Qualcomm CEO Cristiano Amon has put a concrete number on the industry’s ambition to bring advanced artificial intelligence into people’s pockets. In an interview with the tech newsletter Sources, titled “Qualcomm’s CEO on the coming AI phone supercycle” and recorded around October 8, 2026, Amon said that some of the most advanced artificial intelligence companies have given him an explicit requirement: being able to run a model of at least 100 billion parameters on a phone by 2028.
“I have to be careful not to disclose names, but some AI companies right now, some of the companies at the forefront of AI, are telling Qualcomm, ‘Qualcomm, by 2028, I need to make sure that I have the ability to run at least 100 billion parameter model into a phone, and I need it to have the thing running kind of all the time,’” he explained. The nuance matters: this is not about loading a large model for a one-off task and shutting it down, but about keeping it running continuously.
The number does not come from Qualcomm’s roadmap, but from what its customers are asking for. Amon has not tied the date to any specific product or committed to a launch schedule; he presents it as a target the industry is pushing toward. Nor has he identified the companies making the request. Even so, the figure is significant because it comes from the largest maker of chips for Android phones, which supplies processors to many of the world’s major manufacturers.
To size the challenge, it helps to look at the present. Qualcomm’s newest flagship platforms, the Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6, are built on a 2-nanometer process and are headed to phones from brands such as Honor, Motorola, OnePlus, OPPO, vivo and Xiaomi. According to published information about these chips, the Extreme version, with its Hexagon neural unit and a larger shared memory pool, can run mixture-of-experts models of just over 30 billion parameters on-device, with a context window of about 32,000 tokens.
The 2028 target therefore means more than tripling the scale of the models and, on top of that, keeping them resident in memory all day. That drives up the demands on memory bandwidth, power draw and heat dissipation in a device that has to fit in a hand, not overheat and last a full day on a single charge. No manufacturer, Qualcomm included, has yet explained how that equation will be solved within a phone’s battery and thermal budget.
What these companies want is not a bigger version of the assistant we already know. Amon describes a shift toward agentic artificial intelligence: software that watches, reasons and acts on the user’s behalf, using their apps and drawing on their personal context, instead of waiting for someone to open an app and type a prompt. For him, this agent-centric phone, with context stored on the device itself, contrasts with approaches that rely entirely on the cloud.
Amon compares the current moment to the early years of the iPhone, when the applications that would end up defining the platform had not yet been invented. “The smartphone was defined by the third-party applications,” he said, and he argues that agents will transform the phone experience in a similar way. In his view, these new capabilities will unleash an upgrade “supercycle” among users who have long been putting off buying a new device.
That delay has a known cause: rising memory costs. Amon acknowledged in the interview that the smartphone market is going through a weak patch because of soaring memory chip prices, and he has spoken of a “temporary constraint” on supply. The same memory suppliers and the same advanced manufacturing capacity that phones need are under pressure from Nvidia, AMD and the large cloud providers, which are buying components for their artificial intelligence servers.
He also confirmed, according to Dealroom’s summary, that some artificial intelligence companies are starting to build their own phones and that more are coming. Qualcomm makes chips for Android phones, for Meta’s smart glasses and for upcoming devices from companies such as OpenAI, which gives it a role in several competing visions of the next big artificial intelligence device. Amon said smart glasses are the AI wearable category he is most bullish about.
The conversation covered other fronts. Amon discussed the acquisition of Modular, through which Qualcomm wants to offer developers an alternative to Nvidia’s software, and explained that this open platform will be licensed under Apache 2.0. He also said the company will demonstrate 6G at the 2028 Los Angeles Olympics, with prototypes and first silicon between 2029 and 2030 and deployment at scale after 2030. According to Amon, 6G will largely be a network designed for physical artificial intelligence.
He also compared artificial intelligence adoption in China and the United States. In his view, China is moving faster on device-level integration, while the United States has focused more on frontier models and enterprise use. That difference helps explain why Chinese phone makers are among the first customers for the new Snapdragon chips and why the battle for agents on phones is being fought on a global scale.
Amon’s statement adds to other signs that the next chip war will not be fought only in data centers. Analyst Ming-Chi Kuo has reported, for example, that OpenAI is working on an agent phone program involving MediaTek and Qualcomm, although the companies have not confirmed that information. What does seem clear is that the big labs want their models to work without always depending on a connection, for reasons of privacy, latency and cost.
The 100-billion-parameter target for 2028 is, for now, a customer request rather than a technical promise. But it shows how the labs picture the phone at the end of the decade: a device that hosts a permanent agent, able to understand its owner’s context and act for them. Whichever manufacturer manages to build chips efficient enough to meet that goal will have a lot to say about what a phone will be in the years ahead.
