Your phone's AI assistant is about to get a lot bigger. In an interview on October 8, Qualcomm CEO Cristiano Amon said multiple AI companies are pushing for smartphones that can keep a 100-billion-parameter model running continuously by 2028. That is more than triple the roughly 30 billion parameters that the newest flagship chips can run today.
Amon did not name the companies, and there is no delivery promise attached to the 2028 date. Still, the comment, first reported on October 11, offers a rare look at what AI labs are asking chipmakers to build. It also lands in the middle of a smartphone memory shortage. Here is what is actually going on.
What Qualcomm's Chips Can Do Today
In late September, Qualcomm unveiled the Snapdragon 8 Elite Gen 6 and the Snapdragon 8 Elite Extreme Gen 6 at its annual Snapdragon Summit. Both are built on TSMC's 2-nanometer process and are headed to flagship phones from brands such as OnePlus, Xiaomi, OPPO, vivo, Motorola, Honor and others.
The headline feature is on-device AI. According to reporting at launch, the top-tier Extreme chip can run a mixture-of-experts model of about 30 billion parameters locally, without calling the cloud. In plain terms, a mixture-of-experts model is a large model that only activates a portion of its parameters for each task, which keeps speed and power use manageable.
Reported specs also include:
- A new accelerator in the Hexagon NPU and a larger shared memory pool.
- A context window of roughly 32,000 tokens for on-device models.
- A new sensing hub that runs small models of up to 200 million parameters, enabling features like a local personal scribe that can tell speakers apart.
- Security features designed to control how AI agents access sensitive data.
Qualcomm is pitching these phones as an "AI hub": a device that processes AI requests itself rather than sending everything to a data center.
From Chatbot to Always-On Agent
The key detail in Amon's comments is the word "continuously." Today, most on-device AI loads a model, finishes a task, and shuts down. What AI companies reportedly want is an agent that stays resident in memory and keeps working in the background: watching context, reasoning, and acting on your behalf instead of waiting for you to open an app and type a prompt.
That shift is why the number matters. A phone that is merely able to run a large model for ten seconds is a very different product from one that hosts it all day without overheating or draining the battery by mid-afternoon.
Why Memory Is the Real Bottleneck
Larger models need more memory, both capacity and bandwidth. As a rough back-of-the-envelope estimate (not a figure from Qualcomm), a model compressed to 4-bit precision needs about half a gigabyte of storage per billion parameters. That puts a 30-billion-parameter model near 15 GB and a 100-billion-parameter model near 50 GB, before counting the memory the operating system, apps and the model's working context also need. A mixture-of-experts design reduces how much computing each response requires, but the full set of weights still has to be stored and quickly reachable.
Real-world conditions make this harder. The smartphone industry is currently contracting because of sharply higher memory prices, according to CNBC's coverage of the Snapdragon launch. Amon has acknowledged a "temporary constraint" on memory supply, but he frames it as pressure that will push the industry toward more memory-efficient chip designs instead of simply adding more memory.
Phone makers also compete for the same memory and advanced manufacturing capacity that data-center AI buyers are consuming. A 2028 target means Qualcomm, Apple, MediaTek and others will be chasing scarce supply while the server AI buildout continues.
The Bigger Race: Who Defines the AI Device?
Qualcomm is not only thinking about phones. Fortune has reported that Amon is working with nearly all major AI players on undisclosed hardware projects. Analyst Ming-Chi Kuo has separately reported that OpenAI has worked on an AI-agent phone program involving MediaTek and Qualcomm, with possible mass production in the first half of 2027 and combined 2027-2028 shipments estimated around 30 million units. Smart glasses, pendants and other always-on wearables are part of the same pitch.
Apple is moving in the same direction. At its recent iPhone event, the company framed its phones as "intelligent personal hubs" capable of on-device AI processing.
What This Means for Buyers
If you are shopping for a phone in the next year, none of this changes much today. A 30-billion-parameter on-device model is already an upgrade over earlier generations, and the 100-billion-parameter goal is a 2028 target, not a product. But a few practical takeaways follow:
- Expect higher prices for high-memory phones. Memory costs are squeezing the market, and manufacturers are focusing on premium devices.
- Privacy may improve. Models that run locally can keep more of your data on the device.
- Battery and heat will be the real test. Always-on AI is only useful if it does not drain your phone.
- Watch for agent security. As assistants gain access to your messages, files and accounts, how that access is controlled becomes as important as how smart the model is.
The Open Question
Amon was clear that 100 billion parameters is a direction the industry is pushing toward, not a Qualcomm launch. Fitting that much capability into a phone-sized battery and thermal budget is a problem no one, Qualcomm included, has publicly solved yet. Whoever solves it first will have a strong say in what a phone is by the end of the decade.
Frequently Asked Questions
What does "100-billion-parameter model" mean?
Parameters are the learned numbers inside an AI model. More parameters generally allow more capable behavior, but they also need more memory and power.
Can phones run large AI models today?
Yes, to a point. Qualcomm says its newest top chip can run a mixture-of-experts model of about 30 billion parameters on-device.
When could 100-billion-parameter phones arrive?
Amon cited 2028 as what AI companies are asking for. It is a target, not a committed product.
Why does the memory shortage matter?
Bigger models need more memory, and higher memory prices are already squeezing the smartphone market.
Sources
- Startup Fortune, "Qualcomm's CEO says AI firms want phones running 100-billion-parameter models by 2028" (Oct 11, 2026)
- TechCrunch, "Qualcomm launches two new smartphone chips with emphasis on AI" (Sept 22, 2026)
- CNBC, "Qualcomm releases Android chip built for AI as memory shortage weighs on smartphone market" (Sept 22, 2026)
