Gelsinger “retires”

[Daniel Nenni]

Looks like these guys talked to the Co-CEOs. Some tid bits about what happened with Pat Gelsinger.

Weird. He mentioned NDA discussions with the Intel Co CEOs then they rambled on? Must be a different type of NDA than I see. No way would I sign an NDA with these gentlemen.

I do agree that Pat’s refirement is a debacle and a serious vote of no confidence for the Intel BoD. I have seen other CEO debacles, Sam Altman of OpenAI‘s departure and re arrival being the most recent I would say. Intel CEO BK being fired for a years old relationship being another. BK was incompetent and should have been publicly fired for cause. Same with Pat Gelsinger, he was fired for cause, he did not retire. The Intel BoD full on lied again, investors should hold them accountable!

I learned zero from this video. I learned more from reading this thread so thank you all for the conversation.

 

[DepthFirst]

I genuinely don’t understand how a board of directors is comfortable so brazenly lying. No one genuinely believed that BK was fired for fraternization. No one believes that Pat suddenly decided to retire out of no where without a successor. It’s unprofessional and makes them untrustworthy.

 

[dkr1986]

I learned zero from this video. I learned more from reading this thread so thank you all for the conversation.

Isn't that the point of the NDA!!!? Not reveal any of the conversation content explicitly!

Some tid bits I found are
- BoD was not happy with Pat Gelsinger being too focused on the Foundry, Product needed more focus (also evidenced by their press release).
- It looks like there was some friction PG and BoD on not capitalizing much on the AI product front (AMD managed to do that).
(But I think the comment about bringing a knocked down version PVC or a low precision high TOPS product would not have helped Intel in anyway, the problem with adoption of Gaudi 3 is not hardware capability, its software related, compatibility and absence of upgrade path based on what I learned).
- Also PG was very much against splitting the Foundry and Product fully, that is probably what the BoD wanted. CHIPS ACT grant negotiated by PG includes provisions to ensure Intel cannot fully divest Foundry even in a spin off. BoD might have been not happy with that.
- Then some reinforcement that strategy remains the same but with more focus on products (basically the UBS conference stuff).

Regarding the BoD...I think this is a good read on incompetent and what a joke Intel's BoD is.
https://www.fabricatedknowledge.com/p/the-death-of-intel-when-boards-fail?open=false

 

[Daniel Nenni]

I genuinely don’t understand how a board of directors is comfortable so brazenly lying. No one genuinely believed that BK was fired for fraternization. No one believes that Pat suddenly decided to retire out of no where without a successor. It’s unprofessional and makes them untrustworthy.

In regards to BK, do you think he was given a choice? Fired for adultery and publicly embarrass your family or fired for incompetence and ruin your future job prospects?

 

[Daniel Nenni]

Regarding the BoD...I think this is a good read on incompetent and what a joke Intel's BoD is.
https://www.fabricatedknowledge.com/p/the-death-of-intel-when-boards-fail?open=false

This is a very good piece, but you have to look back longer than that to find the root cause. Intel's problems started back when Otellini was hired as CEO in 2005. It was a bad hire and the Intel BoD is responsible for that. Then they hired BK and it was even worse. Then they hired Bob Swan which I thought was okay since he signed the deal with TSMC which saved Intel's product groups after the 10nm debacle. Had Bob continued as CEO Intel would be fabless and the BoD did not want that which is why they fired Bob and hired Pat.

At least Pat Gelsinger had a vision (IDM 2.0) and he was an excellent cheerleader. Unfortunately, Pat set expectations much too high (5N4Y), as cheerleaders do, and that is why he was fired in my opinion. A valid reason by the way. Pat also did not see the big Q2 2024 loss coming. That is conduct unbecoming of a CEO. There should be no surprises and I am guessing that Q4 will have some surprises as well which may have been the last straw leading to Pat's "refirement". Let's see if that term catches on...

So, to me, the many billion dollar question is: Why has Intel had a board problem for the last 20+ years?

