Apple hunting chip startups to fix faltering AI server plan

On Wednesday, The Information reported that Apple is approaching chip startups to gauge interest in acquisition and has consulted with bankers about possible deals, according to people familiar with the matter. The move reflects Apple's effort to bolster its capability to build server processors for running AI workloads.

The reported acquisitions strategy stems from ongoing technical challenges. Apple's current in-house AI servers run on internally designed M2 Ultra chips, but these have faced performance shortcomings. The company had originally planned to ship an updated version of its internal AI server chip, called "Baltra," in 2024, but the project has been pushed back. Earlier this year, when Apple attempted to run Google's Gemini models on its internal servers as part of a Siri overhaul, the Mac-based chips proved unable to handle the large model, forcing Apple to run portions of the revamped assistant on Nvidia chips housed in Google's cloud infrastructure instead.

Apple has historically avoided large acquisitions. Its most recent major deal was in January, when it acquired Q.ai, an Israeli company focused on AI technology for audio. The company reported $45.57 billion（約7.3兆円） in cash and cash equivalents as of March 28, the end of its second quarter. Last week, Apple announced plans to spend over $30 billion（約4.8兆円） under a multi-year chip supply deal with Broadcom, a move aimed at bolstering its domestic sourcing. Apple did not immediately respond to a request for comment, and Reuters could not independently verify The Information's report.

Apple's reported hunt for chip acquisitions exposes a crack in the company's vertical-integration strategy. For years, Apple has built custom silicon for its devices with considerable success, but the demands of running large language models (AI systems that understand and generate text) at scale on its own servers have proved harder than designing mobile and laptop chips. The company's reliance on Nvidia hardware running in Google's cloud — announced as part of a Siri overhaul — is a telling admission that internal M2 Ultra chips lack the architecture or performance headroom to handle contemporary AI workloads.

The delayed "Baltra" project suggests the problem is not a simple engineering fix; it points to structural challenges in Apple's server-chip roadmap. By shopping for external chip startups, Apple appears to be hedging against a longer-than-expected development cycle while maintaining its public commitment to on-device and on-server AI processing. The company's substantial cash position ($45.57 billion（約7.3兆円） as of March 28) and recent commitment to spend over $30 billion（約4.8兆円） on a multi-year Broadcom chip supply deal indicate that Apple is willing to invest heavily in securing chip supply — whether through acquisition or partnership — to reduce future reliance on competitors' hardware.