TSMC basically said that operating cost competitive fabs is impossible in the US. They tried in Arizona, couldn't do it, and delayed the fab. Japan, China, and the EU are already paying out subsidies and fabs are already up and running. Meanwhile the US is bean counting and dragging out the process. Intel explained the problem back in 2020, everyone agreed that it's strategically important for the US to produce chips domestically and that there is a real cost to operating in the US. Here we are in 2024 and Intel still hasn't gotten any money from the federal government, they still need to jump through a bunch of hoops to get those subsidies.
There were two bigger issues with tsmc in the US. They couldn’t get workers, especially for the salaries they were offering. And they didn’t want to give the US the advanced nodes or complete assembly because they don’t want to lose that advantage in Taiwan. TSMC is pretty much the only thing keeping Taiwan safe from China as the US will protect them over it.
US defending Taiwan predates TSMC. It has been US policy since the 50s, a time where TSMC definitely wasn’t a factor. Its geographic importance dwarves its semiconductor importance.
It's crazy, it's almost like if you have to offer salaries to compete with Intel it's hard to undercut Intel on price.
TSMC's business model has always been operating factories better and cheaper than the competition. Taiwan wages were major part of the strategy.
This is accounting nonsense. If you treat products and fabs as entirely separate, you have to pretend the product division is "paying" the fab division. And by changing the "price" the fab charges, you can shift profits wherever you want. The fab may be "losing money", but the product division sure is "doing great".
But they are not simply “making up” the price. The baseline price can be derived from TSMC.
Intel products are pretty profitable if their Intel Fans could build stuff at TSMC prices.
That’s how I interpret it.
> And by changing the "price" the fab charges, you can shift profits wherever you want. The fab may be "losing money", but the product division sure is "doing great".
The idea is to represent what the product division would be paying if they were buying identical wafers from e.g. TSMC. So the conclusion is that the product side is actually pretty cost-competitive, but is being dragged down by their usage of an uncompetitive foundry.
The entire point is for them to be viewable in isolation. And yeah, if the fabs keep losing them money, as they are today, they'll eventually be shut down or sold off.
```Can't eat a $7 billion loss indefinitely.```
Who says they have to?
Just some initial investments until they're caught up with the bleeding edge once again.
It's not like TSMC forgoes r&d costs either, they also spend billions. The difference is that their customers pick up the tab so TSMC can post profits.
```The entire point is for them to be viewable in isolation.```
Explains why Intel is willing to allocate fab space to customers now. Nvidia being one of those customers.
https://www.techspot.com/news/101725-nvidia-turning-intel-foundry-services-expand-gpu-ai.html
> Just some initial investments until they're caught up with the bleeding edge once again.
That's exactly the gamble Intel and investors face. If they succeed in achieving competitiveness with TSMC (across PPAC), then all's well. But it's not guaranteed that they will, and the fabs will burn a lot of money in the meantime.
> Explains why Intel is willing to allocate fab space to customers now. Nvidia being one of those customers.
All that rumor says is that Nvidia is investigating Intel's packaging tech as a second source to TSMC. Also, this isn't about allocating fab space; it's about winning customers. Intel has plenty of fab space free today, but no one wants to use it.
```That's exactly the gamble Intel and investors face.```
There's no 2 ways around it, it's a gamble. Which is why they're trying to separate the foundries from the I.P.
```All that rumor says is that Nvidia is investigating Intel's packaging tech as a second source to TSMC.```
This is the same Nvidia that used TSMC 12nm and Samsung 8nm while AMD was on TSMC 7nm. If it makes business sense Nvidia they'll consider it.
Also a good way for Nvidia to manage costs for gaming related hardware while using bleeding edge nodes for data center/A.I.
```Also, this isn't about allocating fab space; it's about winning customers. Intel has plenty of fab space free today, but no one wants to use it.```
Build it and they'll come. Companies are constantly looking for alternatives, you just have to make it worth their while.
> Just some initial investments until they're caught up with the bleeding edge once again.