The obvious answer is that the Intel products were so strong prior to 10nm they could do no wrong. Then came TSMC and customers (AMD, Nvidia, Google, Amazon, etc...). Or maybe it is Moore than that? (pun intended)

By the way, I married a cheerleader so I have hands-on experience with this.

 

[hist78]

FYI: I am a big fan of Zinsner, MJ, Naga, Kelleher as I think they are more pragmatic and open than Pat. We might get a more balanced report out on future soon (Maybe I am being too optimistic LOL).

It's interesting that both David Zinsner and Naga Chandrasekaran came from Micron.

 

[Daniel Nenni]

It's interesting that both David Zinsner and Naga Chandrasekaran came from Micron.

Is that a good thing? Is Micron the model of semiconductor competitive efficiency?

 

[XYang2023]

I think Pat Gelsinger (PG) has handled the AI trend poorly (in chronological order to my best knowledge):

1. PVC Designed for HPC and Supported oneAPI
Ponte Vecchio (PVC) was designed for high-performance computing (HPC) and supported Intel's oneAPI framework.

1. 2. Reorganization of AXG and Shift in Focus
   Intel reorganized its Accelerated Computing Systems and Graphics Group (AXG), led by Raja Koduri, merging it into the Data Center and AI Group (DCAI) under Sandra Rivera. This restructuring led to the cancellation of Rialto Bridge, a data center GPU project, and a shift in focus toward Gaudi accelerators. However, according to PG, Gaudi lacks programming flexibility, making it less suitable for AI training. Furthermore, Gaudi relies on its own API and does not support oneAPI. At an Intel tech event, Sandra Rivera also criticized NVIDIA GPUs for being power-hungry. This reorganization was justified as an effort to improve "organizational efficiency," but it coincided with the rapid rise of generative AI technologies, such as ChatGPT.
2. 
   
3. 3. Leadership Changes
   Sandra Rivera later transitioned to lead Intel's Altera FPGA business. Intel brought in Justin Hotard, with PG stating that Justin would help drive "GPU" sales.
4. 
   
5. 4. Falcon Shores Announcement
   Intel announced plans to merge its GPU and Gaudi architectures into a unified product called Falcon Shores, marking yet another strategic shift.
6. 
   
7. 5. Warnings from the Board and Establishment of an Acceleration Office
   Reports revealed that Intel's Board of Directors had warned PG years ago about the critical importance of generative AI and the competitive threat posed by NVIDIA GPUs. PG responded by establishing an "acceleration office," led by a person reassigned from India for the role. However, this individual was later reassigned again, reflecting a lack of continuity.
8. 
   
9. 6. Earnings and Focus on AI Inferencing
   Following disappointing earnings, PG announced that Intel would focus more on AI inferencing alongside x86 architectures. However, this announcement lacked clarity, leading to skepticism among analysts about Intel's plans for Falcon Shores and its overall AI roadmap.
10. 
    
11. 7. Confirmation of Falcon Shores Timeline
    Despite earlier uncertainty, it was later confirmed that Falcon Shores remains on track for a 2025 release, with Jaguar Shores planned to follow.
12. 
    
13. 8. Departure of Gaudi Founders
    The founders of the Gaudi architecture left Intel.
14. 
    
15. 9. Speculation and Confirmation on Arc Desktop GPUs
    PG recently mentioned "portfolio optimization," sparking speculation that Intel might cancel its Arc desktop GPU lineup. However, it was later confirmed that Intel plans to release follow-up desktop GPUs.

Under PG's leadership, Intel's AI strategy has appeared inconsistent and reactive. While PG has invested significant time in foundry and CHIPS Act, Intel has fallen behind key competitors like NVIDIA and AMD in the generative AI space. Despite PG's acknowledgment of Intel's missed opportunity in the mobile market under prior leadership, he seemed to be repeating history by failing to effectively capitalize on the generative AI trend. Clear, decisive action and consistent follow-through are needed if Intel hopes to compete with industry leaders such as NVIDIA.

Money wasted under PG:

1. Significant Hiring Followed by Layoffs
Intel spent heavily on hiring only to lay off many employees few years after, reflecting poor workforce planning.

