IoT Connectivity and Compute Choices—AI Is Important, but Business Model/Process and ROI Still Drive Key Decisions

ABI Research’s most recent Artificial Intelligence and Machine Learning: TinyML market data (MD-AIMLT-102) shows spectacular growth for Artificial Intelligence (AI)/Machine Learning (ML) chipsets in the Internet of Things (IoT), reaching 4 billion units in 2030. Digging deeper, it shows that the smart home accounts for over 40% of the shipments through 2030. Automotive and manufacturing grab the 2nd and 3rd spots for shipments depending on the year with manufacturing showing more gains in shipments by 2030. AI/ML can provide huge savings and risk mitigation to manufacturing, given this sector’s breadth of system-level equipment (Programmable Logic Controllers (PLCs), generators, compressors, and battery systems, in addition to gateways and on-premises server) and use in production processes.

On the connectivity side, ABI Research’s Wireless Connectivity Technology Segmentation and Addressable Markets market data (MD-WCMT-198) shows shipments of Short-Range Wireless (SRW) chips reaching 25 billion units by 2030 across Bluetooth®, Wi-Fi, Ultra-Wideband (UWB), Near Field Communication (NFC), and 802.15.4 mesh technologies. AI/ML chipset shipments growth is not as high at 10% through 2030 due to more maturity in the bigger Wi-Fi and Bluetooth® technologies. The smart home is driving 40% of shipments across all technology sectors.

As AI introduces new opportunities for the IoT market, suppliers have been shifting their messaging (and roadmaps) to reflect how they are an important part of the device compute story. But on the ground, the device Original Equipment Manufacturers (OEMs) and customers of IoT solutions still need to make decisions in this new AI-dominated world across both connectivity and compute. Interestingly, connectivity and compute are not necessarily mutually dependent and can be choices that have little to do with each other or potential synergies.

The automotive market is seeing technology changing the automobile on multiple levels. The first is the number of sensors per car. Camera sensors per car currently range from 1 to 8, but the average per car will increase for safety/security and more autonomous features. The second is connectivity. By 2027, every region in the world will have a 5G passenger vehicle with China’s 5G penetration rate on new vehicle shipments reaching 30%. More interesting is that the newest Non-Terrestrial Network (NTN) satellite technologies will see growing penetration in vehicles for safety reasons and eventually to support critical software updates. Third, AI will continue to drive greater processing in vehicles to support car operations, safety, and customer experience. And interestingly, the number of processors in the car will decline from an average of 30 today to 7 by 2030. OEMs want more general-purpose computing for flexibility in their future use.

All these technologies are needed, most importantly the increase in the car’s compute capacity, because the customer revenue model is expected to change. Customers are keeping their cars longer, so auto OEMs are expecting more revenue to come from subscriptions. But OEMs do not know exactly the preferred automotive services of the future, so OEMs are adding greater compute in existing cars to prepare for the unknown new applications and services.

In the supply chain/logistics markets, more “things” are being tracked. Connectivity choices are changing away from more Wide Area Network (WAN) devices to one where a single WAN-connected tracker will collect data from slave devices via SRW, typically Bluetooth®. The business model for asset tracking is also expanding to reusable or disposable tracker models with an increasing share of implementations applying both models to address customer needs.

In the asset tracking market, a key need to support the Return on Investment (ROI) is that tracker devices need to be inexpensive. As a result, OEMs are not looking at adding more advanced hardware compute capabilities. And interestingly, at least today, the business model—either reusable or disposable—does not change the computing need. What is most important is low-cost devices and the right connectivity technology.

Overall, connectivity and compute decisions are completely driven by the IoT use case and vertical market conditions. However, a review across multiple IoT markets provides some common themes.

1. Greater Compute Will Go into Controllers and Machines: Typically, controllers and machines already have more compute because of the complexity of the process they are responsible for. Whether it’s a controller in manufacturing, a generator, a forklift, or a robot, they naturally will be using bigger Microcontroller Units (MCUs). This positions them for more use of AI applications, but it does not mean the MCU will be swapped out for a Graphics Processing Unit (GPU). For example, robot OEMs are not convinced they need to replace controller MCUs with GPUs.
2. Connectivity Cost Is Driving Sensing/Edge Devices; Less Interest in Changing Compute Capabilities: As noted above, computing at the extreme edge is not a priority at this point, not only due to cost, but also need. Gateways or other devices and machines can take on more edge processing tasks. A central aggregation point is also easier to manage and allows better monetization of the compute resources.
3. Greater Compute on Edge Devices Can Improve Safety: Machines in manufacturing such as cobots can use AI to improve safety for workers. Computer vision is a key enabler of improved safety capabilities driven by greater compute at the edge. Interestingly, regulations will also drive more compute at the edge for safety requirements as witnessed by ISO 10218-1 and ISO 10218-2, the recently released robot safety standard.
4. Connectivity and Compute Will Benefit SMEs as Quickly as Large Companies: Technology benefits often tend to go to the largest companies because they have the resources to invest in new technology and ride out initial implementation challenges. However, AI and connectivity have the potential to benefit Small and Medium Enterprises (SMEs) as quickly as the larger companies. As more apps move to the cloud, more things get connected, and AI is applied to detect and troubleshoot, SMEs will also have access to these solutions. In particular, AI and connectivity can be a substitute for on-site engineers and maintenance with OEMs leveraging AI to detect issues remotely and correct, or use teleoperations to guide non-technical personnel in maintenance activities.