It's not just Intel, I think merchant chips will not be the business it was and still is for AMD and Nvidia. Nvidia might be the last mammoth that survives for a lot longer. AMD really depends on the client business remaining x86 to continue their success, and, like I said, it now looks like clients could be iffy for merchant chip vendors. Apple really looks like it could be a trend-setter. Of course, if that happens there will also be a LOT of consolidation among PC OEMs.
Intel on the Brink of Death
- Thread starter XYang2023
- Start date
hist78
Well-known member
If Intel can't fill its fab capacity, a simple and logical solution will be not to build so many fabs. Intel must build new fabs based on customers' demands, not by speculation. Intel will spend at least $100 billion to expand fabs and other factories. There's no room for a gambler to lead Intel.
The volume from the Amazon - Intel deal announced is too small to have enough impact on the huge capacity Intel wishes to build.
Last edited:
Exactly - I think ‘100B is needed to compete with TSMC’ is not the right number to quote. There is a fixed cost of pathfinding / tech development, and another cost to scaling up for manufacturing (building fabs).
Pathfinding is paid by each foundry but scaling up is indirectly paid by the foundry customers. If Intel’s design arm wants to stay at the leading edge, they will need to pay up, but they get to choose to pay themselves or an external foundry. In the latter case they’ll also need to pay for the foundry’s profit margin.
So the real question is really how much money it takes to do pathfinding? 10000 workers’ salary is probably somewhat more than 1 B per year. Intel Hillsboro is about 20k employees, so even if everyone is working on tech development, that’s only about 20-30B over 10 years.
Finally Intel doesn’t need to be cost competitive with TSMC to be successful. Their costs just need to be competitive with TSMC’s cost plus profit margin. Their foundry can sell at cost and make zero profit while trying to get more customers, and their product team can still make a healthy profit.
hist78
Well-known member
When I mentioned that "Intel will spend at least $100 billion to expand fabs and other factories", I was using Intel's own number. In order to receive $10.86 ($7.86+3) billion of US CHIPS Act grants and 25% tax credit for qualified investment, Intel is obligated to spend $100 billion or more to build, enhance, or expand Intel's fabs and factories. I even suspect this tremendous obligation might be one of the reasons got Pat Gelsinger in trouble with Intel's Board of Directors.
Intel, Biden-Harris Administration Finalize $7.86 Billion Funding Award Under US CHIPS Act
Award to support Intel’s $100 billion investment plans to expand American semiconductor manufacturing and technology leadership.
www.intel.com
tooLongInEDA
Moderator
Useful analysis.
But this argument that Intel can run its foundry at cost has never convinced me.
Yes it theoretically could. But it wouldn't count as success in my book.
Let's suppose it could do this and is competing against AMD (using TSMC). Assume AMD can buy its packaged chips from TSMC for $100 (assume TSMC does packaging and test here - even if it doesn't). And sells them to OEMs for $200. And that TSMC can produce these at $50 (selling at $100). Following your argument, Intel can produce at [up to] $100.
But that doesn't actually make Intel competitive against AMD - and certainly not as an investment:
* Intel is carrying the 20K employees in Hillsboro that AMD doesn't ($2-3bn a year - your estimate)
* Intel has tied up a huge amount of capital in fabs for which it is seeing no return - and there is actual cost to this (the money certainly has to be borrowed today, or would be invested at a return if Intel had spare cash)
* Intel's return on capital invested is much lower than AMD's
* Intel's revenue per employee is far lower than AMD's
Also, the net profit generated by AMD/TSMC is higher than that from the integrated Intel model, for [roughly] the same total capital and labour employed.
siliconbruh999
Well-known member
The profit is shared between TSMC/AMD but in turn Intel has the entire profit for the chip produced and also the volume something AMD lacks vs Intel.
there are cons as well like you said but they can improve their cost structure and make better sure it will be higher than TSMC/AMD but they have many inefficiencies they can cut to improve it.iI can never be as low as TSMC cause you are comparing a Subsidized company vs non subsidized company
They can turn their buisness profitable sure it will be less than TSMC+AMD Combined but they have room for improvement we had seen their efficiency someone needs to just clean the house
Last edited:
Well, first, don't take my 20-30B over 10 yrs estimate too seriously. I'm just a random guy. My point is that 100B is too high and includes a lot of fab building that Intel must otherwise pay an external foundry. Many major mature industries have a duopoly structure, so if they can stay at parity with TSMC, keep production costs within ~1.5 - 2x of TSMC, they will do ok (but I would also not call that a success).
