I think Intel HR department is a unique source of awfulness at Intel.
Their practices are idiotic and hurt the company a lot.
To name a few:
1) Freeze and glut of Hiring: Things are bad, hire no one! Things are good, hire everybody!
This "Feast or Famine" Lead to overall worse employees (missing good hires at Freeze, hiring bad at Gluts), increase Churn (Because you over-hire at the Glut) and hurt Morale (Since you fire and hire not-so-good employees).
2) Optional Firings: This means giving the Employee a choice: Get fired with a separation package, or stay. If you stay, you might get fired later, hint-hint, but might not. This basically makes the good people who can find work elsewhere to take the money and quit.
3) Algorithmic firings. The 2016 big firing at Intel (a Decade ago, I know) were basically done by a few metrics. Everyone who didn't get Successful last year or was a long time at the company got an optional separation package. Managers begged their Employees not to take them.
I have known one guy who got a really big separation package, took it, but was needed so he came back to work as a consultant (for higher cost). Another guy didn't take a separation package, and we joked he is now working for free.
Which leads me to:
4) Over-generous separation packages. I mean, It is nice, but 2 years of pay (when not in the contract) for a separation package is a bit excessive, especially when it is optional, and the root cause for the firings is to try to save money.
5) No Trust in managers: Managers have very low flexibility with their personal budget. They can't promote someone, and give someone else higher stocks. They are extremely limited by the system. Why? Also, there are changes all the time (so it might be out of date).
6) Hard to fire: The process to fire someone for a manager is quite onerous and long. Some managers prefer to try to push the employee to another team, or just find a place for him where he can do no damage or at least benefit a little.
So the HR practices of Intel make it a less dynamic workplace with worse people. The first thing that I would torch is the HR department.
I see Nvidia coming out with a high-end, high margin, real AI PC. Then, a year later, coming out with a 3x faster machine at the same price and the original one at half price.
That would take the profits out of AI PC’s for Intel and AMD. Same game as Nvidia played with graphics cards.
They might even come out with ARM and x86 versions of their PC. Maybe have TSMC build the board for the ARM PC and Intel Foundry build the x86 board.
As a software developer, I would love to be able to run Claude Sonnet 3.7 (or a minified version of Sonnet) locally.
What’s the TAM? Maybe 10m units at $10,000 which equals $100B over a couple of years.
That "capacity constraint" in the Intel 7 node is puzzling. It would be interesting to know how many of Intel's people in AZ who are now trying to get the 18 Angstrom node up-and-running for mass production were reassigned from their Intel 7 node. Trying to get a completely new and very challenging node like 18A shipshape requires a large commitment of the most experienced personnel, which means the the 7 nm node might currently simply be short-staffed. Intel 7 is a really complex DUV node with several multiple patterning steps, and keeping that humming along requires lots of ongoing TLC by engineers and technicians who might now be busy with 18A. Just guessing here, of course, but at least that would explain the constraint capacity. It's a problem for Intel, of course, because their Raptor Lake refresh line is the one that is their most profitable right now.
I just recently came across your substack and am greatly enjoying reading your articles- thank you for sharing your perspectives so openly. I write a lot about Intel in my own substack, SemiconAlpha. I've written about this anomaly (first visible in Q125) whereby Intel's customers suddenly decided they wanted 2+ year old Intel client CPUs instead of the latest and greatest. Complete horseshit IMO. I think they decided Lunar Lake was too expensive to ramp any further and made the decision to max out the Israel Fab (running Intel 7, formerly known as Intel 10) in a desperate move to get any kind of product on shelves
DISCLOSURE: I am an Intel shareholder now and have had AMD and Intel stock before. I am NOT A FINANCIAL ADVISOR, I only studied CompSci.
What is refreshing is that Lip Bu does not share of the Gelsinger excessive optimism that was too out of reality sometimes. I like Gelsinger but they really do need to be in the mindset of "only the paranoid survive", given there is no lead in any of the products or foundry.
Meteor Lake has some problems:
- It does not meet Copilot requirements (NPU = 34/40 TOPS)
- It seems to be sharing equipment with Intel 3 that is used on GNR DC chips
- It has unimpressive performance and power compared with RPL.
Why would anybody want the more expensive of the two?
I have purchased a laptop for my wife and settled on an older N7 Ryzen 5700U because it is so much cheaper. What AMD put out as a replacement (on N6) is slower and did not translate into cheaper devices. Of note, this 5700U laptop has been produced over a year after the mainstream replacements have been available.
Intel has just refreshed lower end parts (Twin Lake) on the same Intel 7 process.
Some people (and most analysts) seem to miss is that in 2024, almost 50% of TSMC revenue was N7 or older. So the surprising part, at least for me, is why Intel reduced 7 capacity so early. Would've made some sense to have some kind of improved Intel 6 process but they do have to pick their battles.
I think Intel HR department is a unique source of awfulness at Intel.
Their practices are idiotic and hurt the company a lot.
To name a few:
1) Freeze and glut of Hiring: Things are bad, hire no one! Things are good, hire everybody!
