There are varying views on timing, but no one in the technology field doubts that the coming wave of IoT (internet of things) products will bring about a big shift in how we interact with tech.
IoT products will provide consumers with new ways to access and process information, while also supplying a significant addressable market for the hardware and software companies that get it right.
is not known as a company poised to take on IoT at the edge, where consumer devices will exist. Most have assumed that chip companies more ingrained in the cell phone and embedded markets would dominate there, including Qualcomm
and even NXP Semiconductors
Intel has generally not been considered one of the leading innovators in IoT.
The cloud infrastructure powering the IoT back end was the easy outlet for Intel to compete in. It already has more than 90% market share for enterprise data centers, and even with competition from AMD
and Arm-based products that continue to make noise, Intel will hold serve here. But to find additional revenue and to expand into new growth areas, Intel has its eye on the IoT markets at the edge, especially those that need more intense compute with machine-learning and artificial-intelligence (AI) integration.
Many tech companies will be going after this market as deep-learning revenue (deep learning is the ability to use computing algorithms and data sets to enable devices to learn and process independently) is expected to hit nearly $40 billion by 2025. As much as $11 billion of that number will come from “edge to cloud” spending, where devices handle most of the processing locally before migrating the resulting data to centralized data centers.
Intel is pushing and innovating on many technology fronts, trying to play catchup in areas where it has fallen behind or in new segments it is expanding into. Machine learning and AI are the most obvious examples with numerous internal projects targeting different segments of this space. The company also continues to develop new storage technologies, improvements in consumer processors, and data center capability. There are a lot of projects being balanced within its R&D budget.
One of the most immediate opportunities is with vision processing. Cameras are collecting 1.6 exabytes of data every day but only about 10% of that is viewed by a person or analyzed by a computer. This technology is used by commercial customers to help find product defects during manufacturing, to track inventory and reduce theft at retail, and for tracking vehicle traffic and adjusting roadway systems to decrease congestion. Future opportunities for AI-based vision processing are significant.
Much of the intelligent processing for these takes place on the edge, reducing the need for massive network connectivity and lowering the impact on network infrastructure. These vision products need to be more than just a camera; they need to be smart, and able to run machine learning and AI algorithms locally to find patterns and anomalies for each task.
Qualcomm and Microsoft
partnered to announce a platform for this market to accelerate adoption for security and home-camera installations. The Qualcomm Vision Intelligence Platform uses Qualcomm chips at the edge and Azure IoT systems to offer an easy-to-build product.
Portfolio of products
But Intel believes it provides the best portfolio of products for businesses to build on. That includes a set of three hardware products and a software tool kit to accelerate deployment. Integrated-graphics processors on Intel Core chips that are widely available in the workplace could be used for basic jobs of vision processing, and utilizing hardware already available provides a tremendous cost advantage. With the company’s push into programmable chips (FPGAs), Intel has a solution that is ready for high-performance needs where speed is critical for safety or production. And with the acquisition of Movidius, a technology that provides lower power and highly efficient machine learning compute capability, Intel has an option to provide for individual camera integrations.
A newly announced software tool kit called OpenVINO makes developing applications for all three of these processor types incredibly easy with a “code once, run anywhere” mentality. Intel has a significant history in creating developer tools that make it easy to widely deploy across a range of hardware, and it wants to prove to the machine-learning space that it will be able to do the same for this market. The platform that sees the most support from software developers early in the IoT landscape will have the best chance of capturing revenue in the market.
While the market is growing, and the players are jockeying for position, Intel believes it has a performance and efficiency advantage over what most consider the biggest name in machine learning and AI: Nvidia
Based on Intel’s testing, it can beat a discrete Nvidia graphics chip for data centers in these workloads in performance efficiency through its FPGA solution and run more efficiently with the low-power Movidius chip against an Nvidia Jetson module. The needs of each customers’ implementation would determine which Intel solution to integrate. As always, third-party testing will be needed to validate claims like this, but Intel is painting a positive picture.
Intel isn’t the only company looking to power this subset of machine learning computing. Nvidia has a “smart cities” initiative that includes video analytics and calls out specific use cases of retail, traffic and security. There is overlap of some partnerships, including Hikvision and Dahua, working with both Intel and Nvidia products, so I expect to see an increase in competition and competitive comparisons in this field.
Intel does have an impressive list of additional partners for its vision product and OpenVINO software stack including GE
Dell and Honeywell. Most of these customers are implementing or actively using the Intel vision-product technologies already, which is a strong signal that Intel has a successful launch on its hands.
This vision-processing initiative is part of Intel’s continued AI strategy and it hopes that this multi-chip solution provides better options for customers. But it risks introducing more confusion into a market that is still in its infancy, trying to understand what problems even need to be solved. But the monetary pressure to gain market share in this significant revenue opportunity is strong for a company that has been viewed as slightly behind competitors in the entirety of the machine-learning landscape.
Ryan Shrout is the founder and lead analyst at Shrout Research, and the owner of PC Perspective. Follow him on Twitter @ryanshrout.