More importantly, it is a fallacy that Intel will be forever behind TSMC. That they are behind now does not imply they will forever be in the #2 spot.
In contrast the design space is getting crowded and x86 is not going to change the world (again). Intel needs to run fast toward the problem. Fab parity or even leadership is their key differentiation factor. It would also be nice to identify some key product in a growth market that will become big in 10 years. I feel like they are trying many things but they don't have a continuous strategy.
Finally to the IDM/pure foundry model: none of the foundries are players at the leading edge except for TSMC. Conversely, is anybody thinking about conflict of interest issues when Intel's 10/7nm process is such a disaster? If Intel can capture tech leadership or parity, the customers will come. It is necessary that Intel Design is the first whale customer. The Intel/foundry split must happen eventually, but after tech leadership is established, not before. As someone said, TSMC's currency is ultimately trust. Intel couldn't produce for itself in the last 5 years; why would anyone else trust them? A fab/design split now will maintain the lock between design and fab companies (like AMD/GF), kill the possibility of design/fab integration, and prohibit sharing of resources. It will doom both entities.
siliconbruh999
Well-known member
I think $100 billion also includes the R&D expenses as wellWhen I mentioned that "Intel will spend at least $100 billion to expand fabs and other factories", I was using Intel's own number. In order to receive $10.86 ($7.86+3) billion of US CHIPS Act grants and 25% tax credit for qualified investment, Intel is obligated to spend $100 billion or more to build, enhance, or expand Intel's fabs and factories. I even suspect this tremendous obligation might be one of the reasons got Pat Gelsinger in trouble with Intel's Board of Directors.
Intel, Biden-Harris Administration Finalize $7.86 Billion Funding Award Under US CHIPS Act
Award to support Intel’s $100 billion investment plans to expand American semiconductor manufacturing and technology leadership.
Last edited:
It occurred to me as I caught up on this thread, CEOs previous to Gelsinger did invest in AI. Krzanich acquired Nervana, and Swan acquired Habana Labs. Nervana was a software solution, which even I raised an eyebrow to in terms of competitiveness, but it did align well with Intel's ongoing everything-looks-like-a-CPU strategy. As I've posted before, Habana has a very interesting custom processor design, but from this outsider's view Gelsinger never made it a priority.
One of the few areas we disagree about. Intel is did not have the right fabrication strategy or business model for low cost chips, then or now. Job's price target was too low for Intel.
I agree but Intel could have pivoted into mobile SoCs. Apple wrote some very big checks and it was a collaboration. Apple would have brought a whole host of partners to Intel for collaboration like they did with TSMC. Getting Apple as an EDA or IP customer was all the rage back then. I was right there. Apple business could make or break you. It was an exciting time in the semiconductor ecosystem.
One of the quotes I heard at the time from Intel was that "phones belong on desks!". Seriously, that was the Intel mentality. We have a chapter on Apple in our Arm book. https://semiwiki.com/books/Mobile Unleashed - front to back.pdf
MKWVentures
Moderator
The Apple deal was not possible for Intel to accept. the reason for rejection was correct.... Intels costs were and are way too high. Trust me. They would have lost billions then had to walk away. Intel could not and cannot do mobile as they are too expensive and too slow. They are not good at this despite prioritizing it and trying very har
When do you think this deal from Apple was proposed?
Early 2000. Then why did Intel later come out with the Atom mobile SoCs?
I remember when we were researching the Arm book a friend of mine from eSilicon told me a story about one of the first Apple ASICs they made, it was for the iPod. Apple had originally requested a specific number, let's say 1M chips, so the pricing was set for that amount. Apple sold more like 100M iPods so they tried to renegotiate pricing. The point is that Apple had no idea how many devices they would sell with the first iProducts.
