In a move that could change how we use artificial intelligence on smartphones, news reports revealed that Apple is currently in negotiations with a startup called PrismML for a technology that can shrink large AI models to run directly on the iPhone without needing to connect to the cloud. The story began on Tuesday, July 14, 2026, when the startup announced it had compressed an AI model from 54GB to less than 4GB (15 times smaller).
What does this mean?
Large models like ChatGPT and Gemini require powerful servers to run because they are too big to fit on a mobile device. The current solution is to send your request to the cloud, where the model processes it and returns the response. This process requires an internet connection, takes time, and may expose sensitive data to external servers.
However, PrismML claims it has successfully compressed the model without significantly losing accuracy. The compressed model is called Qwen2.5-27B-Compressed, a compressed version of Alibaba’s popular Chinese model. The end result: a 4GB model that runs directly on the iPhone 15 and later, with all processing happening on-device.
Why is Apple interested in this technology?
Apple is looking for a solution to an old problem: privacy vs. power. Current Siri is limited because most of its processing happens on-device, while ChatGPT and Gemini are more powerful because they run their massive models in the cloud. If PrismML’s technology succeeds, Apple could make Siri as smart as ChatGPT—without sending your data outside your device.
A day before this announcement, Apple released the public beta of iOS 27, which includes a major Siri update. Reports suggest Apple is aiming to compete strongly with OpenAI and Anthropic in the smart assistant space.
How could this apply in Saudi Arabia?
Saudi Arabia is one of Apple’s key markets in the Middle East, with high iPhone popularity. If this technology becomes a reality, iPhone users in Saudi Arabia would benefit from:
- Smarter and faster Siri: Voice commands respond instantly without waiting for cloud processing
- Full privacy: Conversations with Siri never leave your device
- Arabic smart apps: Saudi developers could more easily integrate AI into their apps
- Reduced data usage: Less need to send large amounts of data externally, especially with video calls and browsing
Who is behind PrismML?
PrismML is a startup that emerged from the California Institute of Technology (Caltech), backed by the renowned investment firm Khosla Ventures. Its founder and CEO, Babak Hassibi, told CNBC that Apple and other companies are currently testing their technology, and he stated that negotiations are “going very well” despite being in early stages. Source: CNBC + The Information
When will we see this technology?
Negotiations are still in very early stages. However, experts predict that if the deal succeeds, we could see the first iPhone with this technology in 2027 alongside iOS 28 or a larger update. The important point here is that PrismML has already released its compressed model to the public, meaning any company or developer can use it today.
❓ Frequently Asked Questions
Who needs this technology?
Every iPhone user who cares about privacy and wants a fast, smart assistant. Saudi and Arab developers also benefit from it as it opens the door to powerful AI applications without the complexities of cloud computing.
Where is the technology used?
The technology runs directly on the iPhone device itself, without servers or cloud. This means it works anywhere, even in remote areas or on a plane without internet.
How much does on-device AI cost?
PrismML’s technology itself is free and open-source (the compressed model is publicly available). However, the real cost is the price of the iPhone itself for one of the newer models (iPhone 15 or later).
How long does it take to activate the technology?
If the deal goes through, the technology will be part of an iOS update. The average user only needs to download the update. PrismML managed to compress the model within months of research and development.
How do I use AI on the iPhone?
Currently, developers can test the compressed model from PrismML’s website. For regular users, you can wait for official updates from Apple or use third-party apps that integrate the model.
