WavebreakmediaMicro - Fotolia

Azure IoT Edge tool set stirs AI into Microsoft's cloud

Microsoft extended more of its AI capabilities to connected devices to resolve some of the IoT shortcomings of the public cloud's consolidated data center model.

SEATTLE -- AI and connected devices have coalesced as a central component of Microsoft's Azure cloud strategy, as the company seeks to marry the disparate technologies to complement and amplify each other.

Microsoft plans to spend $5 billion over the next four years on IoT, and AI will play a big part in that push, judging by the myriad overlaps in new functionalities rolled out here at Microsoft's Build developer conference. The updates push more of Azure's capabilities to edge devices, as Microsoft attempts to weave a unified fabric of services that extend out from its public cloud.

The open-sourced Azure IoT Edge runtime improves users' ability to modify and debug IoT applications. Microsoft's edge computing model is similar to that of AWS Greengrass, which Amazon has promoted as a way to add compute capabilities outside AWS facilities.

Public cloud providers have convinced corporations to shrink their own data centers and move workloads to cloud facilities, often hundreds or thousands of miles from their business. For applications that link to connected devices, that can create problems with latency and transfer fees, as well as the potential to lose access to the device due to poor connectivity. To address the public cloud's shortcomings in this area, both Microsoft and AWS have pushed to provide more capabilities for edge devices.

Custom Vision, an image recognition API, can now run on Azure IoT Edge, which opens the door to devices with cameras, such as drones, to analyze data locally, rather than send it back to the public cloud. Microsoft said it will incorporate other cognitive services into Azure IoT Edge in the coming months.

Azure IoT Edge now supports Event Grid, as well as Azure Kubernetes Service, which enables customers to deploy virtual kubelets on a given device and spread a cluster from the cloud to edge devices.

There's also Project Brainwave, an architecture for neural-net processing that offers five times lower hardware latency than Google's Tensor Processing Unit for real-time AI, according to Microsoft execs. And Project Kinect for Azure is a package of sensors, including a new camera, with compute designed for AI on remote devices. Microsoft has worked to integrate Azure's IoT capabilities with Azure Machine Learning for predictive analytics running on devices.

AI capabilities permeated or touched on just about all the new tools. Azure Search can now integrate with Cognitive Services to better index images, videos and PDFs, and Visual Studio IntelliCode can make suggestions for code improvements for better productivity.

Fewer hurdles for AI in the edge, cloud

The ultimate goal is to extend AI tools beyond Azure, so connected devices are more intuitive and responsive, allowing companies to better track the efficacy of critical equipment in factories and in the field.

You can have IoT without AI, and you can have AI without IoT, but there is complementarity of the two together -- particularly if you're talking about [the] edge.
Ezra Gottheilanalyst, Technology Business Research

"You can have IoT without AI, and you can have AI without IoT, but there is complementarity of the two together -- particularly if you're talking about edge and AI," said Ezra Gottheil, an analyst with Technology Business Research in Hampton, N.H.

Microsoft wants to lower the barrier for AI, which traditionally required a big investment in data scientists and data lakes, Gottheil said. By their nature, edge and IoT devices have limited complexity, so they present the opportunity for more reasonable first steps into AI for most corporations.

It makes sense to be able to bring AI models down to the edge, and continuous updates to those models make sense for real-time data processing, said Jeffrey Hammond, a Forrester Research analyst.

"As our clients try to do more complex things, the need for intelligence at the edge grows, either in real-time processing capabilities or just to keep the data volumes down," Hammond said.

Microsoft has also expanded its partner network for connected devices, which includes a software development kit by drone vendor DJI for Windows 10 PCs and a vision API for camera-based applications built jointly with Qualcomm Technologies.

Few companies will have full-stack deployments from a single vendor, so partner ecosystems -- from chip fabricators up to device manufacturers -- is critical to the success of a given IoT platform, said Stacy Crook, an IDC analyst.

And Microsoft isn't the only one vying for a greater mind share in IoT. Other cloud providers and manufacturing companies have staked their claim in this emerging market.

"The rate of change and addition of new capabilities is very high, which itself is a reflection of how it's still early days," Gottheil said. "Everyone is bringing on new things as fast as they can."

Dig Deeper on Cloud deployment and architecture

Data Center