This "Feast or Famine" Lead to overall worse employees (missing good hires at Freeze, hiring bad at Gluts), increase Churn (Because you over-hire at the Glut) and hurt Morale (Since you fire and hire not-so-good employees).
2) Optional Firings: This means giving the Employee a choice: Get fired with a separation package, or stay. If you stay, you might get fired later, hint-hint, but might not. This basically makes the good people who can find work elsewhere to take the money and quit.
3) Algorithmic firings. The 2016 big firing at Intel (a Decade ago, I know) were basically done by a few metrics. Everyone who didn't get Successful last year or was a long time at the company got an optional separation package. Managers begged their Employees not to take them.
I have known one guy who got a really big separation package, took it, but was needed so he came back to work as a consultant (for higher cost). Another guy didn't take a separation package, and we joked he is now working for free.
Which leads me to:
4) Over-generous separation packages. I mean, It is nice, but 2 years of pay (when not in the contract) for a separation package is a bit excessive, especially when it is optional, and the root cause for the firings is to try to save money.
5) No Trust in managers: Managers have very low flexibility with their personal budget. They can't promote someone, and give someone else higher stocks. They are extremely limited by the system. Why? Also, there are changes all the time (so it might be out of date).
6) Hard to fire: The process to fire someone for a manager is quite onerous and long. Some managers prefer to try to push the employee to another team, or just find a place for him where he can do no damage or at least benefit a little.
So the HR practices of Intel make it a less dynamic workplace with worse people. The first thing that I would torch is the HR department.
In the
One reassuring thing is that the effort to transform Intel is at least *visible* and out in the open.
Samsung, on the other hand, isn’t even showing signs of that kind of movement.
I see Nvidia coming out with a high-end, high margin, real AI PC. Then, a year later, coming out with a 3x faster machine at the same price and the original one at half price.
That would take the profits out of AI PC’s for Intel and AMD. Same game as Nvidia played with graphics cards.
They might even come out with ARM and x86 versions of their PC. Maybe have TSMC build the board for the ARM PC and Intel Foundry build the x86 board.
As a software developer, I would love to be able to run Claude Sonnet 3.7 (or a minified version of Sonnet) locally.
What’s the TAM? Maybe 10m units at $10,000 which equals $100B over a couple of years.
Anyone else see this happening?
That "capacity constraint" in the Intel 7 node is puzzling. It would be interesting to know how many of Intel's people in AZ who are now trying to get the 18 Angstrom node up-and-running for mass production were reassigned from their Intel 7 node. Trying to get a completely new and very challenging node like 18A shipshape requires a large commitment of the most experienced personnel, which means the the 7 nm node might currently simply be short-staffed. Intel 7 is a really complex DUV node with several multiple patterning steps, and keeping that humming along requires lots of ongoing TLC by engineers and technicians who might now be busy with 18A. Just guessing here, of course, but at least that would explain the constraint capacity. It's a problem for Intel, of course, because their Raptor Lake refresh line is the one that is their most profitable right now.
I just recently came across your substack and am greatly enjoying reading your articles- thank you for sharing your perspectives so openly. I write a lot about Intel in my own substack, SemiconAlpha. I've written about this anomaly (first visible in Q125) whereby Intel's customers suddenly decided they wanted 2+ year old Intel client CPUs instead of the latest and greatest. Complete horseshit IMO. I think they decided Lunar Lake was too expensive to ramp any further and made the decision to max out the Israel Fab (running Intel 7, formerly known as Intel 10) in a desperate move to get any kind of product on shelves
Good article! Could you please write one for QUALCOMM? QCOM stock is now very interesting after their Q1 report.
this is so, so good
Great write up. I too pwn shares in this dumpster fire - i misspelled owned, but pwned seemed too appropriate.
DISCLOSURE: I am an Intel shareholder now and have had AMD and Intel stock before. I am NOT A FINANCIAL ADVISOR, I only studied CompSci.
What is refreshing is that Lip Bu does not share of the Gelsinger excessive optimism that was too out of reality sometimes. I like Gelsinger but they really do need to be in the mindset of "only the paranoid survive", given there is no lead in any of the products or foundry.
Meteor Lake has some problems:
- It does not meet Copilot requirements (NPU = 34/40 TOPS)
- It seems to be sharing equipment with Intel 3 that is used on GNR DC chips
- It has unimpressive performance and power compared with RPL.
Why would anybody want the more expensive of the two?
I have purchased a laptop for my wife and settled on an older N7 Ryzen 5700U because it is so much cheaper. What AMD put out as a replacement (on N6) is slower and did not translate into cheaper devices. Of note, this 5700U laptop has been produced over a year after the mainstream replacements have been available.
Intel has just refreshed lower end parts (Twin Lake) on the same Intel 7 process.
Some people (and most analysts) seem to miss is that in 2024, almost 50% of TSMC revenue was N7 or older. So the surprising part, at least for me, is why Intel reduced 7 capacity so early. Would've made some sense to have some kind of improved Intel 6 process but they do have to pick their battles.
Lunar Lake is a whole other balancing act.