ArtOfWarfare
macrumors G3
This Wikipedia page lists a ton of companies who made RAM up until 2000-2020... why did all of the companies in this list stop? What happened to Intel and IBM each making memory, for example?
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.
Apple Reportedly Looking to Acquire AI Chip Companies
- Thread starter MacRumors
- Start date
- Sort by reaction score
it's been a long time since I heard that phrase, everybody on the performace/workstation side has moved to AMD pretty much
Happy_John
macrumors 68000
There aren't any available to purchase. Foundries / fabs are not plucky start-ups working out of small office spaces.
Why would any foundry / fab owners who are capable of producing memory or NAND chips sell their company to Apple, or anyone else, in the current climate?
They are using it. They are using it to borrow money at relatively very low interest rates.
Apple borrows $10B and has $9-10B in cash reserve that makes it a safer bond to buy than T-bills. (Apple, unlike the US government, has the money to pay the off the bond when it matures. )
The 'stock buyback' stuff is the capital allocation that should stop if investing in buying overpriced AI company was really going to pay off. To similar extent same with dividends. If Apple buying that company was going to return some sky high profit , then stockholders should be OK with letting Apple invest the money.
Diving into the cash backstop would make Apple just a 'broke' as the other AI-herd than is mortaging their future on a bubble. [ As some of the bubble stocks declined over last coule of days , Apple's went up precisely because they are not trying to go into debt up to their eyeballs. ]
Apple is getting a $10-20B per year check from Google for doing next to nothing. ( Apple users generate the money, not work Apple is doing. ). Skip $2-4B off that and it is still a 'large' aquistsion and has done nothing to be cash pile.
If there was ever a 'dust up' with China and Apple got disconnected from their Chinese contracto manufacturers then would need a large cash pile to survive.
The PA Semi thing was to hire talent they didn't have. They have already have AI compute talent in house. What is happening here is that getting spread thin. Apple isn't going to buy a 'product'. (Never seriously sold the PA Semi products after acquistion past what were contractually obligated for. )
This is more so filling talent drain (and/or backsliding on new talent acquisition ) that has been happening for last several years. It isn't a 'chip' thing , it is a 'people with talents' thing.
Spock
Contributor
And because historically the margins have be extremely highly variable. 2-3 years back Apple was short squeezing the memory makers for lower prices.
The memory makers are trying to wait as long as possible before putting new production online because is there is a downturn they will get stuck with all the costs. Vendors like Apple will use the downturn to squeeze the memory makers margins even lower because they 'desperate' and the memory consumers don't want to absorb any of the downside risk.
Last edited:
jz0309
macrumors P6
The memory biz is very volatile, ups and downs for 40+ years. A fab needs to run at almost capacity to not lose money. History has shown that not to be the case over time so companies go out of biz and others limit investments... and margins are low.
We're currently in an "up" cycle, it won't last forever
rp2011
macrumors 68040
Yes. Obviously. What I am saying is this was all obvious years ago and them not getting involved in the commodity software end of AI made a lot of sense., They were never really "behind" as it was always there a phone call away from OpenAI, Anthropic and Google eager to make any kind of money from the money losing endeavor. The point is that Chip design kept them from becoming irrelevant and that avoiding the server chips has continually bit them in their behind and foolish of them to ignore as it was a constant reminder of their blind spot.
Intel has no AI solution work buying that is substantively better than what Apple already has.
As a 'acqui-hire' Intel is horrible as a target. Way too much overhead and talent pools are much broader than
what Apple is specifically short on .
Nevermind, Intel doesn't work at all if it is not selling to third parties. Apple silicon busines has one, and only one, customer. It is about as opposite of a business model as you can possibly get. It is a grossly large mismatch.
Apple can buy wafers from Intel without having to buy the whole company. ( just like can buy wafers from TSMC without having to buy the whole company).
The dependency on Nvidia is mostly tactical not strategic. It is the part of Google cloud that is generic enough to move to another cloud vendor and run on the same 'stuff'. It is an outlet for Apple to move out of Google Cloud is the pricing isn't good enough for them. They don't 'need' Nvidia chips specifically. Just something that is closer to commodity. [ Google TPM chips also have privacy / virtual image features but if Apple built on top of that ... they would be stuck with just one cloud vendor. Apple gets $10-20B per year out of Google so they are the low friction choice since Apple can barter cloud services out of that pot and really doesn't 'cost' them anything from their direct product and services revenue. Apple users are pragmatically generating all that money for Apple. This is far more 'free' Nvidia chips than needing Nvidia chips for technical reasons. ]
Intel would be a gigantic boat anchor dragging Apple down. Intel wouldn't be 'powering' anything. At best the AI acqui-hire thing would be to fill the resources to build a server chip without dragging the Mac stuff backworks. ( missing M6 Pro / Max ... because why? Spread too thin with glasses chip , server chip , and non-Mac chips that don't have timeslice for Mac SoCs ).
Nice, but we still need to solve the RAM issue or it will continue to take 12-14 weeks to get your machine, with a scaled option on RAM.
zarmanto
macrumors 65816
As others have already noted, volatility, razor thin margins and economies of scale have caused some former memory producers to either go out of business or retool into a different, more profitable, market... and by the same stroke, these challenges have made it all but impossible for new entrants in the market. (It literally costs multiple billions of dollars to start up a new fab.) The three major contenders who remain have each found very tenuous but generally sustainable paths to profitability, and their sheer mass enables them to remain profitable enough to weather the periodic volatility. Right now, they're reaping the rewards. This time next year, or ten years from now, who knows?
