Siri
Apple’s makes an attempt to meet up with the remainder of the market within the generative AI subject might price greater than first thought, with it probably spending greater than $5 billion over two years simply on servers.
On Sunday, it was reported Apple has dedicated to spending within the area of a billion {dollars} yearly simply on its AI efforts. In a Monday submit, TF Securities analyst Ming-Chi Kuo proposes the quantity may very well be a lot larger than that.
Writing on Medium, a survey by Kuo reveals Apple intends to buy between 2,000 and three,000 AI servers in 2023, and probably between 18,000 and 20,000 servers in 2024. These figures are mentioned to characterize 1.3% and 5% of worldwide AI server shipments of their respective years.
Kuo thinks it’s “cheap to imagine” that Apple’s AI server selections would be the “hottest specification” of the yr, utilizing Nvidia’s HGX H100 8-GPU. For late 2024, Apple might improve to a “B100 resolution.”
With a value of round $250,000 per HGX H100 8-GPU server, it is estimated that Apple will spend round $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 decrease spending in 2023 than 2024 is all the way down to a scarcity of Nvidia AI chips, in addition to Apple “putting its orders later than different main clients.”
Regardless of seeming like an important expense, Kuo thinks the purchases are nonetheless far behind different firms. As examples, Meta will allegedly be shopping for round 40,000 AI server items in 2024, whereas Microsoft plans between 80,000 and 100,000 items in the identical yr.
This will not be a difficulty for Apple with its decrease variety of orders, as Kuo believes it is “not acceptable to check” with Microsoft, since it’s assumed Apple will not supply providers like cloud internet hosting of AI initiatives.
Privateness, flexibility, and extra expense
Kuo thinks that Apple prefers to coach its giant language fashions on AI servers it has bought and put in itself slightly than utilizing digital hosts within the cloud, citing safety, privateness, and “design flexibility” behind the transfer.
Although Apple’s AI infrastructure and computing energy will not be as excessive as opponents who’ve already made a head begin, Kuo thinks this implies Apple thinks “it should have superior software program growth capabilities to catch up.”
Kuo factors out that the acquisition of servers does not embrace different prices that Apple might incur, together with labor and the price of working its infrastructure. It is due to this fact “cheap to estimate” that Apple may very well be spending a number of billion {dollars} yearly to meet up with its rivals.
There may be additionally a point out to Sunday’s report, with Kuo providing “If Apple actually intends to spend solely a billion {dollars} a yr on generative AI growth, it will not matter a lot if my survey is mistaken, however I’m genuinely involved about the way forward for Apple’s generative AI enterprise/service.”
Kuo additionally accepts that Apple “can develop its personal AI server chips to avoid wasting on AI server procurement prices.” Whereas believable, Kuo says “the visibility of the event just isn’t clear at current.”
