Qualcomm plans to use the new chip architecture, originally developed for data centers, in smartphones in the future. The goal is for mobile devices to more efficiently execute artificial intelligence models locally, without the constant reliance on remote servers and the need to access them via an Internet connection.
Durga Maladi, executive vice president of Qualcomm, confirmed in an interview that the technology introduced along with the new data center chips will not remain limited to the server market.
“What starts in data centers will not end there,” said Maladi.
Qualcomm is already talking to manufacturers of smartphones, personal computers and cars about the implementation of part of the new technology portfolio. At its heart is the High Bandwidth Compute architecture, which places memory and computing units much closer together.
Vertical stacking of memory and processor enables faster local AI with lower power consumption
Instead of the traditional arrangement of chips next to each other, Qualcomm suggests stacking them vertically. This shortens the path that data has to travel between memory and processing units, increases bandwidth and speeds up the processing of large amounts of information.
A similar concept is not completely new, but until now it has mainly been used in server and data center systems, where high performance and fast memory access are of crucial importance. Its application in phones requires additional adaptation to energy consumption, chip dimensions and limited cooling capabilities.
The first generation of Qualcomm High Bandwidth Compute architecture should be introduced in data centers during 2027, while the commercial availability of the new server chips is expected in 2028. The company has not yet revealed when the same technology could make its way to smartphones and other end-user devices.
When that happens, phones could run larger and more complex AI models directly on the device. This would enable faster responses, greater data privacy and the operation of functions even without internet access.
The new architecture could be especially important for so-called always-on AI agents. They could monitor context, process data, and execute tasks in the background without draining the battery at a rate that would make such functionality impractical.
Qualcomm has lagged behind the leading companies in the data center chip market in recent years, but its decades of experience in developing mobile Snapdragon platforms gives it a different starting position. The company is now trying to transfer technologies from the server environment to phones, computers and cars, while maintaining the energy efficiency required by devices that run on batteries, Semafor reports.