Inquiring minds want to know how all of this business confidential information winds up behind paywalls. Which stool pidgeon is doing the squealing and how much are they getting paid?
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
NYT: Apple's AI Struggles Began with 2023 Chip Budget Dispute
- Thread starter MacRumors
- Start date
- Sort by reaction score
Yet another case of poor leadership from the top. If Tim approved it (CEO) and Luca cut it (CFO), the next time you're meeting with Tim you tell him schedules have slipped by 6 months because your budget was cut. He either accepts it or overrules the CFO.
Then splitting the development between groups is stupid and guaranteed to fail. Does the Board ever review these failures?
Then splitting the development between groups is stupid and guaranteed to fail. Does the Board ever review these failures?
Several caveats.
1️⃣ The CFO may have been persuasive, so it's not a matter of his controlling Cook but making a better case.
2️⃣ This is a report based on sources from within a struggling effort, each player with an axe to grind, so we don’t know what really happened.
3️⃣ According to the article, the AI people were still given a *huge* 50% *increase* in their budget, so it's not as if they were cut back. Groups often ask for more than they need.
4️⃣ We also don't know if the assertion that Cook approved doubling the budget before the CFO chimed in is even true!
5️⃣ I read the NYT article, but I don't recall the reporters giving actual dollar amounts. How much money was involved and what fraction of Apple's overall R & D budget was it? What portion of Apple's overall operating budget was it?
6️⃣ It's fun to speculate about what's going on, and some even enjoy condemning particular people, arguing they should be fired, but let's admit all this is based more broadly on narrative spins than proven facts.
7️⃣ Nevertheless, the fact this was a major article in the New York Times will give institutional investors and board members much to chew on — and likely gives some support to the hyperbolic claims that “Cook is cooked”!
I asked Siri when my iPhone was last backed up. It opened the health constipation graph that I didn’t even know existed.
Apple has roughly $54 billion in cash or cash equivalents. Investors have always wondered what they would do with that extra cash.
Seems like they should've been investing in AI to a large degree. Just like all other major tech companies were doing.
How can one of FAANG, the mag7, or whatever it is now, miss this much? 5 year old GPUs when everyone else is gobbling up all the H100s/B200s they can?
"Cook initially approved doubling the team's chip budget, but CFO Luca Maestri reportedly reduced the increase to less than half that amount, and instead encouraged the team to make existing chips more efficient.The lack of adequate GPU resources meant Apple's AI team had to negotiate for computing power from providers like Google and Amazon.At the time, Apple's data centers had about 50,000 GPUs that were more than five years old – far fewer than the hundreds of thousands of chips being purchased by competitors like Microsoft, Google, and Meta."
I wonder if their spat with Nvidia ended up hurting them here. If ACDC doesn't scale till late 2026, the full LLM Siri possibly moving to iOS20 could make sense. And what happens if ACDC is still far behind Nvidia?
"Cook initially approved doubling the team's chip budget, but CFO Luca Maestri reportedly reduced the increase to less than half that amount, and instead encouraged the team to make existing chips more efficient.The lack of adequate GPU resources meant Apple's AI team had to negotiate for computing power from providers like Google and Amazon.At the time, Apple's data centers had about 50,000 GPUs that were more than five years old – far fewer than the hundreds of thousands of chips being purchased by competitors like Microsoft, Google, and Meta."
I wonder if their spat with Nvidia ended up hurting them here. If ACDC doesn't scale till late 2026, the full LLM Siri possibly moving to iOS20 could make sense. And what happens if ACDC is still far behind Nvidia?
Nope, they put him in charge of Business Services... Reporting directly to Tim.
and what the CFO should have done is just told the CEO most of the computers should be 1.1-1.3 X and make the sign off at 1.5X.
Sounds like the dysfunction where I work. All the execs are telling us to incorporate more ai yet our cybersecurity and privacy teams reject it. We are supposed to be getting ai capable pcs so we can run models locally but so far no one I know in data science has yet. Cloud compute charges they scrutinize. Approvals take months to a year so by the time it is approved the next great thing is out and you are constantly behind the curve. This is where I hate working in enterprises and prefer startups.
The keypost is poorly written, misusing the word "chip" and not understanding what "wrong" can mean for LLMs and similar tools.
Hallucinations are inherent with the entire paradigm that are LLMs.
So, and tool such as Siri that will depend on other tools like LLMs is always going to spit out the occasional "wrong" answer.
