Qualcomm's CEO says AI firms want phones running 100-billion-parameter models by 2028.
Cristiano Amon says the AI industry is already asking Qualcomm for phones that can run 100-billion-parameter models all day, every day, by 2028. That is more than triple what this year's flagship chips can do, and it arrives on top of a memory supply crunch that Amon himself acknowledges is already straining the industry.
In an interview on October 8, Qualcomm's chief executive said multiple AI companies are pushing for this capability, though he declined to name them. They want smartphones that can keep a 100-billion-parameter model running continuously, not just load one for a single task and shut it down. The figure matters because it doesn't come from Qualcomm's own roadmap. It reflects what AI labs are telling the world's largest mobile chipmaker they want in consumers' pockets.
Qualcomm's newest flagship mobile platforms, the Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6, are built on a 2nm process and are slated for flagship phones from brands including Honor, iQOO, Motorola, OnePlus, OPPO, Redmi, RedMagic, vivo and Xiaomi. The higher-end Extreme version's Hexagon NPU, which includes a new Element Accelerator and a larger shared memory pool, can run mixture-of-experts models above 30 billion parameters on-device, with a context window of roughly 32,000 tokens. Amon's 100-billion-parameter target is not a modest step up from that. It is more than three times the scale, and the companies requesting it want it running nonstop rather than on demand.
What's being requested is not simply a bigger version of today's assistant. Amon described it as a shift toward agentic AI: software that watches, reasons and acts on the user's behalf instead of waiting for someone to open an app and type a prompt. Such a system has to stay resident in memory and keep running inference in the background all day, without making the phone too hot to hold by lunchtime or leaving it dead by mid-afternoon.
Today's 30-billion-parameter on-device models already push Qualcomm toward new memory architectures to keep latency and power draw in check. Tripling the parameter count while keeping the model running continuously, rather than spinning up and down, multiplies memory bandwidth and battery demands in ways current flagship hardware was not built for. Amon has acknowledged a "temporary constraint" on memory chip supply for smartphones, but he frames it as pressure that will push the industry toward more memory-efficient chip designs rather than simply more memory.
That supply pressure is not occurring in isolation. The memory suppliers and advanced-node capacity that phone chipmakers rely on are also under strain from Nvidia, AMD and the hyperscalers buying parts for data-center AI servers. A 2028 target for 100-billion-parameter phones would mean Qualcomm, Apple and MediaTek competing for scarce memory investment and leading-edge manufacturing capacity while the server AI buildout is already stretching production.
**Qualcomm wants in on more than one device**
Qualcomm's ambitions extend beyond phones. Fortune has reported that Amon is working with "pretty much all" of the major AI players on undisclosed hardware projects. Separately, supply-chain analyst Ming-Chi Kuo reported that OpenAI has worked on an AI-agent phone program involving MediaTek and Qualcomm. His later reporting put possible mass production in the first half of 2027 and estimated combined 2027–2028 shipments at around 30 million units. Smart glasses, pendants and other always-on wearables are part of the same pitch: devices built around an assistant rather than a home screen full of apps.
That is the real significance of the 100-billion-parameter figure. It is not a spec-sheet boast. It is Qualcomm signaling, two years before the hardware exists, that the chip race for agentic AI will be fought on-device as well as in the cloud. Whoever ships silicon efficient enough to reach that target first could shape what a phone is by the end of the decade.
Qualcomm has not said which companies are making the request. The 2028 date also carries no delivery commitment. Amon has been clear that it is a target the industry is pushing toward, not a Qualcomm product launch. Given how far current flagship silicon sits from that bar, the open question is how to reach it within a phone-sized battery and a phone-sized thermal budget, and Qualcomm has not yet answered it.