Apple finally realized that relying on Google for artificial intelligence creates massive liabilities. The tech giant plans to mass-produce a dedicated AI server chip codenamed Baltra starting late twenty twenty-six. Ming-Chi Kuo claims this silicon focuses on inference. Cupertino currently rents brainpower from Gemini to keep Siri functional while engineers build proprietary hardware.
This strategy mirrors how they handled modem transitions with Qualcomm before switching to internal solutions. Kuo identifies distinct hurdles regarding on-device processing capabilities. Even though the iPhone 17 lineup crushed Samsung globally last year to take the sales crown, those numbers relied on standard upgrades rather than revolutionary machine learning features.
Meaningful on-device intelligence likely arrives around twenty twenty-seven once these custom processors hit the servers. Controlling the full stack eliminates bottlenecks and improves efficiency dramatically. Previous benchmarks showed the M3 Ultra sipping way less juice than x86 rivals during heavy workloads. This shift brings core technology control back home, where the company prefers keeping everything locked down.
This strategy mirrors how they handled modem transitions with Qualcomm before switching to internal solutions. Kuo identifies distinct hurdles regarding on-device processing capabilities. Even though the iPhone 17 lineup crushed Samsung globally last year to take the sales crown, those numbers relied on standard upgrades rather than revolutionary machine learning features.
Meaningful on-device intelligence likely arrives around twenty twenty-seven once these custom processors hit the servers. Controlling the full stack eliminates bottlenecks and improves efficiency dramatically. Previous benchmarks showed the M3 Ultra sipping way less juice than x86 rivals during heavy workloads. This shift brings core technology control back home, where the company prefers keeping everything locked down.