This sub never changes even in the face of extraordinary evidence.
Look at this thread from 2020. https://www.reddit.com/r/hardware/comments/hwnovh/intels_7nm_is_broken_company_announces_delay/
Commentors were sure that just because intel had problems with intel 7 it did not mean that intel 4 would as bad because intel would just put in the effort and everything would be fine.
Intel 4 has not ramped to this day in a meaningful way. Even though after the initial admittance that intel 4 was delayed and telling us it would ship at end of 2022 beginning 2023.
As I tried to explain in the other thread intel going from node to node is not a law of nature or a money problem. Intel has to overcome engineering challenges that it can not resolve because of its culture.
You can downvote me like in the other thread. But four years later I will be right again that intel will not have caught up. Intel 18A will not be as good as advertised. etc.
It's called accountability and the lack of it is what knocked down Intel from their leadership position and put them where they are today.
Gelsinger knows this issue and he's been very vocal about it since they hired him back. The last two CEOs probably knew it too but didn't want to rock the boat.
This depends on how independently ran the products division is. In other words, does the product division have an option to use a non Intel fab, and direct its budget to that fab instead of Intel's?
If so, this metric actually becomes very meaningful and shows how Intel's fabs actually compete with other companies fabs.
Given that 30% of Intel's chips are now produced on external fabs (something that would have been unheard of even a few years ago), it does seem that a large portion of product divisions do have this flexibility.
> In other words, does the product division have an option to use a non Intel fab, and direct its budget to that fab instead of Intel's?
It seems like they have more autonomy than they used to, but still aren't completely free to use what they want.
```does the product division have an option to use a non Intel fab```
Three out of the five tiles Meteorlake mobile uses today are using TSMC.
```this metric actually becomes very meaningful and shows how Intel's fabs actually compete with other companies fabs```
They eventually want to compete, but for right now it's for practical reasons. At the moment they fab chiplets at TSMC to remain competitive with AMD. However, intel's end goal is to compete with TSMC as a foundry as well.
well yea that's the point, the number they chose to "pay" the fab was TSMC rates for the equivalent nodes... and it turns out intel's nodes cost them quite a bit more than that to operate.
>During a presentation for investors, Chief Executive Pat Gelsinger said 2024 would be the year of worst operating losses for the company's chipmaking business and that it expects to break even on an operating basis by about 2027.
>Partially as a result of the missteps, Intel has outsourced about 30% of the total number of wafers to external contract manufacturers such as TSMC, Gelsinger said. It aims to bring that number down to roughly 20%.
So yea Intel is going to be reporting operating losses until around 2027 at which point it expects to break even, so expect a more of these sorts of financial statements.
If Intel decided to report things this way in the past, they would have reported operating losses on the foundry side every year. Because the foundry never made money 'per se', and because this separate way of reporting excludes the sales of Intel products.
And what they're aiming in 2027 is for external clients to cover those operating losses on the foundry.
Not even close. What you're saying is that external costumers were covering their expenses AND the expenses related to the manufacturing of Intel products. No shot.
Huh? These numbers are external *plus* internal customers. Both before and now, internal customers represent the vast majority of Intel's foundry business. The difference is in the relative competitiveness of the nodes.
> Edit: Either way, I highly doubt the foundry itself was ever profitable if reporting was done this way back in the day.
Why not? Simply having node leadership would have been a massive financial boon.
The financial boon would appear on the 'Intel Products' side of the report. It's not like the foundry would compete against itself and raise prices against the company it belongs to just because they have node leadership.
It doesn't work like that. Because when accounting is down for these situations the end product recieves a industry average profit margin and intel's manufacturing division would 100% have seen profits.
This pretty much sums up majority of redditor in this sub. People only read headline and draw stupid conclusion, almost no one in here read the article which is sad.
Intel made tens of billions in profits last year. This is just their old accounting trick of separating profits and costs of operations across different years for tax evasion purposes.
It's based in real costs, just offset by Intel's CPU sales. The product teams are basically saying that Intel Foundry is so uncompetitive they need to sell below-cost to keep them as a customer.
