Will Apple Pull a SpaceX? (hitesh.in)

Apple is developing a new inference chip, codenamed Baltra, in collaboration with Broadcom, utilizing TSMC's advanced 3nm process and set to launch in 2027. This move aligns Apple with other tech giants like Google and AWS who have similarly invested in custom silicon for running AI workloads. The significance of Baltra lies not only in its expected impact on Apple's internal AI infrastructure but also in the potential for Apple to enter the AI cloud computing market, a growing trend where companies with substantial computing power begin to monetize their idle capacity, as demonstrated by SpaceX renting out its Colossus cluster to Anthropic. Apple has three potential paths for its new capabilities. It can choose to keep the infrastructure internal, rent out its processing power like SpaceX, or develop a product that emphasizes privacy and efficiency, distinguishing itself from competitors. This last approach of offering "private by construction" inference services could cater to organizations that face legal restrictions on using public cloud services. With Apple recently crossing $100 billion in annual services revenue, leveraging its own silicon for a new, trust-based inference service could be not only technically feasible but a lucrative business opportunity as well, making it imperative for Apple to carefully consider how it will leverage this capability in the evolving AI landscape.