Hacker News Reader

Kind of disagree. Have read on forums that by all accounts, the tech in the 18A process is amazing (Gate-all-around transistors, backside power-delivery, 3d packaging tech...), and that they are on the path to catchup/pass TSMC in terms of chip-fabrication performance/power-expenditure in a year or two. Yeah, this is an amazing achievement for such a large company.

But have also ready that to make this profitable is like a 5-8 year plan, not something that is profitable in 3. To me, this is his biggest mistake, he needed to keep the ship sailing at least somewhat smoothly while waiting for this big gamble to pay off.

And eventhough I do agree with you that he made a lot of mistakes, still he got the the most important issue right, regaining manufacturing leadership.

I would love to see a source for you first paragraph.

This forum is full of engineers and they talk about 18A (I myself am only a software engineer but who took hardware design classes waay back in college, so can understand 60% what they're are saying but am no expert).

https://semiwiki.com/forum/index.php?threads/intel-18a-too-g...

And obviously, many of these posters are like me and have a bias towards Intel succeeding. Still, it sounds like 18A seems like it's on track in terms of the technologies (at least for now)

Even if 18A can catch TSMC, which I highly doubt, what does it matter if they don’t have any of the big chip designers like Apple and Nvidia?

Depends how similar the PDK is.

That's the only thing? We're talking Apple here. If you can't provide tens of millions of chips per quarter, you're useless to them.

TSMC not only has the best fabs, the chips fly out of those fabs at blinding speed.

There are more custers besides apple. Intel is not a small manufacturer.

No they are not, but one big part of their problem they have in their attempts to catch up to TSMC is that they can't get reliable yields above 30%, while TSMC has a better node AND they have 60% yields.

Yeah, that is why some outsiders say this is more like an 8 year strategy, it takes time to convert customers.

IMHO, if you have a great product and a market with significant demand, seems like eventually the customers will come - and it seems like this is possible if Intel can retain manufacturing leadership for a few years and build out a better ecosystem to support external chip development by foundry customers, but this of course take time.

> * Intel still has more employees in 2024 than in 2019. Instead of trimming fat, he added more fat.

Intel hiring practices under Gelsinger were really puzzling. Did he think that by hiring thousands of random university graduates there will be the next Gordon Moore among them? It was also discussed on HN: https://news.ycombinator.com/item?id=42300028

I can kind of understand the extensive poaching of TSMC engineers in Arizona. Though making TSMC upset is playing with fire as much of Intel's business now depends on them.

2 days late, but my 2 cents:

> Can't win over AMD, Nvidia, Apple, Broadcom as fab customers

Not surprising, and arguably not to be expected. His 5 nodes in 4 years plan ends in 2025, and only then is the fab "good enough" for external customers.

Specifically, they want to see how well Intel's own product on 18A performs before committing. So the real test is in 2025-2026.

> Fab cancellations and delays, wasting a ton of money

The cancellations and delays are on the expanded fab roadmap that he introduced. Compared to before he joined, the fabs are in a much better place. Essentially, this is complaining that he committed to opening 5 new fabs and only opened 3 (or whatever the number is).

> Not hiring the right people to steer Intel's internal fab culture to external quick enough

Perhaps. 5 nodes in 4 years, which Intel may actually achieve, is unprecedented, though. No other company has done it (and likely never will again - it's crazy expensive). So at least as far as execution goes, the fab is in much better shape than when he joined.

> Not simplifying product roadmap.

Definitely agree.

> Didn't stop paying dividends until August 2024

They cut them significantly in early 2023. Yes, they should have done it all earlier, but keep in mind that the board needs to approve it. Or at least, not fire him for doing it. On the plus, he did stop stock buybacks almost immediately after joining.

> Didn't see that Intel was swimming naked during the covid boom and that after the boom, Intel would be in huge trouble due to inferior products

I'm sure he knew the products were inferior. What his failure was is to see that the overall PC market would drop significantly. As an example, in one year it dropped globally 25%. Intel's share dropped 33%, with AMD picking up the remaining 8%. That extra 8% for AMD was something like a 30-50% increase in sales.

> He hired 20% more workers since he joined but 54% less revenue

Totally agreed. What's worse, after the 5% cuts in 2022, Intel was back at the same headcount before the current layoffs. How do you lay off 5% of the company, claiming you need to save money and in 18 months grown the company to the same level?

If you want to see his positives, look here:

https://news.ycombinator.com/item?id=42336143

Reasonable takes.

I think the employee count blowup was his biggest mistake because it shortened Intel's runway towards being an external fab company.

Lip Bu Tan, who was on the Intel board, recently resigned because he did not agree with Pat's employee count plan and company culture. Tan was likely right.

  Specifically, they want to see how well Intel's own product on 18A performs before committing. So the real test is in 2025-2026.

Another mistake of Pat's is that he did not split the company fully. AMD, Nvidia, Apple, Qualcomm aren't voluntarily going to fund a competitor by using Intel fabs. Furthermore, Intel not splitting means there will always be small conflicts of interests with Intel designs teams. Pat wanted the cake and eat it too. He wanted to compete with TSMC in addition to Nvidia, Apple, AMD, Qualcomm. Now Intel is at risk with competing against neither group.

I think not having any major external customer signed onto 18A in 2024, approaching 2025, is a huge failure on Pat's part.

> I think the employee count blowup was his biggest mistake because it shortened Intel's runway towards being an external fab company.

It was a mistake, but in terms of runway, would not have made a large difference. The capital costs of running a fab dwarf the cost to pay salaries. That's why they still do fabs in the US - outsourcing to another country doesn't save much money.

> Another mistake of Pat's is that he did not split the company fully.

I don't know that anyone knows what his plans were on this. Internally within the company (and perhaps externally), he was clear on a few things:

- If Foundry cannot become profitable, he'll shut the fabs down and kill Foundry altogether.

- The Products side will be charged market rates, and they can choose if they want to use Intel's fabs or TSMC

- If an external company is willing to pay more than Intel Products, they'll prioritize the external company.

So he was trying to set it up so that it was effectively two separate companies. I guess it was either harder than expected, or there was significant pushback. When he joined, the two sides were heavily intertwined. His goal for 2023 was to get things separated enough so we could get separate P&L statements for Foundry and Products. He achieved it - but I don't know if it could have been done quicker.

The other (obvious) reason he didn't want to break the company up: Intel Foundry is unprofitable. They need the revenue that the Products brings in to keep it alive. The goal was to hit profitability in 2027-2028, when there are enough customers. Splitting it up now will kill Foundry altogether (and now with the CHIPS act, they are not allowed to kill Foundry).

As for the concern about funding competition - not sure how much that is true. The vast majority of TSMC's customers (including AMD) do not want to use the leading edge node. It's more expensive, and has higher quality problems. They are happy to let Apple take it. Likewise, the narrative was that most of these companies will prefer not using Intel's latest node, and would be happy to let Intel use that capacity. So I think the plan was to make 18A mature, move to 14, and then external customers will use the mature 18A node.