Artificial Intelligence (AI) and the Internet of Things (IoT) are two of the most impactful and far-reaching technology developments of our time. Increasingly these two technologies are deployed together and the term ‘AIoT’ has come to the fore.
IoT devices can generate significant amounts of potentially valuable data and AI can be applied to this data in many locations, including ‘cloud’ data centres, various network edge locations, and on board the actual IoT devices themselves. All of these approaches have the potential to unlock significant value and new insights from IoT data.
When we refer to ‘AIoT’ we mean specifically the deployment of AI Use Cases on board IoT devices.
Applying AI to IoT data on board the source IoT devices can bring significant benefits, including improved performance, enhanced compliance, privacy and security and potentially reduced operational costs. These benefits are discussed in more detail below.
Meanwhile, applying AI to IoT data in cloud locations allows for access to the greater compute power that is typically available in those locations. It also enhances the potential scope of AI applications that can more readily draw inputs from multiple IoT devices in different locations and can potentially augment these data streams with other data (including from non-IoT sources) to support more sophisticated analyses. Deploying AI in cloud locations also allows for significantly simplified AI application management, primarily in terms of version control and the evolution of AI algorithms over time.
In many cases, therefore, AIoT (i.e. AI deployed on board IoT devices) does not exist in isolation and often AIoT devices may generate data that is further processed using potentially more sophisticated AI algorithms either in cloud or edge locations (or both).
As described above, the deployment of AI on board IoT devices can potentially unlock significant benefits. In the following paragraphs we discuss these in more depth.
AIoT can improve IoT application performance by allowing for the analysis of more inputs at greater resolution. For example, in the case of AI-enabled CCTV, AIoT algorithms can support activity recognition analyses of high-resolution video feeds locally, so averting the need to transmit vast quantities of data to remote locations for analysis. Inevitably this enables those video feeds to be analysed in more detail compared to crude redaction-at-source approaches that might otherwise be applied.
AIoT enables alerts, analytics and applications to run faster closer to the originating source of data. This can be particularly critical in the case of applications such as self-driving cars and similar autonomous systems, or in critical situations such as ‘emergency stop’ processes in industrial contexts that might apply when a machine is operating outside its design parameters. Applications like these are not tolerant of the lags that would be inherent in transmitting data to a remote location for processing and subsequently receiving the results of such processing.
AIoT can also increase the resiliency and uptime of IoT applications, which is particularly critical for applications that must continue to work in the scenario that a wide area connection fails. Again, autonomous vehicles and critical industrial processes provide good examples, but the potential benefits of increased resilience and uptime also extend to improving the value proposition and user-experience associated with a range of less-critical IoT applications such as, for example, AIoT enabled HVAC systems.
AIoT can help to protect privacy by processing data locally, so that personal data is never communicated beyond the AIoT device. For instance, an AIoT smart speaker could process voice commands locally, rather than streaming those commands (and likely other background sounds, including potentially unrelated conversations) to remote locations for processing. When data is sent to remote locations, for instance to trigger a web search, it can be transmitted in an anonymised way.
In enterprise contexts, similar techniques can help to ensure that critical or confidential information remains local and is not transmitted beyond corporate boundaries, so enhancing compliance and security.
AIoT can help to reduce operational costs in several ways. Firstly, it can be deployed to support sophisticated on device performance monitoring, optimisation and pre-emptive maintenance, also improving the value proposition of an IoT application.
By supporting the undertaking of sophisticated analyses locally, AIoT can significantly reduce the communication costs associated with data-intensive IoT applications. This benefit is particularly applicable in the case of applications that might otherwise generate significant volumes of data that would need to be analysed in remote locations.
AIoT can also unlock the potential for IoT solutions to be deployed in more marginal situations, so reducing human resource costs. For instance, AIoT enabled CCTV cameras can be deployed to monitor locations where activities are expected to be infrequent, potentially to monitor an ‘out of bounds’ location in an industrial complex. Such a solution could be configured to only communicate small amounts of data when activity is detected, so minimising bandwidth costs while reducing the need for security guards to monitor the same location.
Of course, the deployment of AI on board IoT devices can also reduce the costs of cloud processing that might otherwise be applied to the same source data, and associated storage costs.
The potential for AIoT is significant. Transforma Insights forecasts that total AIoT connections will grow from 1.4 billion at the end of 2023 to 9.1 billion at the end of 2033. This is a more than 6-fold growth in 10 years, resulting in a CAGR of over 20%.
Overall AIoT represents a significant market with net additions growing from less than half a billion in 2023 to just over 900 million in 2033.
For context, we forecast a total of 39 billion IoT devices at the end of 2033, up from 16 billion at the end of 2023. Accordingly, as illustrated below, we forecast that 9% of IoT devices will have on board AI in 2023, rising to 23% in 2033. The rate of growth of AIoT penetration of IoT devices slows towards the end of the forecast period, primarily due to the AIoT penetration of key IoT applications reaching saturation.
Further forecast information can be found in our recent 'Global AIoT Forecast, 2023-2033' report and the associated webinar 'AIoT: the impact and opportunities of combining AI and IoT'.
The emergence of advanced AI applications is acting as a catalyst for the development of advanced and sophisticated AI chips. While superior computational performance has traditionally been the primary focus for chip makers, the diverse requirements of AIoT applications demand a balance between computational power, memory, latency, energy efficiency and chipset area and cost to meet context-specific needs. Meanwhile, the rise of smaller, application-specific AI models is further catalysing the development of use-case optimised edge AI and AIoT chips.
Accordingly, markets for AIoT compute power are becoming more fragmented to match the diverse needs of AIoT applications, with a priority placed on optimising cost, power consumption and processing power for specific use cases resulting in significant levels of innovation and opportunities for new market entrants.
For instance, autonomous vehicles demand higher compute performance and can accommodate higher power consumption for image processing while a smart camera processor may require a lower cost chipset that is more power-efficient.
With the growing number of IoT devices and the integration of AI into these devices, the demand for hardware that efficiently can support AIoT applications will grow, and so the development of application- and use case-optimised AIoT components will accelerate.
Newly emerging AI-optimised SoCs (Systems-on-Chips) can efficiently support specific target applications and rapidly evolving requirements for specialised AI chips will support a proliferation of companies developing AI SoCs for different use cases in an increasingly diversified and specialised marketplace.