If you believe in MAD, mutually assured destruction you will understand that Intel is currently an undervalued company with tons of upside.
X86 is effectively a duopoly between AMD and Intel. AMD has made gains with the Ryzen Architecture and TSMC manufacturing. However, Intel continues to dominate laptops and enterprise solutions.
Frankly, Intel is probably very much alright with AMD posing a small threat to their overall sales tk get regulatory pressure off them.
"If you believe in MAD, mutually assured destruction you will understand that Intel is currently an undervalued company with tons of upside.
X86 is effectively a duopoly between AMD and Intel." - redditracing84
x86 is about to compete with ARM. Historically your statement was accurate, but times are different, and even AMD understands this.
A.R.M has arrived on windows with the Snapdagon X Elite this quarter and results are looking promising.
https://www.xda-developers.com/amd-nvidia-arm-chips-pc-what-it-means-for-the-industry/
https://www.extremetech.com/computing/snapdragon-x-elite-dethrones-intel-meteor-lake-in-pre-launch-tests
\`\`\`If you believe in MAD, mutually assured destruction you will understand that Intel is currently an undervalued company with tons of upside.
X86 is effectively a duopoly between AMD and Intel.\`\`\`
x86 is about to compete with ARM. Historically your statement was accurate, but times are different, and even AMD understands this.
A.R.M has arrived on windows with the Snapdagon X Elite this quarter and results are looking promising.
https://www.xda-developers.com/amd-nvidia-arm-chips-pc-what-it-means-for-the-industry/
https://www.extremetech.com/computing/snapdragon-x-elite-dethrones-intel-meteor-lake-in-pre-launch-tests
```If you believe in MAD, mutually assured destruction you will understand that Intel is currently an undervalued company with tons of upside.
X86 is effectively a duopoly between AMD and Intel.```
x86 is about to compete with ARM. Historically your statement was accurate, but times are different, and even AMD understands this.
A.R.M has arrived on windows with the Snapdagon X Elite this quarter and results are looking promising.
https://www.xda-developers.com/amd-nvidia-arm-chips-pc-what-it-means-for-the-industry/
https://www.extremetech.com/computing/snapdragon-x-elite-dethrones-intel-meteor-lake-in-pre-launch-tests
```If you believe in MAD, mutually assured destruction you will understand that Intel is currently an undervalued company with tons of upside.
X86 is effectively a duopoly between AMD and Intel.```
x86 is about to compete with ARM. Historically your statement was accurate, but times are different, and even AMD understands this.
A.R.M has arrived on windows with the Snapdagon X Elite this quarter and results are looking promising.
https://www.xda-developers.com/amd-nvidia-arm-chips-pc-what-it-means-for-the-industry/
https://www.extremetech.com/computing/snapdragon-x-elite-dethrones-intel-meteor-lake-in-pre-launch-tests
```X86 is effectively a duopoly between AMD and Intel.```
x86 is about to compete with ARM. Historically your statement was accurate, but times are different, and even AMD understands this.
https://www.xda-developers.com/amd-nvidia-arm-chips-pc-what-it-means-for-the-industry/
```X86 is effectively a duopoly between AMD and Intel. ```
x86 is about to compete with ARM. Historically you're statement was accurate, but times are different, and even AMD understands this.
https://www.xda-developers.com/amd-nvidia-arm-chips-pc-what-it-means-for-the-industry/
> If you believe in MAD, mutually assured destruction you will understand that Intel is currently an undervalued company with tons of upside.
Intel has a ton of potential upside, but also a ton of potential downside. What if this gamble on the fabs doesn't work, and they continue burning billions a year until it's shut down or spun off? What if Intel fails to gain traction in AI, while suffering continued erosion of their existing positions in datacenter and client? Based on their history, these are all very realistic outcomes, and that's going to weigh down their stock.
> However, Intel continues to dominate laptops and enterprise solutions.
