During the AIGC trend, Apple has been keeping a low profile, and now multiple sources indicate that the technology giant intends to accelerate progress.
According to Guo Mingyi's latest survey, Apple will purchase 2000-3000 units in 2023, accounting for 1.3% of global AI server shipments; In 2024, 18000-2000 AI servers will be purchased, accounting for 5%.
Guo Mingyi believes that Apple's main AI servers purchased are the Nvidia HGX H100 8-GPU, which is the most common and used for training and inference generative AI. The purchased chips in the fourth quarter of next year will be added to B100. Calculated at a price of 250000 US dollars per H100, it is estimated that Apple's procurement cost on AI servers in 2023 will be 620 million US dollars, and it is expected to reach 4.75 billion US dollars next year.
Guo Mingyi also stated that Apple needs to invest at least billions of dollars annually to have a chance to catch up with competitors - which coincides with Bloomberg's report, but the prediction of investment amount is more aggressive. According to Bloomberg on October 22nd local time, Apple is expected to spend $1 billion annually on developing its generative AI products.
Bloomberg sources say that the $1 billion annual investment plan comes from three senior executives responsible for Apple's AI business - Senior Vice Presidents for AI and Software Engineering, John Giannandrea and Craig Federighi, as well as Senior Vice President for Apple Services, Eddy Cue.
Technology giants vying to hoard hardware computing infrastructure is expected to usher in the golden age
According to the news, Apple built a large language model Ajax in July and launched an internal chat robot called "Apple GPT" to test its functionality.
The latest Bloomberg report indicates that its internal team is urgently testing the chat robot Apple GPT and plans to incorporate new generative AI capabilities into the "family bucket". It is said that a more intelligent Siri will be available as soon as next year. This is equivalent to the current capabilities of ChatGPT and Bing.
Jeff Pu, an analyst at Haitong International Securities, believes that Apple plans to integrate generative AI technology into the iPhone and iPad as early as the end of 2024.
To achieve the above goals, Apple indeed needs to accelerate its efforts and stockpile more AI hardware in the next two years.
Guo Mingxuan stated that he believes that when training its AI big model, Apple should prefer to purchase and build AI servers on its own based on security, privacy, and design flexibility considerations, rather than using virtual hosting services provided by other cloud service providers.
At present, Apple lags behind competitors such as Microsoft and Google in terms of both product and computing power reserves.
According to Guo Mingyi's article, starting from the second quarter of 2023, Nvidia AI chips have experienced a shortage, and coupled with Apple placing orders later than other major customers, Apple's AI server procurement volume in 2023 is significantly lower than in 2024. Even if the supply of Nvidia AI chips improves in 2024, Apple's AI server procurement still lags behind its competitors.
Guo Mingyi stated that Meta's AI server procurement volume in 2024 is approximately 40000 units. Morgan Stanley's report in July pointed out that due to an increase in orders from major cloud service providers in the United States, especially Microsoft, and new AI demand from Saudi Arabia, the shipment volume of AI servers this year has been increased, with an annual growth rate of 70%.
The high spending of technology giants such as Apple on AI servers fully demonstrates the strong demand for intelligent computing power in the industry.
Song Hui from Huaxi Securities stated that the growth of computing power in the AI era far exceeds the rate at which Moore's Law doubles every 18 months, and the arrival of the singularity of AI will also achieve a golden era in the field of computing infrastructure. The industry benefits not only from AI servers, server OEM, and server power, but also from liquid cooling solutions, optical modules, lasers, AI computing centers, and more.
Dongguan Securities Chen Weiguang stated that in the context of independent and controllable computing power, domestic computing power such as Huawei, Haiguang, and Cambrian are expected to further rise and deeply participate in the construction of domestic computing power projects. Among them, Huawei's "Kunpeng+Shengteng" computing power system is at the forefront of domestic computing power in terms of technical level, ecological construction, and application implementation. It is recommended to focus on Huawei's "Kunpeng+Shengteng" industrial chain.