Daniel J. Arbess, the CEO of an investment firm that holds Apple stock, recently wrote a piece for the Wall Street Journal about Apple’s AI strategy (paywalled article).
The gist of it is that the major AI players have been investing enormous sums in AI infrastructure that are unlikely to be recouped (at least not anytime soon), while Apple has been investing much smaller amounts in hardware capable of running local AI models that are good enough for most purposes, preserving privacy, and only offering to use larger centralized models when a local model isn’t enough.
Basically, Apple is betting that the power of relatively inexpensive distributed computing models under the local control of end users will be much more important than enormously expensive, centralized big iron approaches. It’s not a new argument, but I thought it was well presented.
He also wrote a Substack article making substantially the same points:
Most of the big players are trying to invent a magical chatbot that knows everything, is always correct and can be so convincingly human as to pass the Turing test. This is certainly a fascinating field of research, but I believe that this goal is fundamentally impossible.
Apple, on the other hand, is not trying to create an actual “AI”, but is using neural network technology to produce software that does useful things - like photo editing or summarizing text documents. These are much smaller-scope activities, but in the final analysis, I think we’ll find them much more useful and less likely to cause damage when they don’t work as expected.
And this is why I don’t believe the doom-and-gloomers who are claiming that we’ll never again see affordable prices on DRAM and storage.
If companies are flushing 95% of their budget down the toilet, that bubble is going to burst. And when it does, all that capital expenditure will stop. Not just return to historic background levels, but stop, as the big players realize that there’s more profit in repurposing those trillions of dollars worth of servers for other purposes.
And then you and I will be able to get a personal computer with 128 GB of RAM and 8TB of SSD without breaking the bank.
I don’t know if it will happen this year, next year, or the year after that, but it will have to happen because no company is so big that it can hemorrhage money like this forever.