Enterprise, Intel has to basically give away their chips. Unless they radically close the competitive gap, these markets will remain on the decline for Intel.
The reason Intel is so undervalued, imo, is failure is baked into their existing price. The potential upside to INTC isn't that they may retake former market share - it's that you're betting money they'll make big strides into external fabbing, AI, and dGPUs in the next 5 - 10 years, which are markets they currently are either not functionally in or have just begun entering.
> The "real cost" is way above competitors'.
That's my point, yes. The fabs are paid for the product (the node) they deliver, not the actual manufacturing cost.
> Real cost is the actual incurred cost.
*And that's what I'm referring to*. It's gap between the high real cost vs the competitive/market cost that's making them lose so much money.
I wonder if it's because of their sluggish CPU sales and dwindling market share, especially in the data center sector?
After all, it's not like anyone else is using their foundries in any 'meaningful' capacity. Intel only has four 'fab partners,' at least to my knowledge, and I've only heard of one (Siemens):
>At a customer event in San Jose, Calif., Intel Chief Executive Pat Gelsinger introduced key ecosystem partners for the company's Intel Foundry initiative. Those partners include Cadence Design Systems (CDNS), Siemens (SIEGY), Synopsys (SNPS) and Ansys (ANSS), which Synopsys is acquiring.
[https://www.investors.com/news/technology/intel-stock-chipmaker-building-foundry-for-ai-era/](https://www.investors.com/news/technology/intel-stock-chipmaker-building-foundry-for-ai-era/)
MediaTek is also supposed to be in the list, but things have gotten quiet in the past year or so. Their entire current SoC line-up is on N5/N4 and it seems like they're currently in talks with TSMC for the N3.
It was never going to be a fast transition. Intel is trying to build up a business in 2-3 years that TSMC built over decades, and they're doing it while cash is tight.
They say they're shooting for profitability in 2027 but it's probably going to be a bumpy road until then.
Siemens is one of the largest industrial controls companies in the world. They're present in PLCs/controllers/automation equipment in every single factory in the world.
The comment above clarified that it's referring to Mentor, a similar EDA tool vendor.
> I'm sure they are an ecosystem partner, but they certainly have not-insignificant chip making demands as well.
Irrelevant for this, and probably Intel in general.
> I wonder if it's because of their sluggish CPU sales and dwindling market share, especially in the data center sector?
The next big wave of prebuilts being sold will be next year when Windows 10 becomes EOL. Windows 11 has minimum requirements to be used. This will give Intel a nice long boost because AMD can never match their volume. Datacenter is hard with some companies rolling their own designs or going with AMD's Epyc because the products benefit them more than Intel's Xeons unless Intel rights their ship soon.
I think a quote from [Semi-Analysis](https://www.semianalysis.com/p/is-intel-back-foundry-and-product) on this really helps summarize a lot:
>As an IDM, Intel’s CPU-heavy product portfolio meant that their manufacturing focus was on leading edge processes. Intel didn’t maintain these “forever” nodes (although for a time it appeared as if Intel 14nm was going to be a forever node but not by choice) because their only customer (Intel themselves) had little need for them. Consequently, Intel didn’t have this ever-growing pool of mature wafer capacity that was capable of generating cash flow seemingly into perpetuity.
>
>
Instead, nodes would only be kept around for a few generations before being deprecated to make way for newer processes. While a lot of the equipment was reused and transitioned to newer technologies, they would also need to reinvest in new tooling for that transition. Now, as a foundry, Intel can sustain wafer demand for their nodes by filling capacity with external customers as internal demand goes down over time. This now supports their forever node strategy.
It is no longer financially viable to run leading node fabs strictly internally because:
1) The volume necessary to sustain the bleeding edge will exceed the potential internal capacity
2) You need a large backlog of fully depreciated nodes providing a high margin cash line.
​
The transition from internal foundry to external is going to require **substantial** capex. Nothing here has fundamentally different
I'm pretty sure Intel is going to take on major investment partners to fund CAPEX, something like a joint venture with majority Intel ownership.
