Apple’s Custom Server Chips Set for Mass Production in H2 2026, Facing Short-Term AI Development Challenges

Apple’s AI Future: The Baltra Chip

Apple is not just focusing on workstation-grade chipsets with its ongoing Apple Silicon transition. The company is reportedly developing a new chipset, codenamed Baltra, which will primarily handle AI inference tasks. An industry analyst has indicated that the mass production of these proprietary server chips is expected to begin in the latter half of 2026.

This news follows recent reports of Apple’s collaboration with Google to integrate the Gemini AI model into an updated version of Siri. The discussion around these in-house server systems has been ongoing for some time, with the same analyst outlining specific obstacles that Apple must overcome in its AI development strategy. As artificial intelligence becomes increasingly integral to both hardware and software, Apple finds itself in a precarious position, currently lagging behind competitors.

On-Device AI Rollout Expected by 2027

Much like its partnership with Qualcomm, which lasted until Apple developed a sufficiently advanced 5G solution for the iPhone 16e, the tech giant appears to be aligning with Google until it successfully launches its own Large Language Model. According to TF International Securities analyst Ming-Chi Kuo, Apple is encountering two significant short-term challenges in its pursuit of in-house AI technology.

Fulfillment of Apple’s previous pledges regarding Apple Intelligence and an enhanced Siri may not be adequate. There is an evident need for a more sophisticated on-device AI model. While the iPhone 17 has seen soaring sales that positioned Apple ahead of Samsung in market share for 2025, Kuo suggests that the immediate impact of on-device AI on shipment figures will likely be minimal.

A quick take on the Apple–Google AI partnership. I was on the MM Podcast about a month and a half ago, where we talked about the Apple–Google AI partnership. Here are the main takeaways from that part of the conversation (from ~38:50 to 45:20).I’ve lightly edited the text below… https://t.co/gHgHg8tMmc

— Ming-Chi Kuo (@mingchikuo) January 13, 2026

Despite the temporary respite provided by Google’s partnership, Kuo warns that this collaboration will only mitigate immediate pressures. The integration of AI is poised to become a pivotal element in hardware design, operating systems, and overall user experience. Looking ahead, Apple faces the long-term challenge of consolidating greater control over its core AI technologies. The initiative to create in-house server chips is promising, as history shows that Apple’s custom silicon offers substantial processing power and impressive memory bandwidth, both critical for efficient AI operations.

Perhaps most importantly, these in-house server chips can function at nearly half the power consumption compared to traditional x86 processors, as illustrated by the M3 Ultra, which uses 55% less energy when operating software like HandBrake. Kuo foresees that while mass production of these server chips will start in the second half of 2026, a meaningful rollout of on-device AI capabilities could begin to emerge significantly from 2027 onwards.

News Source: Ming-Chi Kuo

Source & Images