1. 2. Failed Tower Semiconductor Acquisition
   The attempted acquisition of Tower Semiconductor felt apart, resulting in wasted resources and time and also penalties.
2. 
   
3. 3. Discontinued Initiatives
   Programs like the RISC-V initiative were started but later discontinued, resulting in sunk costs with no long-term benefit.
4. 
   
5. 4. Paused Fab Construction in Israel and Germany
   Intel paused its fab construction projects in Israel and Germany, causing wasted investments in planning and initial development.
6. 
   
7. 5. Overestimating PC Demand During COVID
   Intel anticipated significant PC growth during the COVID-19 pandemic but faced major write-downs recently.
8. 
   
9. 6. Failed investments such as Stability AI

I guess there are more to the list.

1. 
   
2. Key Point:

The amount of money wasted by Intel under PG appears to rival the scale of rewards expected from the CHIPS Act. Instead of misallocated resources, these funds could have been used to:

1. Fund Rialto Bridge Development:
Continuing the development of their data center GPU could have provided a competitive edge in the generative AI space.

2. Seed Strategic Investments:
Intel could have made early investments in promising AI companies, such as Anthropic, to stay relevant in the AI ecosystem.

 

[siliconbruh999]

These are valid points outside of Tower Semi that is no way his fault

 

[Xebec]

Intel CEO BK being fired for a years old relationship being another. BK was incompetent and should have been publicly fired for cause.

I always assumed these scenarios were because the lawsuits for 'lost revenue and reputation' (or whatever) that ex-CEOs could sue for was considered too costly.

(But def. agree BK was wholly incompetent.)

 

[XYang2023]

These are valid points outside of Tower Semi that is no way his fault

I agree with you. If Intel could acquire Tower. The foundry side might be a bit better. It is just puzzling to me that AMD could acquire Xilinx while Intel could not acquire Tower at that time.

 

[Xebec]

Unfortunately, Pat set expectations much too high (5N4Y), as cheerleaders do, and that is why he was fired in my opinion. A valid reason by the way.

A little OT - but discussion question for all -- would anything less than 5N4Y (or "get to 18A with heavy EUV use, BSPD, and GAAFET in 4 years) have been viable given what Intel was facing in 2020-2021?

I'm not sure this was too high expectations vs. what was required to get the business back on track.

 

[MKWVentures]

Is that a good thing? Is Micron the model of semiconductor competitive efficiency?

Model is a tough word. but for financial efficiency and manufacturing efficiency, Micron is miles ahead of Intel . Naga knows how to get things done at half the cost and with less people.. but there are tradeoffs that Intel is not used to and may have trouble accepting. Ann and Naga should make a good team assuming they are working together.

 

[MKWVentures]

A little OT - but discussion question for all -- would anything less than 5N4Y (or "get to 18A with heavy EUV use, BSPD, and GAAFET in 4 years) have been viable given what Intel was facing in 2020-2021?

I'm not sure this was too high expectations vs. what was required to get the business back on track.

Intel 3 and 18A are the only real new nodes since Feb 2021. they need to deliver on those and ramp them.

It is not financially possible to do advanced processes unless you have higher volume than just Intel. So the real goal is ~1 fab worth of external foundry minimum or the finances will continue to look bad (This is all very obvious math).

Intel was supposed to have foundry volume on 16 (ie 22) and intel 3, both of which should be long foundry nodes. Whatever happened to that?

I have heard rumors since foundry day but Intel has not announced anything.

 

[osnium]

Model is a tough word. but for financial efficiency and manufacturing efficiency, Micron is miles ahead of Intel . Naga knows how to get things done at half the cost and with less people.. but there are tradeoffs that Intel is not used to and may have trouble accepting. Ann and Naga should make a good team assuming they are working together.

What are those trade offs?

 

[Xebec]

Pat talking about yields today on X: https://x.com/PGelsinger/status/1865438772013494730

"speaking about yield as a % isn't appropriate. large die will have lower yield, smaller die - high yield percentage. Anyone using % yield as a metric for semiconductor health without defining die size, doesn't understand semiconductor yield. yields are represented as defect densities."