As for "chip" continually being repeated: that's because these things are magic to writers in media. The high priced gadgets from Nvidia are so performant because of the memory architecture. There's not magical instruction in whatever piece of silicon Nvidia has its supplier (usually TSMC) produce. And there's little "G" in "GPU" that is important in these devices. It's all about array mathematics.
Siri has never been of use to me. I think most people also look at Siri that way.
Hence my own shrugging of my shoulders when people seem to get bent out of shape about it.
Hallucinations are inherent with the entire paradigm that are LLMs.
So, and tool such as Siri that will depend on other tools like LLMs is always going to spit out the occasional "wrong" answer.
As for "chip" continually being repeated: that's because these things are magic to writers in media. The high priced gadgets from Nvidia are so performant because of the memory architecture. There's not magical instruction in whatever piece of silicon Nvidia has its supplier (usually TSMC) produce. And there's little "G" in "GPU" that is important in these devices. It's all about array mathematics.
Siri has never been of use to me. I think most people also look at Siri that way.
Hence my own shrugging of my shoulders when people seem to get bent out of shape about it.
While I fully agree with you on that suggestion I'm sorry but this article
Screams pure ********!
Since 2021 Apple has been pushing and presenting Apple Silicon M series and A series chips as having built in NPU's which is EXACTLY where testing devices connected to a testing g board ans USB-C interface directly to a PC/Mac is done!
If you've worked in such an industry and had access to see this then you KNOW!
Blaming a damb chip budget onto an exiting executive Luca who's to retire this year or has is DISPICABLE at the very least!
Lucas is in charge of hardware!
Both Luca and this half twit looser that was made an executive in 6 months for DOING ABSILUTELY NOTHING answer directly to Tim Cook!
Anyone that REALLY believes this crap story that a sub executive would go behind the CEO's back at Apple or not do what the CEO asked or approve without reprocautions
Need I remind you of Apple Maps and Forstall?!
Yeah that's why this story is both bull and dispicable through Luca under John's lame excuse to save face for failing everything his jobs entailed at Apple!!!
Wow, huge red flag when the CFO is empowered to make major technical decisions that could affect the company's core products, and brand value, for decades. That has to be on Cook. Even though he initially approved the increase, the fact that he let his decision get overridden, and sided with the finance guy over the technical team is ominous, and would frankly be demoralizing as an employee. This from the company that, as far as I know, has more free cash than any other in the world. At some point hoarding all that cash is supposed to pay off, and this seems like exactly the kind of generational moment that demands some of it be used.
I don't like people reflexively criticizing Tim Cook for every screwup. However, if Cook let the CFO slash the budget, then it is on Cook's head.
Five-year-old GPUs in 2023 would be a Tesla V100... about 14 TFlops. A 2023 H100 Tensor Core would be about 60 TFlops. Apple bought back $84 billion in stock in 2023. To update those 50,000 GPUs to H100s (at retail prices) would have cost a fraction (less than 2% or $1.5 billion) of what they spent buying their own stock. Such shortsightedness is stunning.
Luca Maestri should possibly be let go. Apple has dropped the ball on Siri since version 1. Lots of us who like Apple have been telling them this for years. The current Siri is worse to use than Google Assistant was at its first release, and Google Assistant is now a dead product. Apple has lacked innovation in a software for years at this point. Do we blame this guy?
The Siri/Apple Intelligence debacle is as bad or worse than the Apple Maps situation was.
The Siri/Apple Intelligence debacle is as bad or worse than the Apple Maps situation was.
Are you sure you're not looking at the use of LLMs only as chatbots? Apple has decided to use an Apple-built LLM to run much of Siri too, without fully accounting for the fact that LLMs are by their very nature not accurate or capable enough for that task. Siri is expected to be reliable, but Apple has made achieving this even more difficult by trying to graft an LLM into it, without taking enough care to deal with the fact that LLMs are inherently "creative", and so they're often fuzzy and prone to error. Add to that Apple’s privacy-first approach, which makes cloud-based LLM implementation for Siri harder, while implementing on-device LLMs with current hardware limitations is also difficult.
Last edited:
Wow 😮 bog oops 🙊 silly to be limiting AI investment and cut off development - big huge problem there…
That is often easier said than one; and knowing the CEO and CF and long standing relationship I suspect the CEO knew what was going on. In addition, keeping lower limits meant people who didn't need more power didn't overbuy. In the end, it worked for us.
In the end it’s Tim Cook’s fault.
It was obvious to most people in tech or who follow tech by early 23 that ChatGPT was game changing and that he seems to have not seen this, is v concerning.