William Gallagher
Apple's artificial intelligence work has now extended to it needing a series of servers for testing, and Foxconn is its exclusive supplier.
Tim Cook has said that Apple has been working on AI for years, and it's Machine Learning tools also pervade everything it does, but it hasn't got a ChatGPT-like app as rivals have. Or at least, it doesn't have one outside of internal use.
Now according to the South China Morning Post, Apple has ordered an unknown number of servers that are specifically for the training and testing of AI services. The servers are to be made by a division of Apple supplier Foxconn, called Foxconn Industrial Internet, and built in Vietnam.
Foxconn is reportedly already Apple's biggest supplier of servers for its data centers. The South China Morning Post says that Foxconn accounts for approximately 43% of the server market globally, not just within Apple.
Neither Apple nor Foxconn have commented. However, it is believed that this division of Foxconn already supplies servers to ChatGPT OpenAI, as well as Nvidia and Amazon Web Services.
News of the order follows recent conflicting reports from analysts about Apple's plans to compete against AI chatbots. Bloomberg has said Apple intends a major launch in 2024, but Ming-Chi Kuo says that's unfeasible.
Charles Martin
Malcolm Owen
Christine McKee
Wesley Hilliard
Andrew Orr
15 Comments
If true, this is long overdue and if the servers are Apple designed and specifically for training it will be very much a step in the right direction.
Although, more than servers in the traditional sense I would expect them to be cluster boxes housing thousands of accelerator cards. Then all the interconnect technology, software frameworks etc.
Amazing what can happen with no budget constraints :)
Seems kind of odd to see this headline just a couple months after ChatGPT et al put AI into the spotlight. What are the odds of Apple just happen to need some AI servers at this time - especially, if they've been working on AI for years, as Tim Cook suggested. To someone more cynical than I, it might seem as if Apple is trying to show in any public way they can that they're not behind on AI.
I hate to beat a dead horse to death, but if Apple had developed some worthwhile generative AI over the years, would Siri be in such a sad state? I mean, it's been 12+ years since this digital assistant appeared on Apple's platform. Has anyone noticed any huge improvements in those 12 years? After a few years, I've pretty much been giving up on my hopes for it and been using it mainly to set an occasional timer or to check on the day's weather. In the last couple years, I've added asking Siri to turn lights on/off for me. Anything beyond that, is just too much of an exercise in frustration. Siri mishears or needs an Internet connection for the most mundane requests (e.g. requests for stuff that's on my friggin' phone!). I've never expected Siri to engage in serious two-way conversations, but even asking simple follow-up questions to a previous question have mostly been beyond its ken (I think recently some rudimentary capability was added there for some limited scenarios). If Apple has spent billions and years on AI, they sure haven't gotten their money's worth with Siri.
AI, of course, is in a lot more things than just conversational AI - Apple has applied it successfully to its object recognition in Photos, voice transcription, and probably other areas. But Apple's best known AI representative - Siri - remains as dumb as a stone.
So I wonder what the hardware specs are for those servers. Are these Intel/AMD CPUs with Nvidia GPUs running Linux? Or are they some kind of custom ASi-based servers running macOS?
It would be cool if they packed a bunch M2 Mac mini mainboards together in a box or some such. But I wonder how likely that is...