Generative AI to your smartphone is the next big thing… Qualcomm’s trajectory underlines this

At a time when smartphone sales are cooling and phones are becoming increasingly predictable in terms of their look, functionality, and performance to some extent, Qualcomm is turning to generative AI as a means to stimulate demand for high-end devices in a competitive landscape where having just a good camera or display may not be enough to justify a purchase. But with Generative AI, companies like Qualcomm and handset makers such as Xiaomi and OnePlus might be looking at new possibilities to create demand and increase upgrade cycles for phones.

Cristiano Amon, CEO of Qualcomm, speaking during the opening day of the Snapdragon Summit on Tuesday. (Anuj Bhatia / Express photo)

Which is why Qualcomm isn’t placing much emphasis on the raw performance of the Snapdragon 8 Gen 3, its latest flagship mobile processor that will power all the high-profile Android smartphones from the world’s largest handset manufacturers. Certainly, the chipset is faster and more efficient than the previous-generation processor, offering improved gaming performance and camera upgrades, but it’s evident that this year Qualcomm is dedicating more time to discussing the on-device AI capabilities of the Snapdragon 8 Gen 3, and the reason for this is becoming clear.

During the on-stage presentation and later in the demonstrations given to the press, the Snapdragon 8 Gen 3’s generative AI capabilities were front and centre. Under the hood, the chipset’s AI Engine supports multi-modal generative AI models, resulting in on-device Stable Diffusion. This allows users to suggest prompts like “a man playing football in the desert” and receive a custom image rendered on their phone in about one second. Additionally, there’s something Qualcomm calls “Photo Expansion,” which relies on generative AI to expand the image’s background, and a feature known as Video Magic Eraser, capable of removing subjects from a video to eliminate distracting elements.

Qualcomm also showed the 8 Gen 3 processor supports a chatbot trained on Meta’s Llama 2, and it can accept text, image, and voice input. The list of features goes on, but the point Qualcomm is trying to make is that the magic of the silicon allows powerful Generative AI models to run locally on phones and can be used even when the device is offline and without an internet connection.

On-device AI features have been present on smartphones for years, but their capabilities were limited. More advanced Generative AI features, in recent months, have only been accessible through the cloud, requiring users to have a constant internet connection, such as accessing the free version of ChatGPT on their smartphones. However, Qualcomm wants to change that with on-device AI. During the keynote, Qualcomm highlighted the advantages of on-device AI, which include user privacy, optimised battery life, rapid response to queries, and enhanced personalization. Truth be told, there are also economic benefits of using on-device AI, which is much cheaper to implement than maintaining large servers for complex machine learning models.

Right now, you can’t really run large models on-device. But experts say as things progress and Gen AI gets more mature, it will be possible to run complex models on the device itself.

But it’s not just Qualcomm that is banking on-device Generative AI. Google, Microsoft and even Apple are all eyeing this as the next frontier and supercharging their efforts in on-device AI.

“On-device Generative AI is essentially running a more distilled version of cloud models on a local device, and while it is still very early in the development of this technology, it’s very exciting. I think the possibilities are significant,” Google Senior VP of Devices and Services, Rick Osterloh, told indianexpress.com last week when he was in India to announce the manufacturing of Pixel 8 smartphones in the country.

Google’s newly announced flagship, the Pixel 8 Pro, is the first hardware to run Google’s generative AI models on-device. By using on-device models, Google has not only improved Magic Eraser, its post-processing tool for photo enhancement but also extended its capabilities to audio recording. Google said Tuesday it is also working with Qualcomm to allow a number of its AI models to run on Android devices.

Apple, too, is not far behind in the AI race, even though it may not be talking as much as others do in Silicon Valley. In fact, Apple is incorporating AI-led features more than ever in its products but in a subtle manner. Consider the new Journal app, which has the ability to make personalised suggestions based on your activities in other apps, such as Photos, Music, or even your location. Or think of iPadOS 17’s “enhanced autofill” feature on the iPad, which identifies fields in a PDF and then auto-fills personal information using saved data from Contacts. But the question is how quickly Apple can add Gen AI to its iOS and macOS baked at the core level. Running on-device AI models would be in line with the company’s privacy policies, a move many anticipate in the industry.

The writer was in Maui, Hawaii, attending the Snapdragon Summit on the invite of Qualcomm India.