You would have to be wholly fatuous to leak the details of an ongoing negotiation with a company as private as Apple, and from the mouth of the CEO no less, as PrismML appears to have done recently, indicating that the tidbit about its impending acquisition by Apple was nothing more than a glorified PR campaign.
You don't leak details of an ongoing negotiation, especially when a company as sensitive as Apple is involved, if those talks have even a snowball's chance in hell of succeeding
While speaking with CNBC earlier this week, PrismML's CEO casually lobbed a bombshell: Apple was evaluating its technology, and that the discussions were still of a preliminary nature but that things were progressing "nicely." This came after The Information reported last week that negotiations between the two companies were continuing, while implying that an acquisition by Apple was also on the table.
For the benefit of those who might not be aware, PrismML specializes in shrinking huge AI models into smaller ones that are more suitable for edge cases such as mobile phones. Just last week, PrismML announced that it shrank the 27-billion-parameter Qwen 3.6 model, and then ran it on an iPhone 17 Pro.
This brings us to the core of today's topic. The fact that PrismML publicly disclosed its supposed talks with Apple only goes to nullify the probability of anything solid emerging from those discussions. After all, Apple is notoriously secretive about its internal deliberations, and remains loathed to their spilling out into the public arena prematurely.
Yet, there is an even more fundamental reason why these supposed talks are unlikely to succeed: Apple is already kind of doing what PrismML regards as its specialty.
As we detailed in a dedicated post, the all-new Siri AI and Apple Intelligence is powered by 3 cloud-based models, with the AFM 3 Cloud Pro reportedly distilled from a 1.2-trillion-parameter Google Gemini model, and two on-device models, called AFM 3 Core Advanced and AFM 3 Core, respectively.
While the AFM 3 Core has just 3 billion parameters, the AFM 3 Core Advanced takes the parameter count to 20 billion, but only activates between 1 and 4 billion parameters at a time, depending on the task at hand. Critically, it is this model that unlocks a revamped dictation function and more expressive Siri AI voices.
What's more, at a granular level, Apple's AFM 3 Core Advanced sports an innovative architecture, where model attention blocks - which look at the broader meaning within a given prompt - reside in the DRAM, while Feed-Forward Network (FFN) weights - which explore each prompt string or word individually to surface its specific meaning within the given context - reside in the NAND.
Even though Apple's AFM 3 Core Advanced model only loads between 1 and 4 billion parameters into the DRAM at any given time, it still requires a RAM size of at least 12GB.
Coming back, Apple is already distilling Google's Gemini models to develop bespoke versions suitable for its use case. And, the iPhone giant is reportedly paying Google as much as $1 billion per year to use its models in such a manner. Against this backdrop, we fail to see any rationale behind Apple going to the trouble of acquiring PrismML, especially as it has proven to be particularly frugal with the way it dispenses its cash hoard.
About the author: Writing is my one incontrovertible passion. Over the past six years, he has authored over 2,200 distinct articles on financial and tech-related topics, spanning nearly 1 million words. And he has been a member of Wcctech mobile team since 2025. As an alumnus of the University of Toronto, Rotman Commerce Program, I bring nuance, in-depth knowledge, and a unique perspective to every topic that I cover. When I'm not writing, I'm traveling the world, exploring hidden confectionaries and restaurants as an aspiring food connoisseur.
Follow Wccftech on Google to get more of our news coverage in your feeds.
