Apple's earnings will be flat ahead of AI-enabled iPhone
Investment analyst firm TD Cowen is predicting that Apple's next earnings call will show little growth, but says forthcoming AI features could prove significant.
iPhone 15 Pro range
Analysts such as JP Morgan in March 2024, and Morgan Stanley in November 2023, are consistently predicting that AI will boost Apple's iPhone sales. JP Morgan, though, has predicted that this boost will come in 2025 with the iPhone 17 range.
Now in a note to investors seen by AppleInsider, TD Cowen repeats the expectation of an AI boost, but is less bullish about it than others, despite keeping its $220 Apple price target. TD Cowen set that price in late 2023, based in part on increasing sales in India, and it is again expecting to see shipment in that country increase.
Its analysts even argue that the declining iPhone demand in China may be at least somewhat countered by increased sales in what it describes as the "underpenetrated" India.
Overall, TD Cowen expects that global iPhone sales will be down approximately 3% year over year, it predicts a seasonal recovery around the launch of the iPhone 16 range. It also predicts that Mac sales will have dropped around 4% year over year, while Services will have grown around 11%.
TD Cowen expects that iPad sales will have stabilized, even ahead of the May 7, 2024 launch of the new models.
Its analysts, though, say that its supply chain sources back up claims that demand for the Apple Vision Pro is falling. So far the Apple Vision Pro is only available in the US, however, and demand will always be highest at the time of launch.
They also say that the Apple Watch has been affected by the US import ban and removal of the blood oxygen feature.
Read on AppleInsider
Comments
The famous investor Warren Buffett said the following about bitcoin - it produces nothing and is therefore worth nothing. Well, let’s see what AI can do or can’t do.
The brain has billions of neurons:
https://aiimpacts.org/brain-performance-in-flops/
100 billion neurons, 1000 connections, 200 calculations per second. The estimates there put this processing power around 10^16 FLOPs or 10,000 TFLOPs, some estimates are higher. Personal computers today are around 10-100 TFLOPs. 30 years ago, GPUs were well below 1 GFLOPs ( https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units ).
To train a model needs even more power:
https://www.tomshardware.com/tech-industry/artificial-intelligence/elon-musk-says-the-next-generation-grok-3-model-will-require-100000-nvidia-h100-gpus-to-train
Training is simulating what a human goes through from birth to adulthood:
https://www.ndtv.com/world-news/elon-musk-on-parenting-says-raising-a-kid-is-18-years-of-prompt-engineering-5459887
People often overestimate human intelligence. Babies know absolutely nothing and start out with a very powerful training engine. From birth they are bombarded 24/7 with patterns of association. Sounds matching shapes and colors, all at high fidelity audio and Retina quality visuals and it's clean training data that is curated by thousands of years of human development. It's also data weighted by importance, driven by motivations - hunger, curiosity, survival, affinity etc.
It's all just data patterns in the end though and all it needs is more power and accurate simulations. The right simulations depend on people coming up with algorithms that behave like how a brain works.
https://en.wikipedia.org/wiki/Transformer_(deep_learning_architecture)
When people do creative work, they are mixing patterns together. These patterns are accumulated through a lifetime of training.
Today, we have enough compute power in data centers to build powerful enough simulations to be useful.
https://www.sciencealert.com/worlds-first-human-brain-scale-supercomputer-will-go-online-in-2024
Realistic voices is one application:
https://elevenlabs.io/
These can be used to read books in any accent and any language. They can be used to do real-time two-way translation.
AI models will change how people search for information. Search engines are link services that rank information based on how popular they are and often they contain garbage links that are marketed well. AI language models are like a personal assistant with huge amounts of data and can generate conversations answering questions on any topic.
The image generators are being used by artists:
https://www.midjourney.com/showcase
https://www.instagram.com/midjourney.gallery/
Mainly for pre-viz, moodboards and concepts rather than final output but a time-saver compared to doing web searches for matching image references.
Software developers are using AI models to be more productive, they can write tedious code and help solve problems faster than digging through obscure, poorly written documentation:
https://medium.com/plandek/how-to-measure-the-impact-of-github-copilot-on-engineering-productivity-70698b41e4c
AI models won't be useful to everyone directly but they will be used in most industries eventually.