The overwhelming majority of smartphones shipping today have silicon that drives some integrated Artificial Intelligence (AI) capabilities. The industry is moving fast to embrace next-generation chips that will drive new and exciting features and interaction modes.
IDC defines “next-gen AI smartphones” as devices with a system-on-a-chip (SoC) capable of running on-device Generative AI (GenAI) models more quickly and efficiently leveraging a neural processing unit (NPU) with 30 Tera operations per second (TOPS) or more performance using the int-8 data type. These new devices are generating a lot of interest amongst consumers and OEMs, making AI the focal marketing message at recent flagship launches, with more to follow this year.
IDC’s latest forecast estimates that GenAI smartphone shipments will grow 364% year-over-year in 2024, reaching 234.2 million units, growing to 912 million units in 2028 implying a compound annual growth rate (CAGR) of 78.4% for 2023-2028.
While AI will impact all device categories, smartphones will lead the charge from a pure reach and volume perspective, quickly outpacing the forecasted volumes of AI PCs. What follows is IDC’s initial definitions around AI smartphones and an early look at the impact we expect it to have on the market.
AI Smartphone Definition
Smartphones capable of running on-device AI have existed for nearly a decade; however, more recently, the term AI smartphone has come into play to describe some of the latest flagship devices with on-device GenAI capabilities that are creating increased interest and excitement in the industry. IDC categorizes AI smartphones into two distinct groups, with these newer AI smartphones as the second category below:
- Hardware-enabled AI Smartphones (≤30NPU TOPS): These smartphones use accelerators, or specialized processors aside from the main application processors, to run on-device AI at lower power. More recently, this includes a shift to the use of neural processing unit (NPU) cores with up to 30 TOPS performance using the int-8 data type. Examples of on-device AI are natural language processing (NLP) and computational photography. These smartphones have been on the market for nearly a decade.
- Next-gen AI Smartphones (>30 NPU TOPS): These smartphones use SoCs capable of running on-device GenAI models more quickly and efficiently and have an NPU with at least 30 TOPS performance using the int-8 data type. Examples of on-device GenAI include Stable Diffusion and various large language models (LLMs). This category of smartphones first hit the market in the second half of 2023.
The smartphone SoCs being designed and marketed by silicon vendors with next-gen AI smartphones in mind will proliferate in the future as they continue to push forward the NPU technology. However, to date, here are a few that qualify based on the definition above:
- Apple A17 Pro
- MediaTek Dimensity 9300
- Qualcomm Snapdragon 8 Gen 3
The AI Smartphone Journey
The first thing to know about on-device AI in smartphones is that on-device AI has been a part of smartphones for years in the form of natural language processing (NLP) and computational photography. The data models to do this are usually smaller than the models that run in the cloud on servers but work well enough to get the job done.
AI is run on devices for speed of response, privacy, and security. On-device GenAI is newer, and the industry’s discussion of AI smartphones is centered around this. IDC will use the term next-gen AI smartphones when referencing these newer AI smartphones focused on device GenAI as defined above.
Next-gen AI smartphones are about performing inferencing on the device and specifically on-device GenAI through the inclusion of LLMs and text-to-image models, among others. However, just like putting a game on a PC does not make it a gaming PC, putting an LLM or two on a smartphone does not make it a next-gen AI smartphone. It is the fact that the smartphone’s SoC is designed with specific accelerators, or specialized processor cores, that are optimized to run LLMs quickly and efficiently with less power consumption than if the main processor cores were the primary workhorse. These specialized cores are called neural processing units (NPUs) typically.
There have been phones on the market with NPU cores already, but typically with fewer cores and less performance. Processors just hit the market in the second half of 2023 with more powerful NPU cores and a greater number of cores since they were designed with on-device GenAI in mind. One could define an AI smartphone based on quantitative specifications such as core count, the aggregate processing power of those cores, or total processing power across all application processor cores and all accelerators. The typically minimum amount of DRAM could be part of the definition, too. However, these are all moving targets and will keep changing. Furthermore, GenAI models can be compressed and quantized further to fit into phones with lesser specifications.
To create a knowable segmentation between hardware-enabled AI smartphones and next-gen AI smartphones, IDC has drawn the line at 30 TOPS. The smartphone SoCs powering today’s next-gen AI smartphones listed above are in the range of 30 to 45 TOPS of NPU performance using the Int-8 data type.
Customer Demand and Future Applications
The sales of many next-gen AI smartphones in the first year or two will likely be driven by the sheer fact that they are flagship phones. But the arrival of phones capable of running GenAI on the device will lead to more application development and next-gen AI smartphones will become increasingly capable. A later evolution could include a very large AI model that is a much more personalized and proactive assistant. This is likely what Google is aiming for with Gemini, Microsoft is aiming for with a future version of Copilot on smartphones, and what Apple is likely also planning. And this is where the excitement of consumers and industry really comes into play, at the untapped potential of what this technology could bring in the next phase of evolution, rather than the basic applications and use cases that exist today.
Hardware Implications – An Opportunity and Challenge for OEMs
Before we get to the next phase, OEMs will need to address the other hardware requirements of next-gen AI smartphones outside of SoCs. One of the biggest variables will be not just the premium cost of the SoC, but the cost of additional DRAM that will be necessary to support the super powerful SoCs of these devices.
Where 16 GB is a large amount of RAM for most smartphones, 16GB is already considered a minimum requirement for next-gen AI smartphones. A general doubling of memory included in these phones will be far more than double the recent historical cost of memory, as the cost of DRAM is increasing. Most flagship phones will tend to be loaded with premium configurations and components, including better screens and cameras that would optimize the use of multi-modal GenAI around imaging.
This means that next-gen AI smartphones will also come with a larger bill of materials (BOM) cost, which presents both an opportunity and a challenge for the OEMs. OEMs can leverage the innovative technology and its obvious benefits to pass the added BOM cost to the end consumer with a higher sticker price and use the opportunity to raise the value of the industry or absorb the additional cost between the channel and OEM.
Learn how to take a data-driven approach to managing and developing the right partnerships in IDC’s recent playbook.
Although a combination of the two is the likely outcome, the good news is that smartphone ASPs have been going up in recent years, and the share of premium devices continues to grow with no signs of slowing down. Consumers are getting used to and even willing to pay more for their smartphones. As flagship smartphone prices increase, increasing availability of multi-year payment plans and other promotions like aggressive trade-in offers, will make it easier for consumers to purchase these higher-priced next-gen AI smartphones.
Looking Ahead
IDC forecasts 170 million next-gen AI smartphones to be shipped in 2024, representing almost 15% of the total smartphone market. This will be a significant leap from the 51 million devices shipped in 2023, more than tripling in volume in just one year. Next-gen AI smartphones will continue to grow rapidly in the coming years as use cases evolve and as the OEMs, silicon vendors, and industry players continue to drive processing power and adjoining hardware specifications to support the growing demand for these intelligent smartphones.
Next-gen AI PCs are expected to take off this year, too, but the scale of the smartphone market means the volume of next-gen AI smartphones will quickly surpass the PC. IDC forecasts AI PCs to ship 167 million units by the end of 2027, while smartphones will cross that number this year alone. Therefore, while AI as a technology will impact all corners of the devices market, smartphones will be the device driving the AI revolution into every home.
Contributing authors:
