Apple's attempts to catch up with the rest of the market in the generative AI field could cost more than first thought, with it potentially spending more than $5 billion over two years just on servers.
On Sunday, it was reported Apple has committed to spending in the region of a billion dollars annually just on its AI efforts. In a Monday post, TF Securities analyst Ming-Chi Kuo proposes the amount could be much higher than that.
Writing on Medium, a survey by Kuo reveals Apple intends to purchase between 2,000 and 3,000 AI servers in 2023, and possibly between 18,000 and 20,000 servers in 2024. These figures are said to represent 1.3% and 5% of worldwide AI server shipments in their respective years.
Kuo thinks it is "reasonable to assume" that Apple's AI server choices will be the "most popular specification" of the year, using Nvidia's HGX H100 8-GPU. For late 2024, Apple may upgrade to a "B100 solution."
With a price of around $250,000 per HGX H100 8-GPU server, it's estimated that Apple will spend around $620 million in 2023, and $4.75 billion in 2024 on AI servers alone. This works out to be about $5.37 billion over two years.
The lower spending in 2023 than 2024 is down to a shortage of Nvidia AI chips, as well as Apple "placing its orders later than other major customers."
Despite seeming like a great expense, Kuo thinks the purchases are still far behind other companies. As examples, Meta will allegedly be buying around 40,000 AI server units in 2024, while Microsoft plans between 80,000 and 100,000 units in the same year.
This may not be an issue for Apple with its lower number of orders, as Kuo believes it's "not appropriate to compare" with Microsoft, since it is assumed Apple won't offer services like cloud hosting of AI projects.
Privacy, flexibility, and more expense
Kuo thinks that Apple prefers to train its large language models on AI servers it has purchased and installed itself rather than using virtual hosts in the cloud, citing security, privacy, and "design flexibility" behind the move.
Though Apple's AI infrastructure and computing power won't be as high as competitors who have already made a head start, Kuo thinks this means Apple thinks "it must have superior software development capabilities to catch up."
Kuo points out that the purchase of servers doesn't include other costs that Apple could incur, including labor and the cost of operating its infrastructure. It's therefore "reasonable to estimate" that Apple could be spending several billion dollars annually to catch up with its rivals.
There is also a mention to Sunday's report, with Kuo offering "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 accepts that Apple "can develop its own AI server chips to save on AI server procurement costs." While plausible, Kuo says "the visibility of the development is not clear at present."