To wit: when only a two-ton gorilla can ever hope to heft the barbell without the risk of breaking an arm, nobody else tries.
Economics is a cruel mistress.
Tim dropped the ball not investing in AI infrastructure years ago. Now Apple is trying to play catch up and is way behind the 8 ball.
The cash isn't in a pile. It is invested. It is circulating in the economy. People are using it in ways that Apple has no direct control over. It's nice to hear you encouraging Apple to take more direct control over the capital that they have created and accumulated.
sw1tcher
macrumors 604
Samsung, SK Hynix, and Micron are the largest RAM manufacturers. But there are a few smaller Taiwanese RAM fabs -- Nanya Technology and Winbond
And there's more than one (CXMT) in China
On Tim’s way out, he should have Apple purchase Seneca Chip company. They make some excellent chips and they are “fabricated” in the U.S. too.😉
Fast tracked or just skipping M6 Pro/Max that cannot afford to do?
Decent chance the 'missing' M6 Pro/Max are more because the Server chip is using M6 tech (or M5 tech), not M7 and needed more work.
M6 is suppose to be N2 process.
" ... At least a trio of companies are believed to be involved with the chip. Apple is said to be handling the overall design of the chip, while Broadcom is said to be providing some networking technology for it. TSMC is expected to begin mass production of the chip in 2026, using its third-generation 3nm process, known as N3P. That is the same process expected to be used for the A19 chips in the iPhone 17 models next year. ... "
https://www.macrumors.com/2024/12/11/apple-intelligence-servers-new-chip-report/
The substantive issue with the sever chip is that it isn't just Apple making the chip. Broadcomm is in the loop also. Apple will have to coordinate with them. Server chip going into production 2H 2026 wouldn't surface until 2027.
If Broadchip has a chiplet and Apple has a chiplet that are both part of the package , then going to have issues willy-nilly substituting something else in their mid-stream in the R&D process and still hit the same sizes , costs, power efficiency were targeting.
From the lead entries link about Baltra
"...
The server chips that Apple is working on are part of its "Baltra" project, and the chips are expected to be finished by 2027. Apple is working on multiple types of chips, including those with double, quadruple, and eight times the number of CPUs and GPUs as the current M3 Ultra. ... "
Since the M7 Ultra isn't suppose to come until 2028... that really isn't going to help in the next year or so.
Very likely that these server chips are more expensive and complex to product then the M7 Ultra package will be.
Broadcomm isn't going to sell they contribution at rock bottom prices or free. So very likely those features are not in the M7 Ultra. ( the networking that the Mac Studio needs to do is nothing like what Apple is going to need for a more competitive sever chip. )
Apple's better server chip probably isn't going to lighten the load of their "AFMx xxx Pro" workloads.
The server chip likely is going to be so that can handle higher amounts of the "non Pro" foundation models at higher inference user rates in the data centers that Apple has now. Very energy efficiently scaling in their current data center footprint. Not trying to build more buildings than Google Cloud.
If the AFM non-pro models are successfully adopted over 2027-2028 there will be a bucketload of just that inference work to handle.
Reportedly the Apple Private Cloud Compute server utilization levels are crazy low at the moment , but that is because the adoption rate is also crazy low. Decent model that should radically change the amount of workload.
Decent chance Apple is still going to outsource the most exterme compute because not going to all that 'Green energy'. The current iCloud servers already use Google Cloud ( and AWS and Cloudflare and etc.). Apple regularly outsources scaling compute to 3rd party vendors all the time. It is their standard operating model. "AI" isn't going to change that.
Scarcity of wafer starts. Scarcity of packaging processing slots. Lack of RAM modules available to buy.
The biggest 'flaw' in the current Apple PCC SoC solutions is I/O and networking; not 'AI compute'. the baseline M7 architecture is likely not some magic bullet for those issues.
The deal with Broadcomm is to back up talent they Apple does not have excess of. There are a number of folks that left the Apple Silicon group to do projects elsewhere. Nuvia is one. There are about 2-3 other start-ups chasing alternative compute servers that folks from Apple ran off too.
Apple really isn't into maximum data center focus. iCloud is not solely run with just Apple servers. Termus is more a 'product guy" than a "hardware guy". Apple's services business is certainly growing but that isn't a "physical product for sale" business.
How is it possible that Apple has debt? Oh, I forgot that in this society we reward debt and make sure being free of debt is a disaster.
Lots of pontificating going into this comment section. And some good points made. But also I’m pretty sure anyone in the know isn’t posting their official prospectus on Macrumors.
Forget Vision Pro or the Butterfly Keyboard, this is Timmy's biggest screw up ever. Instead of wasting money buying back shares, he should have used that to start making chip fabs. Had he done that, Apple would be worth at least twice what it is now. And would be even more vertically-integrated. Isn't it Business 101 to vertically integrate when you can? Duh!
Oh what a dum-dum Timmy!
Oh what a dum-dum Timmy!
CarlJ
Contributor
I mean,mov course there's a pony - it serves as a safety mechanism. I'm not trying to remove the seatbelts and airbags from my car "because I never use them".
Totally agreed with this. Apple bringing hardware component design / manufacturing capability in-house is generally a good thing.
Debt (bonds) issued by a company is a form of financing. It is Apple borrowing at low interest rates (they are about as safe a bet as the stock market has right now) to fund projects. Think about it this way: If Apple borrows money at 3.25% (that's the rate paid on their bonds which mature in 2029) and can use it to fund projects that will earn at a higher rate, why wouldn't they? In this economy, most large companies use debt financing because to do otherwise means to let opportunities go to their competitors.
Last edited: