Skip to main content

Apple in Talks with Startup to Squeeze Big AI Models into iPhones

Apple might be cooking up something that could finally make Siri smart enough to hold a real conversation. According to reports, the tech giant is in early talks with a Silicon Valley startup called PrismML to develop a way to compress large AI models so they can run natively on iPhones. Think of it as putting a supercomputer's brain into a phone-sized package.

Why does this matter?

Right now, most advanced AI tasks—like the ones powering ChatGPT or Google's Gemini—happen in the cloud. Your phone sends your request to a distant server, which processes it and sends the answer back. That works, but it has downsides: it's slower, requires an internet connection, and raises privacy concerns. Apple has long championed on-device processing for privacy, but its own Siri has lagged behind competitors. This compression tech could change that.

PrismML's CEO confirmed that Apple has started evaluating the technology, though the discussions are still in the early stages. The timing is interesting: the news broke just a day after Apple released the iOS 27 beta, which includes the first public test of a significantly revamped Siri. It seems Apple is finally serious about catching up in the AI assistant race.

How does the compression work?

PrismML's approach is pretty clever. They take a massive AI model—like Alibaba's open-source Qwen, which originally took up about 54GB—and shrink it down to less than 4GB. That's a 93% reduction in size, allowing the model's 27 billion parameters to run on an iPhone 15 or newer. The company has already released a compressed version of Qwen publicly, and they plan to tackle other large open-source models next.

On the technical side, their solution can reduce a cloud model that once required eight GPUs to just one. It also allows models that typically rely on data centers to migrate to smartphones and laptops. But here's the catch: while it cuts the memory and computing power needed for a single AI task, it doesn't mean overall chip demand will drop. You still need powerful hardware to run these models locally.

What does this mean for users?

If Apple successfully integrates this technology, you could see Siri become much faster and more capable—without needing to be online. Complex tasks could still be handled in the cloud, but sensitive information (like your personal messages or health data) would stay on your device. That's a big win for privacy.

Analysts like Pathak from Counterpoint believe the combination of cloud and on-device AI will provide a more comprehensive, efficient, and privacy-focused experience. "Complex tasks are handled in the cloud, while sensitive information is processed on the device," he said.

The bigger picture

PrismML was initially incubated at Caltech and received backing from Khosla Ventures. Their technology isn't just about making Siri smarter—it's about enabling a new class of AI applications that can run entirely on your phone. Imagine real-time language translation, advanced photo editing, or even AI-powered health diagnostics, all without sending your data to a server.

Of course, the talks are still early, and there's no guarantee a deal will happen. But the fact that Apple is exploring this path shows they're serious about bringing powerful AI to their devices in a way that respects user privacy. For now, we'll have to wait and see if this partnership bears fruit.

Key Points:

  • Apple is in early talks with PrismML to compress large AI models for on-device use.
  • The technology could enhance privacy and give Siri a major upgrade.
  • PrismML has already shrunk a 54GB model to under 4GB, compatible with iPhone 15 and newer.
  • On-device AI reduces latency, lowers cloud costs, and works offline.
  • The deal is not finalized, but the potential is huge.