Apple AI chip acquisitions: evaluating a potential pivot in server hardware strategy

Apple is reportedly exploring acquisitions of semiconductor startups to accelerate its AI server chip development, according to The Information. The move comes as its next-generation server chip, Baltra, has slipped past its planned 2026 debut, and the company currently relies on Nvidia GPUs in Google Cloud for demanding AI workloads like the Gemini-powered Siri.

What happened

The Information reported that Apple has been in talks with bankers and has approached semiconductor startups to gauge interest in selling themselves, aiming to boost its efforts to build server chips for running AI. Apple's next-generation server chip, code-named Baltra, has reportedly slipped past its planned 2026 debut. In the meantime, Apple uses M2 Ultra-based systems for some AI processing, while more demanding tasks are handled by the Gemini-based model powering the new Siri, which runs on Nvidia GPUs in Google Cloud.

Apple has a history of selective larger acquisitions, including Beats for $3 billion in 2014, most of Intel's smartphone modem business for $1 billion in 2019, and Israeli AI startup Q.ai for nearly $2 billion earlier this year. The Information also mentioned that Apple held talks with PrismML, whose technology can shrink large AI models to run directly on iPhones, though no deal was reported.

CFO Kevan Parekh recently signaled a shift away from the company's net cash-neutral goal, which could free up capital for larger acquisitions. Additionally, leadership changes are on the horizon, with hardware chief John Ternus set to become CEO in September and chip executive Johny Srouji taking expanded responsibility over hardware engineering and semiconductors.

Why AI builders should care

If Apple accelerates external chip acquisitions, it could alter the competitive landscape for AI server hardware and potentially affect silicon supply and pricing dynamics. A shift toward larger acquisitions could expand Apple's in-house chip capabilities beyond the Baltra and M5 Ultra roadmap, impacting timelines for AI server deployment. PrismML's model-shrinking approach, if pursued, could influence on-device AI efficiency strategies, including potential implications for model sizes and inference speed. Any move away from a traditional net cash-neutral policy could enable bigger bets, potentially changing capital allocation in the AI hardware ecosystem.

Practical implications

For AI builders, potential Apple acquisitions could shift supplier dynamics for server chips and accelerators. If Apple pursues Baltra-like or M-series server chips in larger package deals, there could be downstream effects on chip supply, R&D collaboration, and pricing pressures. Apple is currently developing a server chip based on the M5 Ultra, with plans for an M7 Ultra capable of up to 1.5 TB of memory, though a server chip based on M7 Ultra is unlikely to be ready before 2029. The company also extended its collaboration with Broadcom on an AI server chip through 2031. Any partnership with Broadcom or other suppliers tied to AI server chips could gain or lose momentum depending on acquisition outcomes.

Caveats

These claims rely on The Information report and related coverage; actual deals, terms, and timelines are uncertain and could change. Details about Baltra's schedule, Gemini workloads, and the specifics of PrismML discussions are not confirmed as deals and may be speculative. The allowed sources focus on journalism coverage; future outcomes depend on strategic decisions by Apple and market conditions.