100 Billion Parameters on Your Phone: A 2028 Dream or Reality?
Leading AI companies are requesting smartphones capable of continuously running 100-billion-parameter AI models by 2028, a monumental challenge that faces significant memory and thermal hurdles, making it more an industry aspiration than a guaranteed outcome.
AI-generated image
Is the smartphone as we know it about to be completely redefined, or is Qualcomm painting a picture that’s still years, if not decades, away from reality? Qualcomm CEO Cristiano Amon has ignited a fervent debate with his recent assertion that leading AI companies are pushing for smartphones capable of running 100-billion-parameter AI models continuously by 2028.
Speaking on the Sources podcast on October 8, 2026, Amon revealed that these AI pioneers are not merely asking for incremental improvements, but a monumental leap in on-device intelligence. This isn't Qualcomm's product roadmap or a delivery commitment; it's an industry aspiration, a demanding request from those at the forefront of AI innovation for devices that can host truly agentic AI. Amon highlighted this target during a period when the broader smartphone market is maturing, with memory costs putting additional pressure on device prices, arguing that AI could provide the next major reason for consumers to upgrade.
The Memory Wall: A Herculean Task
The most immediate bottleneck for on-device AI models of this scale is memory. Qualcomm's newest flagship mobile platforms, the Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6, unveiled on September 22, 2026, are designed to run mixture-of-experts models above 30 billion parameters locally. Going from 30 billion to 100 billion parameters is more than a threefold jump, inside the same rough physical footprint of a phone, and companies want these models running nonstop, not on demand. As Wccftech notes, this is a Herculean task.
To put this into perspective, running a 4-bit-quantized 100-billion-parameter model would require a smartphone to have at least 50GB of RAM, with sufficient memory left for the operating system, applications, and context cache. Given that current flagship chips like the Snapdragon 8 Elite Extreme Gen 6 are specified with up to 24GB of RAM, the gap to 50GB is substantial. This memory constraint becomes even more pronounced considering that a smartphone cannot dedicate all its RAM to a single program. Beyond just the quantity of memory, continuous operation of such dense models would immensely contribute to heat generation, leading to thermal throttling and significant battery drain, demanding new low-power architectures.
Beyond Memory: The Agentic AI Shift
The vision Amon outlines extends beyond raw parameter counts; it's about a shift towards
No topics yet: start the first one.
More stories
Smartphone Overload: The Industry's Unsustainable Race to Nowhere
With a new smartphone appearing every 12 hours in Europe, and giants like Apple and Samsung slashing production, the market is signaling a critical breaking point.
Midnight Mimosa: The Malware That Infects Your Phone Before You Do
The discovery of Midnight Mimosa pre-installed in inexpensive Android phones, especially those with MediaTek chips, reveals a deep-seated supply chain security flaw, leaving budget-conscious consumers unknowingly compromised from day one.
HyperOS 4 isn't just an update, it's Xiaomi's play against the walled garden
Xiaomi's HyperOS 4 makes an aggressive push for cross-platform integration, offering a glimpse into a future where traditional mobile ecosystem walls might finally begin to crumble.
The Thermal Tipping Point: Why Advanced Phone Cooling Is No Longer Niche
The advent of integrated RGB liquid cooling in phones like the RedMagic 12 Pro Plus and compact external solutions from OnePlus signals a critical shift where thermal management, once niche, is now essential for the future of mobile performance.
