Apple is in talks with PrismML to test technology that compresses large AI models for native operation on the iPhone [1].
This move represents a strategic shift toward on-device processing. By reducing reliance on cloud servers, Apple can lower operational costs while delivering faster response times and enhanced privacy for its users [1, 3].
PrismML is a Caltech spin-out backed by Khosla Ventures [1, 2]. The startup specializes in AI model compression, a process that allows massive neural networks to fit within the hardware constraints of a smartphone without losing critical functionality [3].
Apple intends to use this technology to privatize and accelerate Siri and other integrated AI functions [1, 2]. Currently, many complex AI tasks require data to be sent to remote servers for processing, which can introduce latency and create potential security vulnerabilities [3].
If the partnership proceeds, Apple could deploy some of the largest AI models ever seen on a mobile device [3]. This would allow the iPhone to handle sophisticated generative tasks locally, meaning the device processes the information without an internet connection [2, 3].
The discussions are taking place between Apple's headquarters in Cupertino, California, and PrismML's base in California [2, 4]. The integration of PrismML's compression techniques would allow Apple to maintain its focus on user privacy, a core pillar of the company's branding [1, 3].
“Apple is in talks with PrismML to test technology that compresses large AI models for native operation on the iPhone.”
This potential partnership signals a broader industry trend toward 'Edge AI,' where the intelligence resides on the hardware rather than in the cloud. By successfully shrinking large models, Apple could gain a competitive advantage in privacy and latency, effectively removing the cloud bottleneck that currently limits the real-time capabilities of mobile virtual assistants.
