Apple is evaluating technology that could make future iPhones capable of running much larger AI models without relying as heavily on cloud servers.
CNBC reported the company is testing software from PrismML that makes a much larger AI system small enough to run on a high-end phone.
If the technology works at scale, it could eventually give iPhone owners faster, more useful AI without making every feature depend on a strong internet connection.
Apple begins testing the technology
PrismML CEO Babak Hassibi told CNBC that Apple is “really evaluating our technology right now” as it tests the models on its devices.
Testing focuses on speed, power use, and answer quality. Discussions remain preliminary, though Hassibi said: “things are progressing nicely.”
Apple declined to comment, and no partnership or licensing agreement has been announced.
A 54GB AI system made small enough for an iPhone
Most advanced AI systems need more memory than a phone can provide. PrismML says it reduced one from about 54GB to 3.9GB, allowing it to fit within the space available to an app on a high-end iPhone.
PrismML stores the system’s information in a simpler form, reducing the memory required to run it. Lower memory requirements could let a phone handle AI work that would normally require a much larger computer or a remote server.
According to the startup, its AI can understand text and images. Longer instructions can also be completed across several steps, including tasks that involve other tools.
Making an AI system this small can affect the quality of some answers, Hassibi acknowledged. More testing will be needed to show how reliably it performs during everyday use.
Stronger AI could make future iPhone models more useful
Apple already runs some Apple Intelligence features directly on iPhones while sending more demanding requests to its Private Cloud Compute infrastructure.
Running a larger share of that work on the device could let future devices handle longer requests involving text, images, and several connected steps. Responses could arrive faster because less information would need to travel to a server before an answer appears.
Local processing could also keep more personal information on the phone and allow some features to keep working when service is weak or unavailable. Fewer server requests could reduce Apple’s computing, energy, and data-transfer expenses as AI use grows across its devices.
Apple has not confirmed plans to use PrismML. Still, the test could influence how the company divides future AI work between iPhones and its data centers.
Also read: Apple is accusing OpenAI of using stolen confidential information to accelerate its push into consumer devices.