I was not aware of that proposal. I was only aware of the proposal for the iPad, years later. The Atom mobile SoC wasn't launched until years later; the news stories available say 2012. I personally heard Otellini say he wasn't building extra fabs for Apple's pricing requirements. He even mentioned the potential deal in an all-hands meeting, though I don't remember the year. "Everyone has an iPad on their kitchen table. I do." Sometimes Paul was amusing.
MKWVentures
Moderator
I am well aware of the entire Intel Mobile history ... years of Product and process reviews.
Simple level ... lets call this hypothetical.
1)Intel was VERY committed to getting into mobile. Intel proposed and did prototypes for a handheld PC /phone LONG before Apple
2) Intel developed Arm processors and communications chips. A lot of them. Some didnt even get cancelled for being late.
3) Apple proposed a partnership for chips. The numbers were very low. Lets guess the price was ~1/2 the cost. it was not good. The major miss was saying no one could meet the apple price.
4) Intel manufactured AP and comms products for Mobile. All products were 2 years late, and yet somehow we had the AP for the most popular blackberry/TREO/semismart phones for about 9 months.
5) Intel lost billions. 90% of products missed the window. Intel was doing comms for all mobile too. Intel probably spend more on mobile arm processors/comms than any other company. It was also a high corporate priority
6) a certain IA manager said "we can do a x86 mobile product and it will be better than strongarm and we will deliver on time".
7) Intel Killed the strongarm business and sold it off ... mostly to Marvell. Marvell immediately moved fab work to TSMC and others
8) Intel didnt do much better with Atom. It was OK for netbook and tablet. No good for phone. Shockingly cost was much higher than the ASP
No one tried harder at mobile than Intel over the course of a decade or more. Turns out that Slow and expensive is not a good skill to have in the mobile or graphics market
Steph Curry cannot play QB like Patrick Mahomes. Patrick Mahomes cannot hit like Juan Soto. Juan Soto cannot shoot like Steph Curry. Nvidia Can't make cars. It is not because they didnt have the right strategy LOL
Enough ranting for a Friday
hist78
Well-known member
Why didn’t Intel try to cut down costs to compete when Apple was asking?
hist78
Well-known member
Don't forget Intel's "Contra Revenue" rebate scheme that made Intel mobile division to have "negative revenue"!!
lol - literally wrote this same thing *then* read your message. The Contra Revenue / Atom scheme is super ironic given the Apple / cost / fab "dilemma" Otellini faced.
I forgot about contra revenue. That was iconic! Can the Intel Foundry have that too?
thank you for the belly roll.I think it has to be thought of in annual costs vs a single number.
It is estimated that Intel spent ~ 20bn on 18A development ( nearly enough to purchase 2 US Navy aircraft carriers). Now, some of that could well have been catch-up spending; however, based on what I have been reading about wafer costs and EUV equipment costs, it seems like the next step beyond this generation will likely be even more insane in costs.
It seems like "EUV" isn't a precise enough term for the ASML equipment either. You have to specify High NA for the latest equipment (as an example). So even though EUV has been around for some time, the equipment from 10 years ago can't be used to do the best processes today from what I understand.
So there is an on-going cost for every new node as more equipment is purchased for the new line process. And every so many years the equipment, process and libraries are all revamped for something quite new it seems.
All foundries must then have enough revenue to cover this exponentially increasing cost. As a result, it seems obvious to the most casual observer that the rate of process changes can not continue as it is economically unfeasible.
Intel would require continued large investments while it attracts customers to cover its volume. It would need to have onboarding capabilities and tools that it must develop and mature to do this in a meaningful way. All the while the cash bleeds red.
It can be done, but it wont be pretty for some time. Intel would need to reach steady state financially for this to be viable.
I very much want Intel to be successful. We in the US need them to be for our collective security. I am not opposed to the government helping them regain their footing even understanding how an expensive proposition that can be as the potential alternative may well be much much more expensive in the long run.