Apple’s on-device ambitions just got more interesting. The company has reportedly been talking to a startup that claims to run a massive, server-grade language model on the iPhone 17 Pro — no cloud required.
If it holds up, it would be a big deal. Here’s why.
The startup fueling Apple’s on-device AI ambitions
Most advanced AI features today rely on powerful models running in remote data centers. That means your requests, documents or conversations often have to leave your device before an answer can be generated.
Apple Intelligence reduces this as much as possible with on-device models, but some of the smartest Apple Intelligence features still offload their heaviest tasks to Private Cloud Compute. This not only costs money, but also carries a small privacy tradeoff. On-device AI that can match that muscle would enable Siri AI to get sharper without ever leaving your pocket.
But there might be a fix. Apple is reportedly talking to PrismML, who claims it pulled off something Apple engineers haven’t been able to, according to a new report by The Information.
PrismML reportedly squeezed Alibaba’s open-source Qwen 3.6, a 27-billion-parameter model that normally takes up 54GB — and compressed it down to 4GB without compromising on how smart it is.
One thing to note is that it’s PrismML’s own claim, not something that’s been independently verified.
How this stacks up against Apple’s own AI
Apple’s latest on-device model — AFM 3 Core Advanced — already powers the expressive Siri AI voices introduced in iOS 27. And while it’s a 20-billion-parameter model, it only uses 1 to 4 billion parameters to stay lightweight.
But PrismML says its compressed Qwen model skips that shortcut entirely, meaning all 27 billion parameters are active at once. That’s a big part of why it can handle complex reasoning, extended chats and even autonomous coding tasks. It is exactly the kind of heavy lifting usually reserved for cloud-based features.
Why would Apple want in?
Apple is reportedly currently in early talks with PrismML about how it could use the compressed Qwen model. Still, nothing suggests a deal is close, or even on the table.
It won’t be the first time Apple has shopped in this space. The company paid around $2 billion to acquire AI startup Q.ai. Pair that with Apple using Google’s infrastructure to help power the new Siri AI coming with iOS 27, and a pattern emerges. Apple knows its in-house AI still needs outside help to compete, and it is willing to pay for it.
What it means for you
None of this changes anything about the iPhone in your pocket today, and it may never see the light of day. But if the claim holds up, it could signal where Apple wants Siri AI to go next — less reliance on servers, more capable intelligence features running on the silicon you already own.
If this shows up before next year’s WWDC keynote is anyone’s guess. But if Apple really is shopping for a smarter, cloud-free AI, don’t be surprised if PrismML’s name appears before iOS 28 rumors start swirling.
Anurag Chawake is a tech-focused writer specializing in smartphones, apps and consumer technology. His interest in computers began during the Windows 98 era, eventually leading him to explore everything from operating systems to mobile devices and PC hardware. Anurag previously contributed to The Indian Express, covering Apple, Android, gaming and the broader technology landscape.
One response to “Apple talking to a startup that could cram GPT-class AI into your iPhone”
If it works, I sure hope Apple would buy that company and immediately hire all those employeesl