In-House Chips Fall Short, Apple Turns to Acquisitions to Break Through AI Server Bottlenecks

Apple's success in consumer-grade chips has not extended to AI servers—a shortcoming that is now pushing the company toward mergers and acquisitions.

Apple is seeking to acquire chip companies to compensate for its deficiencies in AI server capabilities. According to a recent report by The Information, Apple has been negotiating with bankers regarding potential deals in recent months and has proactively reached out to multiple semiconductor startups to inquire about their willingness to sell.

The direct trigger for this move is that Apple's self-developed M2 Ultra server chip cannot run Google's Gemini large language model, forcing some features of the new Siri to rely on NVIDIA GPUs hosted on Google Cloud.

Apple and NVIDIA have historically had a tense relationship. Previous reports indicated that some Apple executives considered NVIDIA "difficult to work with." Now, having to rely on their competitor's chips puts Apple's management in a rather passive position.

M2 Ultra Falls Short, Baltra Delayed Again

Apple's AI servers currently run on the M2 Ultra chip. Originally designed for Macs, this chip struggles when facing ultra-large-scale models like Gemini.

According to insiders speaking to the media, Apple attempted earlier this year to deploy Google's Gemini model on its own server infrastructure to power the upgraded Siri, but "these chips simply could not handle such massive models." Ultimately, Apple was forced to transfer these workloads to NVIDIA chips on Google Cloud.

The root of the problem lies in the fact that Apple's chip team has long focused on energy efficiency optimization for mobile devices, whereas the core requirements for AI server chips are vastly different—high power consumption, high concurrency, and large memory bandwidth. These are two distinct technical paths.

Apple had originally pinned its hopes on the next-generation AI server chip, codenamed "Baltra," to solve this issue. The chip was scheduled for launch this year, but insiders reveal that the project has been delayed.

Catching Up by 2029: How to Fill the Immediate Gap

According to Bloomberg, Apple is currently advancing server upgrades based on the M5 Ultra as an interim solution. Greater hopes are placed on the more competitive M7 Ultra, which boasts a maximum memory capacity of 1.5TB—roughly double that of the M5 Ultra—with performance expected to rival NVIDIA's Blackwell chips.

However, the server version of the M7 Ultra will not be ready until 2029.

This means Apple faces a technological window of at least three years. During this period, the AI competition will not wait. Acquiring a startup that possesses server chip technology is the most direct way to fill this gap.

Apple is also exploring other avenues. The Information previously reported that Apple partnered with Broadcom in 2024 to develop AI server chips. Last week, Broadcom disclosed in securities filings that the two parties are extending their "long-term technical cooperation through 2031," though specific details of the collaboration were not provided.

M&A Doors Are Opening

Apple has historically been restrained regarding large-scale acquisitions. Its largest deal to date was the $3 billion acquisition of Beats Electronics in 2014.

However, this convention is showing signs of loosening this year.

In January, Apple acquired the Israeli startup Q.ai for nearly $2 billion—marking the second-largest acquisition in Apple's history. Q.ai's technology interprets speech content through facial micro-expressions.

Signals are also appearing at the financial policy level. During the second-quarter earnings call, Chief Financial Officer Kevan Parekh announced that Apple would abandon its long-standing "net cash neutral" policy, which maintained a balance between cash reserves and total debt. While Apple did not explain the reason for the policy adjustment, The Information noted that one possibility is using part of the cash for major acquisitions.

Apple's journey into self-developed chips itself began with an acquisition. In 2008, Apple bought the startup PA Semi for $278 million, laying the foundation for over a decade of in-house chip development.

Leadership Changes May Bring a Bolder M&A Style

Apple is currently in a CEO transition period. This September, hardware head John Ternus will succeed Tim Cook as CEO. Meanwhile, chip lead Johny Srouji will be promoted to oversee all hardware engineering, with semiconductor business remaining within his responsibilities.

The arrival of new leadership may bring a more aggressive approach to mergers and acquisitions.

Apple's day-to-day acquisition efforts are led by Steve Smith, who reports to Adrian Perica, Vice President of Corporate Development. In addition to server chips, Apple is also seeking to acquire AI companies that can help compress model sizes, making them better suited for on-device operation on iPhones, as previously reported by The Information.