In retrospect, Intel is probably kicking themselves for not switching to a foundry model before getting rid of their planar silicon nodes. That kind of capacity would have been gold for the rush to reshore critical silicon.
Intel has been losing money now and then, this isn't really new. Remember the division bug with early Pentium? That was an expensive one as Intel had to recall and replace thousands CPU at their expense.
> The actual margins of the foundry have been allocated to their products division.
No, they haven't. That's the entire reason for this accounting split.
Except they did. They have inflated gross margins of their product division. This is a function of the wafer costs. Zinsner literally mentioned how their foundry and product gross margins will evolve as foundry will start adopting fair market value for their wafers. He indirectly admitted that the wafer costs they attributed were not fair market value.
You're reading that the opposite of its intended meaning. This *is* what fair market value looks like for Intel Foundry. Again, literally why they made this accounting split. It just so happens that with fair market wafer prices, Intel Foundry is losing a lot of money. That's what happens when you have a grossly uncompetitive product.
If you're talking about this part:
> The bulk of wafers are in the life for external - we're pricing our wafers with competitive technology at below market prices to drive revenues and volumes
Then he's basically claiming that to get volume to the fabs, they need to offer a discount. Which is identical to what they'd do if Intel Foundry was independent. So really, it's just that they don't include the risk factor of IFS in their "market price", but *do* in the financials, because that's the reality. I don't see that as a distortion.
91 Comments
ThePandaRider@reddit
sittingmongoose@reddit
Exist50@reddit
Famous_Wolverine3203@reddit
DaBIGmeow888@reddit
Famous_Wolverine3203@reddit
gnocchicotti@reddit
advester@reddit
pascalsAger@reddit
raulgzz@reddit
DaBIGmeow888@reddit
DaBIGmeow888@reddit
Exist50@reddit
raulgzz@reddit
magistrate101@reddit
Exist50@reddit
HandheldAddict@reddit
Exist50@reddit
pascalsAger@reddit
Exist50@reddit
HandheldAddict@reddit
ElementII5@reddit (OP)
YepperyYepstein@reddit
gnocchicotti@reddit
DrBoomkin@reddit
Exist50@reddit
HandheldAddict@reddit
capn_hector@reddit
gajoquedizcenas@reddit
Flowerstar1@reddit
gajoquedizcenas@reddit
Exist50@reddit
gajoquedizcenas@reddit
Exist50@reddit
gajoquedizcenas@reddit
Exist50@reddit
gajoquedizcenas@reddit
nugurimt@reddit
Exist50@reddit
no_salty_no_jealousy@reddit
makememoist@reddit
Exist50@reddit
AThousandNeedles@reddit
allahakbau@reddit
no_salty_no_jealousy@reddit
Plus_Quarter_3340@reddit
imaginary_num6er@reddit
CarbonTail@reddit
SkillYourself@reddit
Exist50@reddit
redditracing84@reddit
Strazdas1@reddit
noiserr@reddit
HandheldAddict@reddit
HandheldAddict@reddit
HandheldAddict@reddit
HandheldAddict@reddit
HandheldAddict@reddit
HandheldAddict@reddit
Exist50@reddit
soggybiscuit93@reddit
dotjazzz@reddit
Exist50@reddit
dotjazzz@reddit
Exist50@reddit
mycall@reddit
GenZia@reddit
gnocchicotti@reddit
monocasa@reddit
Exist50@reddit
Brostradamus_@reddit
Exist50@reddit
TiL_sth@reddit
Exist50@reddit
GenZia@reddit
U3011@reddit
soggybiscuit93@reddit
gnocchicotti@reddit
12A1313IT@reddit
tomytronics@reddit
Exist50@reddit
RulerofKhazadDum@reddit
Exist50@reddit
RulerofKhazadDum@reddit
Exist50@reddit
RulerofKhazadDum@reddit
Exist50@reddit
RulerofKhazadDum@reddit
Exist50@reddit
synthjunkie@reddit
wizfactor@reddit