October 24, 2023 – Noted analyst Ming-Chi Kuo has released a market research brief predicting Apple’s AI server procurement for 2023 to be in the range of 2,000 to 3,000 units, accounting for a mere 1.3% of the global AI server shipments. Kuo’s estimates indicate that in 2024, Apple’s AI server purchases will surge to a range of 18,000 to 20,000 units, capturing a substantial 5% market share.
Kuo suggests that Apple’s AI server acquisitions predominantly consist of the widely-used Nvidia HGX H100 8-GPU servers, primarily designed for training and inferencing in generative AI. By the fourth quarter of the next year, Apple will also introduce the B100 chip into its procurement strategy.
Calculating at a cost of $250,000 per H100 unit, Apple’s estimated expenditure on AI servers in 2023 is projected to reach $620 million. In 2024, this figure is anticipated to soar to a staggering $4.75 billion.
Additional insights from Kuo’s research brief reveal that even with an expected alleviation in Nvidia AI chip supply in 2024, Apple’s AI server quantity remains notably lower than its competitors. For example, Meta is set to acquire approximately 40,000 AI servers in 2024, eclipsing Apple’s procurement. Moreover, Meta already outpaces Apple’s AI server holdings by a considerable margin before 2024.
Kuo hypothesizes that Apple, considering security, privacy, and design flexibility, is likely to favor in-house procurement and deployment of AI servers over relying on virtual hosting services provided by other cloud service providers, such as Microsoft, which is projected to purchase between 80,000 and 100,000 AI servers in 2024.
Apple’s AI infrastructure capacity currently lags behind key competitors, and to catch up, it must enhance its software development capabilities significantly. These calculations do not account for other expenses such as labor costs and infrastructure operational expenses. Consequently, it is reasonable to assume that Apple would need to invest several billion dollars annually to have a chance at catching up with its rivals. If Apple intends to allocate only $1 billion annually to the development of generative AI, there are valid concerns about the prospects of Apple’s generative AI business and services.
While some may argue that Apple could reduce its AI server procurement expenses by developing its own AI server chips, this remains an uncertain venture with limited visibility at present.