Kuo: Apple to Spend $5 Billion on AI Servers in 2024 to Catch Up

Apple could spend upwards of $5 billion on servers alone in the next two years as it attempts to catch up with competitors in the generative AI field.

On Sunday, Bloomberg’s Mark Gurman reported that Apple will spend in the area of a billion dollars annually on its Artificial Intelligence efforts. However, in a Medium post on Monday, TF Securities tech industry analyst Ming-Chu Kuo says the amount Apple will spend could be much higher.

In his post, Kuo says a survey indicates that Apple has plans to purchase 2,000–3,000 units of AI servers in 2023 and 18,000–20,000 units in 2024. He says Apple will account for about 1.3% and 5% of global AI server shipments in 2023 and 2024, respectively.

Kuo says it’s “reasonable to assume” that Apple’s purchases of AI servers this year, are primarily the Nvidia’s HGX H100 8-GPU, which is designed for generative AI training and inference. Kuo expects to see Apple upgrade its AI server purchases in 4Q24 to a “B100 solution.”

HGX H100 8-GPU servers go for a price of around $250,000 apiece, meaning Apple will spend approximately $620 million in 2023, and $4.75 billion in 2024 on AI servers alone — or about $5.37 billion in the next two years.

Kuo says Apple would likely spend more during 2023 if it were not for the current shortage of Nvidia AI chips and Apple’s “placing its orders later than other major customers.”

While the amount Apple will likely spend sounds like a huge outlay, Kuo believes Apple’s expenditures will lag behind competitors, as Meta is expected to buy around 40,000 AI server units in 2024, and Microsoft has plans to purchase 80,000 to 100,000 units in the same calendar year.

Kuo says it’s “not appropriate to compare” Apple’s lower number of server purchases to Microsoft’s level of purchases since it is likely that Apple won’t offer services like cloud hosting of AI projects.

Kuo says he believes that when Apple trains its large language models, it prefers to use physical AI servers it has purchased and installed itself rather than virtual hosts from other cloud service providers. Kuo says that’s due to Apple’s desire for security, privacy, and design flexibility.

While Apple’s AI infrastructure might not provide as much computing power as its competitors, Kuo believes Apple thinks that it has “superior software development capabilities to catch up.”

In addition to the price of the servers, there are several other expenses that Apple could incur, such as labor and infrastructure operating costs. Kuo says it’s reasonable to estimate that Apple will spend billions on an annual basis to catch up to the head start its rivals currently enjoy.

Kuo also mentioned Sunday’s report, saying, “If Apple really intends to spend only a billion dollars a year on generative AI development, it won’t matter much if my survey is wrong, but I am genuinely concerned about the future of Apple’s generative AI business/service.”

Kuo also sees Apple eventually developing its own server chips to save on costs. Apple already makes its own chips for most of its devices, including those used in its iPhone, iPad, Apple Watch, and Mac lineups. However, Kuo said, “The visibility of the development is not clear at present.”