place Microsoft Azure clouds.
It's a playbook that's worked for Google before: Kubernetes, a very popular open source tool for managing large-scale cloud infrastructures, became a cloud standard because developers love it so much. If developers come to love Google Coral, too, it could make Google Cloud a more attractive place for developers in the next big thing.
Meanwhile, Google's rivals are doing their own kinds of outreach to AI developers. Microsoft recently resurrected the Xbox's failed Kinect accessory as a $400 AI-powered camera for developers, while Amazon is letting developers program their own self-driving toy cars.