Apple's AI catch-up ends at WWDC and with the iPhone 16, says Kuo
Ming-Chi Kuo believes Apple will roll out big AI improvements during its WWDC keynote that will need the iPhone 16, and the bigger features may take some time fully roll out.

WWDC will bring many announcements, but all eyes are focused on AI news
Kuo, in an online question-and-answer post on Medium, summarized Apple's WWDC AI strategy as strong based on integrating helpful on-device AI into nearly every aspect of its operating systems and apps. However, he predicts that bigger benefits like server-side LLMs and an iPhone refresh cycle will only fully arrive over the course of the year.
Kuo expressed confidence that Apple's intensive efforts in integrating Apple Intelligence technologies into its apps and OS will showcase its "advantage" over competitors in user interface design. While Siri's improvements will be a key highlight, it's on-device LLM technology may be constrained by the 8GB RAM limitation in the next iPhone lineup.
Apple is expected to rely on both on-device and server-side LLMs to dramatically improve Siri responses to user queries.
Thinking long-term
Starting with the release of iOS 18, users will see noticeable improvements. That said, Kuo doesn't believe the company has yet achieved a breakthrough that will exceed market expectations.
Kuo seemed doubtful that the new iPhone hardware and operating system coming in Fall 2024 will trigger a user-replacement super-cycle. He does expect that the investments Apple has made in recent years to boost its machine learning and AI capabilities to pay off over time.
Kuo noted that his surveys indicate that iPhone shipments in the second half of 2024 could be slightly lower than for the same period in 2023. As additional features roll out in OS updates across 2024 and 2025, the picture for new iPhone sales could improve, he suggested.
He added that cloud-based LLM models typically take longer to train. This could mean that a full rollout of Apple's server-side AI solution will be phased in rather than fully implemented at launch.
Significant revenue and profit growth due to the forthcoming AI features may take a longer time to fully realize, Kuo said. He predicted the company will quickly gain "on par" status with its AI competitors, but that WWDC will dispel notions that the company has tried to quickly "catch up" to its industry peers.
In response to a question about collaborations with AI companies, Kuo downplayed the importance of using outside firms. He believes Apple will focus such deals mainly on "wrapper models" for interfacing with users and apps.
Apple is recently reported to have reached a deal with OpenAI to augment some on-device AI services.
Rumor Score: Likely
Read on AppleInsider
Comments
Whatever Apple does, I hope it removes these throughput issues.
https://www.tomshardware.com/news/chatgpt-nvidia-30000-gpus
The top GPUs are also from Nvidia, which Apple has been reluctant to partner with over the years:
https://www.newegg.com/p/1FT-0004-00814
30k of those would be $750m.
Apple is working on their own AI models with properly licensed content and processed on Apple Silicon. They can build custom Neural Engines for their server chips to be most cost-effective.
Having GPT as a fallback is good to have while they build out their own solution.
They have built an Apple Silicon server system already with Swift software, it's not clear which LLMs they will use on those but they can integrate custom models on those whenever they are ready and switch them transparently. They will use their MM1 models for some things:
https://applemagazine.com/apple-researchers-announce-breakthrough-in-ai-with-mm1-multimodal-learning/63593
https://techxplore.com/news/2024-03-apple-mm1-multimodal-llm-capable.html