InvestNot yet confirmed elsewhere1 publisher2 min readPublished
AI firms want 2028 phones to run models more than three times what Qualcomm's 2nm chips handle
Qualcomm CEO Cristiano Amon says AI firms want phones running 100-billion-parameter models continuously by 2028, more than triple his newest chips' limit. Most of Qualcomm's plan for closing that gap is memory, the cost Amon says already deters phone upgrades.
The Investor · Invest desk
What happened
- At its September 22 Snapdragon Summit, Qualcomm unveiled the 2nm Snapdragon 8 Elite Gen 6 and Extreme Gen 6, built to run mixture-of-experts models of up to 30 billion parameters on the phone.
- Qualcomm's methods for fitting bigger models include loading parameters from flash storage, giving the neural processing unit more memory and using a low-power memory architecture.
- Amon linked anticipated growth in smartphone sales to these AI capabilities and said rising memory prices have held back users from upgrading.
- OpenAI is reported to be looking at hardware that would target comparable AI capabilities by 2028; no firm details have come out.
Why it matters
- cost If memory prices stay high, buyers pay for bigger on-device models through dearer handsets, just when the industry needs them to upgrade.
- constraint With no shipping part above 30 billion parameters, the 2028 request cannot yet be turned into unit or price estimates for handset and chip suppliers.
- exposure Amon's case that the phone stays the personal-AI hub because it holds a user's context is exposed if AI firms ship their own devices for the same 2028 capability.
The 3.3-times gap [9] compares the request with the largest model the new chips can hold. Continuous operation is what Amon said, in an October 8 interview, AI companies want [1]. The models Qualcomm runs continuously today sit in its always-on sensing hubs and are about 200 million parameters [5]. Measured against those, the request is 500 times larger [10]. Qualcomm's figure of about 1 million tokens a day [5] comes to roughly 11.6 tokens a second spread across 24 hours [11].
Each method Qualcomm named for fitting bigger models is a way around memory limits [4]. The memory price Amon blames for stalled upgrades [6] attaches to the part a 100-billion-parameter phone needs more of.
The evidence fits more than one outcome. Model sizes may climb from 30 billion while memory prices ease; then Amon's sales case holds as he made it. A slower version has sizes rising only as fast as handset makers pay up for memory, with the first money going to memory suppliers. Flash loading might instead carry most of the jump and move the extra cost into storage.
We think the second is the likeliest on this record. Qualcomm's methods are mostly memory workarounds, and its chief executive says memory prices are why buyers hold off. The third is the counter-case, and it comes from Qualcomm's own design: parameters are loaded from flash so the full model does not have to sit in active memory [4]. If that approach scales from 30 billion to 100 billion, the upgrade money goes to storage and to Qualcomm's silicon, and much less of it to memory.
Qualcomm's own commitment stops at 30 billion parameters [3]. The 2028 figure is what AI companies are asking for, not a product anyone has promised, according to Crypto Briefing [2]. The person relaying the request runs the company that would sell the chips [1]. We would be wrong about memory setting the pace if Qualcomm's next platform lifts the on-device ceiling well past 30 billion without giving the neural processing unit more memory.
What to watch
- Whether Qualcomm or any AI company turns the always-on, 100-billion-parameter target into a committed product with a ship date and a price.
- The direction of memory prices between now and 2028, since Amon says they are holding back phone upgrades.
- Whether OpenAI puts real hardware details on the table, and whether its device depends on a phone at all.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence35
- Adoption
- Insufficient
- Hype gap+25
- Incentives75
- Confidence40
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
In an interview on October 8, 2026, Qualcomm CEO Cristiano Amon said AI companies are already asking for smartphones that can run 100-billion-parameter models continuously by 2028.
ReportedSupportedSource: Cristiano Amon, per Crypto Briefing2 sources— create a free account to open themView cited source - [2]
The 2028 target reflects what AI companies are asking for, not a product anyone has promised to deliver.
ReportedSupportedSource: Crypto Briefing analysis2 sources— create a free account to open themView cited source - [3]
At its Snapdragon Summit on September 22, 2026, Qualcomm unveiled the Snapdragon 8 Elite Gen 6 and Extreme Gen 6 platforms, built on 2nm technology and designed to run mixture-of-experts models of up to 30 billion parameters on the device.
- [4]
To support larger models Qualcomm loads model parameters from flash storage so the full model does not need to sit in active memory, expands the memory available to the Neural Processing Unit, and points to intelligent flash management and a low-power memory architecture.
- [5]
Always-on sensing hubs run much smaller models, at approximately 200 million parameters, continuously in the background; Qualcomm says its new chips can process approximately 1 million tokens per day.
- [6]
Amon linked anticipated growth in smartphone sales to these AI capabilities, noting that rising memory prices have held back users from upgrading.
- [7]
Amon argued that phones will remain the central hub for personal AI because they hold the context about a person's life that makes an assistant useful.
- [8]
If memory costs stay elevated, the hardware needed for these models could push phone prices higher at the moment the industry hopes to lure buyers back.
- [9]
The 100-billion-parameter request is about 3.3 times the 30-billion-parameter ceiling of Qualcomm's new 2nm Snapdragon platforms.
- [10]
The 100-billion-parameter continuous model AI firms want is 500 times the roughly 200-million-parameter models Qualcomm's sensing hubs run continuously today.
- [11]
About 1 million tokens per day equals roughly 11.6 tokens per second averaged over 24 hours.
- [12]
OpenAI is reportedly exploring hardware aimed at similar AI capabilities by 2028, though firm details have not been disclosed.
Sources
1 independent publisher whose own reporting we read for this story.
- cryptobriefing.comQualcomm CEO says AI firms want 100B-parameter models running on phones by 2028
1 article · October 10, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.