Apple delays 'Baltra' AI server chip amid infrastructure struggles, turns to Nvidia

Apple's in-house AI server processor, code-named 'Baltra,' has been delayed as the company struggles to build adequate infrastructure for its generative AI and Private Cloud Compute services. Originally targeted to ship this year, the custom silicon project slipped after engineers found that existing servers using Mac-based M2 Ultra chips could not handle the compute demands of Google's Gemini models for a revamped Siri. Apple reportedly turned to Nvidia GPUs hosted on Google's cloud infrastructure for the heavy lifting—a significant strategic compromise for a company that prides itself on independence and privacy through local processing.

The delay has prompted Apple to break its conservative M&A playbook, approaching semiconductor startups and consulting investment bankers for potential buyouts. CFO Kevan Parekh signaled a move away from 'net cash neutral,' freeing up Apple's $45.6B cash pile for larger acquisitions. Already, Apple completed the nearly $2B acquisition of Israeli AI startup Q.ai. The underlying issue: Apple's chip design team excels at efficiency-per-watt for mobile devices, but AI servers need brute-force thermal design, massive memory bandwidth, and complex interconnects—a radically different engineering challenge. For consumers, the Baltra delay means slower improvements to Siri and on-device AI features like Private Cloud Compute.