Apple's AI situation is more awkward than it appears from the outside. The company built its reputation on controlling every part of the experience, from the chip inside your phone to the software running on it. But for its most ambitious AI features, it is currently renting computing power from Google, which runs Nvidia hardware. Every time the upgraded Siri handles a complex request, it goes through infrastructure Apple does not own.
The root cause is that Apple's own server chips, designed primarily for consumer devices like MacBooks, were not built for the kind of heavy, parallel computing that modern AI models require. When Apple tried running advanced AI models on its own servers, the hardware could not keep up. The chips are good at many things. Running large AI models at scale is not one of them.
Apple had a plan to fix this. An internal project, codenamed Baltra, was supposed to deliver a purpose-built server chip this year. That timeline has slipped. A more capable server chip based on a future design is not expected until 2029. That is a long time to keep renting from a competitor.
So Apple is now doing something it almost never does: looking to buy its way to a solution. According to The Information, the company has approached chip startups and spoken with investment bankers about potential acquisitions over the past several months. The goal is to bring in specialized talent and technology that can accelerate the server chip work.
This is a genuine shift. Apple has historically avoided large acquisitions. Its biggest deal ever was $3 billion for Beats, over a decade ago. Earlier this year it paid nearly $2 billion for an Israeli AI startup, which immediately became its second-largest acquisition. Now it is signaling openness to even larger deals. Apple's CFO told analysts the company is no longer targeting a cash-neutral position, which frees up the $45.6 billion in cash it had on hand as of March.
The timing matters for another reason. Apple's incoming CEO, John Ternus, takes over on September 1. Ternus spent his career in hardware engineering, overseeing the iPhone, iPad, and Mac chip programs. His background suggests the push toward owning more of the AI hardware stack will only intensify once he is in charge.
For anyone who uses Apple products in their business, this is worth watching for a simple reason: whether Apple fixes its AI infrastructure or not directly affects how capable Apple's AI features become over the next two to three years. A Siri that runs on Apple's own hardware, rather than rented cloud servers, would also be faster, cheaper for Apple to operate, and potentially more private. Apple's whole pitch on AI has been that your data stays protected. That promise is easier to keep when you own the hardware.
The $30 billion chip supply deal Apple signed last week with Broadcom covers different chips, mostly wireless and connectivity components, and does not solve the AI server problem. That deal is partly about US manufacturing politics. The server chip gap requires a different fix, and acquisitions appear to be the path Apple is now seriously considering.
Chip startups in the AI server space are expensive right now. Any company Apple buys in this area will command a premium. But given what is at stake for the long-term health of Apple's AI offering, the cost of not solving it is higher.