Semi News: Intel 18A is not ready for Foundry customers (preparing for more bad news?), TSMC exit GaN (doesn’t matter), UMC working on 6nm (ready by 2030?), Microsoft AI chips delays
The real point of Reuters’ article on Intel CEO “new” Foundry strategy is, imo, to pre-announce the possibility of a very large write off
TSMC discontinues GaN doesn’t matter, it’s not even 1% of revenue. There’s 2 points 1) TSMC understands that Auto demand is slow, GaN is lower margins and there’s a lot of competition 2) management has better things to do (AI demand)
UMC 6nm rumor, maybe yes but most likely with Intel’s manufacturing after 2030. UMC and Intel first start with 12nm in 2027-28.
Microsoft AI chips delays (confirmation)
Intel 18A could be an temporary immature node, leading to a large write off?
Reuters has its big article on Intel CEO Mr. Tan
The core of the article is not new news. Reuter’s sources Tan confirm that 18A is seen limited or no demand from Foundry customers and Intel has to restart its Foundry efforts with 14A. That means after 2027. This was already mentioned by CFO Mr. Zinsner in the JPM tech conf on 13 May.
“Now the one thing about 18A was it was developed initially as just something for Intel. And we intercepted it relatively early, and it allowed us to develop PDKs for the industry, but it still was not from the grounds up developed as a foundry node. 14A is developed from the ground up as a foundry node”
Reuters mentioning that Intel has a “new strategy for Intel's foundry business would mean offering outside customers a newer generation of technology” is exactly what Zinsner said 1 month ago. Not news.
What is news is this: a big write off could be coming
“to put aside external sales of 18A and its variant 18A-P, manufacturing processes that have cost Intel billions of dollars to develop, the company would have to take a write-off. Industry analysts contacted by Reuters said such a charge could amount to a loss of hundreds of millions, if not billions, of dollars. Intel declined to comment on such "hypothetical scenarios or market speculation."
This is also new: Intel has commitments to make some chips for Amazon and Microsoft
But Intel confirms that there is very little external customers for 18A (again, we know that):
Intel “said the lead customer for 18A has long been Intel itself, and it aims to ramp production of its "Panther Lake" laptop chips later in 2025”
But “Intel will produce a relatively small volume of 18A chips that it has guaranteed for Amazon and Microsoft, with deadlines that make it unrealistic to wait for the development of 14A”.
Side news: glass substrate can be outsourced
That’s from wccftech:
“it is claimed that Intel wants to reduce its operating costs and focus on core businesses, like CPU and manufacturing. Given that the market opens up opportunities for glass substrates, Intel would likely opt for external sourcing, indicating that the firm is determined to trim down its operations by cancelling projects that aren't considered worthwhile”
I guess that’s part of Mr. Tan’s cutting down on R&D spending and pet projects that won’t see market demand within 5 years.
Brace ahead of Intel’s 2Q25 conf call on 24 July.
TSMC out of Power GaN
We’ve learned from Navitas Semiconductor press release that TSMC will wind down its Gallium Nitride GaN wafer foundry services 2027. We learn from Taiwan media that TSMC has disbanded its automotive power GaN R&D team last week. Here we are talking about high voltage 100V to 650V power modules for electric vehicles.
TSMC has an 6-inch GaN capacity at 3-4 wpm and this means that revenues are very small. Definitely less than 1%.
A lot of positives here:
TSMC’s management is busy enough with higher-margins endeavors such as AI and AI packaging and Power GaN is a distraction
This allows TSMC to reshuffle it’s complicated revenue sharing with subsidiary Vanguard International Semiconductor (VIS) by consolidating 6-inch and 8-inch fabs.
TSMC plans to repurpose this GaN Fab 5 for advanced packaging – as CoWoS capacity is a bottleneck. This saves time.
UMC developing a 6nm process? That’s sort of strange but why not
The source for this is a tenuous source of Nikkei, in quite unspecific terms. UMC would be “considering 6nm production” in a partnership with Intel, adding 6nm chips to the ongoing 12nm collaboration in Arizona.
Maybe UMC has worked on a 6nm PDK for communication chips (WiFi, RF, Bluetooth), that’s UMC strong suit but as Nikkei also mentions, UMC CFO Liu Chitung adds that the company would likely adopt an “asset-light” approach, relying on partners for equipment investments.
So this looks like an expansion of the 12nm agreement with Intel. It started in 2024, production is planned for 2027. Hence, if UMC-Intel start working on 6nm now, don’t expect revenues before 2030.
Reiteration: Microsoft Scales Back Ambitions for AI Chips to Overcome Delays
From The Information, paywall