 

[MKWVentures]

What are those trade offs?

Its long discussion on how the culture is different but both Dave and Naga touched on it.

Some hypothetical examples:
When the forecast is to get 100 wafers per hour for step/tool, but you are only getting 50, Do you buy 2x the tools, or plan to hit the wafer out goal and prioritize process change

When you are running a process and testing a 10% wafer output improvement on a tool, do you run the qual and make the change based on data or do you say "its too risky"

Can a process engineer in a HVM fab make a change to reduce spare parts cost in under 6 months if the data supports the change.

Do you spend capital on extra equipment based on possible growth or not.

Do you continuously make changes to the process to improve cost. or do you say "its too complex to change"

Dave and Naga mentioned why Intel has behaved differently than other companies historically. There is whole culture change and organizational change that is needed to be successful on cost and foundry.

 

[Jure]

Intel was supposed to have foundry volume on 16 (ie 22) and intel 3, both of which should be long foundry nodes. Whatever happened to that?

Exactly my toughts. Why are they trying to develop PDK's and all the associated IP that external customers need, on a leading edge node that has not been proven. It would be much easier to get customers on a node that is proven to yield well. Having customers on Intel 16 or Intel 3 would not make them rich, but would gain them experience and knowledge what is necessary and what customers need. In the past Pat has commented that developing PDK's has turned out more challenging than they thought and this is exactly the thing that they should learn before going all in on foundry.

 

[Jure]

...
Key Point:
The amount of money wasted by Intel under PG appears to rival the scale of rewards expected from the CHIPS Act. Instead of misallocated resources, these funds could have been used to:

1. Fund Rialto Bridge Development:
Continuing the development of their data center GPU could have provided a competitive edge in the generative AI space.

2. Seed Strategic Investments:
Intel could have made early investments in promising AI companies, such as Anthropic, to stay relevant in the AI ecosystem.

Agree with many points, especially with fab spam. Intel showed plans for many fabs, bought land, negotiated subsidies, just to cancel projects due to complete financial miscalculation. Pat seems to have serious issue with financial planning.

I still think that Intel is / was in no position to release data center GPU to compete with NVDA. The latter has been developing cuda for 15 year. They have excellent software support (language, compiler, drivers) that works fantastically with their hardware. Developing Rialto Bridge further would've been complete waste of money. Even Google that has developed their own TPUs and also has very good networking, has ordered massive amounts of Blackwell. It just shows how optimized / efficient NVDA hardware and software is. Intel's oneapi is good, but not on the level of cuda.

Inference is easier. Gaudi is a solid option there, but it has many competitors (AMD's MI300, Trainium,...).

Imo, it's a good strategy to develop first lower tier GPU - like what we see now with Battlemage, Celestial, Druid and try to break into inference and edge AI. If AI is to succeed, it should spread massively and AI accelerators will become an integral part of every PC.

 

[XYang2023]

Agree with many points, especially with fab spam. Intel showed plans for many fabs, bought land, negotiated subsidies, just to cancel projects due to complete financial miscalculation. Pat seems to have serious issue with financial planning.

I still think that Intel is / was in no position to release data center GPU to compete with NVDA. The latter has been developing cuda for 15 year. They have excellent software support (language, compiler, drivers) that works fantastically with their hardware. Developing Rialto Bridge further would've been complete waste of money. Even Google that has developed their own TPUs and also has very good networking, has ordered massive amounts of Blackwell. It just shows how optimized / efficient NVDA hardware and software is. Intel's oneapi is good, but not on the level of cuda.

Inference is easier. Gaudi is a solid option there, but it has many competitors (AMD's MI300, Trainium,...).

Imo, it's a good strategy to develop first lower tier GPU - like what we see now with Battlemage, Celestial, Druid and try to break into inference and edge AI. If AI is to succeed, it should spread massively and AI accelerators will become an integral part of every PC.

I think continuing with data center GPU development is part of the learning process, particularly through customer engagements. Intel could even deploy software engineers to client sites to ensure their code works effectively and collect feedbacks.