I'm not saying GM should stop building the Corvette for you and I. I'm just saying they should keep building the Equinox for everyone else.
HA! That's funny. I was going to use a car analogy in my reply. I was going to say that both the needs and wants should be catered to, otherwise we'd all be driving Civics and Corollas. Crazy minds think alike.
A month ago, Pat Gelsinger was talking about industry leading yields of 2 nano and a resurgence of Intel lithography. Now, they are moving to buy more capacity from AMD. When a back straggler talks about leading the race with a miraculous recovery, its rarely a good idea to put much credence in the claim. Cinderalla stories happen but they are freakishly rare. I heard it through the grapevine that Intel is in an existential moment. While they aren't dead, they desperately need to turn things around and all indicators are currently bad.
Intel claims to have their 2nm node under control with better than expected leads but they bought the hell out of TSMC 2nm capacity. I'm trying to figure imagine a scenario in which Pat Gelsinger is not lying. Maybe their 2nm node is the best thing ever but they need way more capacity than they have at 2nm? It seems most likely Intel is buying 2nm capacity because they don't have 2nm game online, as they claim. It also seems well possible Intel is buying 2nm capacity to prevent AMD/nVidia from getting it. This is precisely within the character profile of Intel. They are one of the most anti-competitive companies in operation.
I feel like Intel made a terrible decision, walking away from on die RAM. Intel ditched it due to consumer feed back but consumers are scared of change. That decision will return to haunt them. I remember when integrated graphics caused a "yuk" factor. The machine I'm typing on right now has an IGP. What benefits most by being close to the CPU? GPU or RAM? Think about it. Intel could tune RAM to the nut, certainly beyond what can be done for a standards compliant memory bus. They could lower the pin count on their CPU complex. They could lower system price with reduced connectorization. More control over power management. It seems like a slam dunk.
Gelsinger now walking at street level carrying a box of squeezy foam stress toys topped by his name plate.
I wish I could be a CEO, fail, and make a butt ton of money in doing so. Gotta love it. Two in a row for Intel.
Give it a flyer for a couple of years. If you crater the company, you'll still have $75M and you can try your hand at another company as CEO to see if maybe your luck improves.
So, here's a thought. I understand the government plans to pay Intel to build out new fabs in the US and do so very quickly. This is undoubtedly in response to the looming war with China. Intel could be under tremendous pressure and/or incentive to spawn their fab business in the interest of national security. If that happens, it is entirely possible AMD wafers could be made by Intel. I don't know where they would be packaged and bumped. Hopefully, the US government is smart enough to know they need more than just wafers to make a complete integrated circuit part. I'm 99% sure they are. lol!
First, 18A yields were super high; way better than expectations. Now, 18A yields are said to be abysmal. The latter is more easily believed. It takes months and sometimes years to develop a process to a reliable state. The thing is, none of that means a thing. Total BS. We would need to know the average number of inclusions and the wafer size, to understand how well the process is performing. Meanwhile, it has become clear that only foundries on the leading edge node will be profitable. Once the fall back much at all, customers leave and the money to move forward dries up. It's a winner take all business. Samsung and SK Hynix are fighting for their lives and they have a real shot at surviving by making extreme 3D NAND for SSD. Intel would have a shot of Intel designs are made in their foundry but if the foundry business is split from the design business, the foundry is going to have a tough time surviving.
Pat Gelsinger now saying "yield %" is not a good metric. No kidding. I swear, the world is full of morons. "yield %" means as much as "chance of rain". Chance of rain when? Today? This hour? This week? It matters profoundly. Yield % is only meaningful when die size is also cited. A better metric would be inclusions per wafer. That, of course, would pivot on the process being perfect and flaws arising from dust in the environment. Some process really are that good. Others are not.
