UMC must have a large back order to clear before their customers leave coz tapeouts are up
All that's left is the UMC 12 Partnership with Intel in US
UMC has no operations in Europe and has little to lose from TSMC Germany. UMC has many fabs in TW, a few in China, a couple in Singapore, and one Japanese fab. As for GF, I doubt TSMC Germany hurts them very much. Just like the TSMC JV fab in Japan, TSMC's JV fab in Germany is serving specific local IDMs and local automakers who want a guaranteed capacity corridor. Reads to me more like existing TSMC customers wanting to shorten their supply chains more than Japanese and German customers ditching their internal or local foundry capacity. Also, TSMC only does bulk, and GF Dresden's bread and butter is SOI. My understanding is that SOI body biasing doesn't really have a bulk equivalent, so you would presumably need a full from the ground up redesign of all IPs and final layout to move a chip that fully utilizes the capabilities of GF's SOI to TSMC's bulk processes. All that effort just to have the "privilege" of worse power consumption and performance on that TSMC process. I doubt clientele who made the deliberate choice to pay a premium for GF's SOI over bulk from SMIC, Samsung, TSMC, or UMC have any interest in TSMC Dresden bulk at an even higher price point than TSMC TW. To make matters even worse, since it is a new fab and TSMC already said they planned to pass all of the added cost of foreign fabs onto their customers, they are presumably passing much of their startup costs onto their customers too (like they do when they start up a new advanced CMOS process). So I wouldn't be shocked if GF's fully deprecated SOI line in a mostly deprecated fab shell is actually able to make and sell wafers for alot cheaper than TSMC's "new" old node fabs even if they run bulk rather than SOI.
Market is just bad (especially in automotive/power/analog/embedded). Just look at TI, Altera, Xilinx, Mobileye, GF, UMC, Analog Devices, Microchip, etc. When even TSMC says things are more challenging than they would like, you know it is ROUGH out there. That embedded/industrial sector is slower to react, given the fact it isn't as close to the end products as say iPhone chips which are just one step removed. A great example is Intel and AMD. Their businesses crashed while their FPGA divisions were having record sales. Their CPU businesses started recovering after inventories were digested and BAM FPGA sales collapse as the oversupply finally made its way through the channels. I think UMC will weather the storm. They seem to have been doing a bit better job at it than GF anyhow. It seems like UMC is the best foundry service provider after TSMC and it seems like they can be a little more flexible than TSMC wants to be with the smaller fish (at least small fish by TSMC standards). As for GF, I am not super worried about them losing their existing business. They at least has designer lock in/differentiated process technology. TI should also crush it. They don't seem to be hurting as much as the others, and they have those fat IDM stacking margins to prop them up better than any fabless our foundry peer. That is before we talk about all of those new fangled 300mm fabs to act as huge margin tailwinds in a couple of years (having much of their capacity at 300mm also being a boon for GF and UMC vs say onsemi, analog devices, Infineon, etc.). Unless these new Chinese 8in fabs are running some specialty substrate (like SiC or InGaAs), they simply can't be cost competitive (even with heavy subsides) versus a 300mm fab (especially vs already deprecated ones like the UMC Fab12's, GF Fabs 1/8, DMOS, RFAB_1, LFAB_1).
Micron Q3 was up 82%. Just don't ask what revenue looked like from late 2022 to late 2023, you really don't want to know
22nm and above is a pretty large sandbox to play in. After all, TI alone has double-digit fabs happily running on 45nm and above. 22nm and above is even in the 30s percent of TSMC's revenue and the majority of their total wafer starts. As for Samsung, I know they licensed SOI, but I don't actually know if they ship much SOI (internally or to foundry clients). STM LOVES their SOI, but they are an IDM not a foundry. STM also uses GF as second source. As for additional cows, they got their new Singapore fab and they are building a new NY fab. Considering how long it took to fill NY and the inability to fill their old China fab before they sold it, I can't exactly say going hog wild in Germany sounds like a great idea without committed customers. Heck, even the 800lb gorilla that is TSMC would have never built their German fab without double-dipping on Government subsidies, and customers sharing construction cost and providing a guaranteed ROI.
Market is just bad (especially in automotive/power/analog/embedded). Just look at TI, Altera, Xilinx, Mobileye, GF, UMC, Analog Devices, Microchip, etc. When even TSMC says things are more challenging than they would like, you know it is ROUGH out there. That embedded/industrial sector is slower to react, given the fact it isn't as close to the end products as say iPhone chips which are just one step removed. A great example is Intel and AMD. Their businesses crashed while their FPGA divisions were having record sales. Their CPU businesses started recovering after inventories were digested and BAM FPGA sales collapse as the oversupply finally made its way through the channels. I think UMC will weather the storm. They seem to have been doing a bit better job at it than GF anyhow. It seems like UMC is the best foundry service provider after TSMC and it seems like they can be a little more flexible than TSMC wants to be with the smaller fish (at least small fish by TSMC standards). As for GF, I am not super worried about them losing their existing business. They at least has designer lock in/differentiated process technology. TI should also crush it. They don't seem to be hurting as much as the others, and they have those fat IDM stacking margins to prop them up better than any fabless our foundry peer. That is before we talk about all of those new fangled 300mm fabs to act as huge margin tailwinds in a couple of years (having much of their capacity at 300mm also being a boon for GF and UMC vs say onsemi, analog devices, Infineon, etc.). Unless these new Chinese 8in fabs are running some specialty substrate (like SiC or InGaAs), they simply can't be cost competitive (even with heavy subsides) versus a 300mm fab (especially vs already deprecated ones like the UMC Fab12's, GF Fabs 1/8, DMOS, RFAB_1, LFAB_1).
Micron Q3 was up 82%. Just don't ask what revenue looked like from late 2022 to late 2023, you really don't want to know
22nm and above is a pretty large sandbox to play in. After all, TI alone has double-digit fabs happily running on 45nm and above. 22nm and above is even in the 30s percent of TSMC's revenue and the majority of their total wafer starts. As for Samsung, I know they licensed SOI, but I don't actually know if they ship much SOI (internally or to foundry clients). STM LOVES their SOI, but they are an IDM not a foundry. STM also uses GF as second source. As for additional cows, they got their new Singapore fab and they are building a new NY fab. Considering how long it took to fill NY and the inability to fill their old China fab before they sold it, I can't exactly say going hog wild in Germany sounds like a great idea without committed customers. Heck, even the 800lb gorilla that is TSMC would have never built their German fab without double-dipping on Government subsidies, and customers sharing construction cost and providing a guaranteed ROI.
