If you haven’t seen - I went on YAVB to do a podcast appearance about my recent long piece on SiTime. I would encourage you to go watch it if you found the SiTime thesis interesting but a bit difficult to digest.
Also, this substack is a reader-supported endeavor, and sharing and subscribing makes a meaningful difference to me.
If you’ve found my work helpful in the past, every little bit matters to me. Thank you.
This week I need to talk about quite a few things. First, I want to talk briefly about US export restrictions. I read the documents, started to write about the export controls, and then realized many of my friends and fellow newsletter writers had really good takes, and a 4th or 5th piece would be of little consequence. I instead want to just talk about the impacts and some specific knock-on effects.
Here are some of my favorite overviews and quick “takeaways” from each.
Dylan at Semianalysis had a great overview. My takeaway is that the broadness of the technology is pretty astounding, dating even “old” supercomputers as critical, and the restrictions for 16nm are vague. Some copper interconnects trailing edge technology, so this seems more expansive than direct interference of leading-edge chips. This is economic warfare that is broader in scope and intentionally punitive.
Ben Thompson had a great piece I appreciate the timeline and context of the recent chip bans. This started with ZTE and Huawei, got pushed along by Trump, and now is a bipartisan issue. The unverified list is just a continuation of this saga.
Lastly, I enjoyed this piece by Gregory Allen at CSIS, which framed it specifically in terms of AI and warfare. As much as I’d like to say this is fanciful, I think this is the correct lens. By focusing on AI, we are fighting the war of tomorrow today by crippling China’s ability to keep up.
Okay, so instead of focusing on the minutiae of the unverified list, I want to review some of my speculative first and second-order impacts.
First-Order Impacts:
-
YMTC is screwed. This was as punitive as possible for YMTC to prevent their ability to produce memory chips.
-
Nvidia’s datacenter business will likely never recover that ~400 million in revenue - expect their comps in DC to be much worse in the coming quarters.
-
Most American Semicap revenues in China should go down 70-90%. Given that some lagging edge tools can be sold, it’s unknown if it’ll go to zero, but it’ll be close. Applied did a preannouncement that seems like they will do better than this. (More below)
Second-Order Impacts:
-
Korea is going to get strong-armed into Chip 4 (soon to be Chip 5)
-
The memory cycle might be flipped to undersupply in NAND, given that YMTC capacity shouldn’t be able to grow bits at all
-
WFE spending is going to be terrible next year, given that YMTC is cut off
-
Chinese retaliation likely comes in the form of US consumer goods. Like Apple iPhones sold in China or Nike shoes. But given their broad economic weakness, I think they are not in the position to return a strong body blow.
I want to spend some time on analogies that I think are appropriate. During the South Korea and China spat over THAAD missiles, Chinese consumption of Korean cars went down meaningfully.
Hyundai Motor reported on Apr. 4 that it had sold 56,026 automobiles in the Chinese market the previous month, a 44.3% drop from Mar. 2016. Kia Motor reported 16,006 vehicles sold over the same period, down by 68%. Together, the companies sold 72,032 vehicles, with a decrease of 52.2% from the same period last year.
Another good analog is during the Japanese and Chinese island disputes — where a similar sales decline happened.
Toyota said sales of new vehicles in China dropped 48.9% in September from a year earlier to 44,100 vehicles. Honda said that September sales plunged 40.5% to 33,931 vehicles. China sales for Nissan slid 35% last month to 76,100 vehicles.
The difference this time is that China’s economy is in a weak place as is. They are enacting stimulus packages in the third quarter to counteract their zero-Covid policy. It’s not like there’s a burgeoning consumer demand available to cut off foreign goods. Ask consumers not to spend might not be in the cards.
The real fear is that this forces China’s hand to something much more punitive. Namely invading Taiwan. That scenario is hard to discount, but I feel like the US delivered a solid blow in this round of economic war. It’s unlikely China can return the blow economically, so it might have to resort to something different. Time will tell.
One of the things that surprised me this week was the intense pushback on this tweet. Reminder for everyone: I have a Chinese and end market comp sheet, and Intel has the highest exposure to PCs.
So my reasoning is this is extra bad for Intel. However, the comments and pushback go like this: “Intel foresaw this, and that was last quarter’s guide down. They are ahead of this weakness”. I disagree.
Let’s recount the history of the PC unit shipment expectations so far this year. AMD has always been a bit ahead of Intel in terms of expecting the PC units for this year to be lower. In Q1, they were high single digits, while Intel was in mid-single digits.
In Q2, AMD said:
Our PC outlook, now at mid-teens, would kind of put the market at somewhere around, let's call it, 290 million to 300 million units. So I do think we've appropriately derisked the PC business.
Intel, in their Q2, still is behind AMD’s PC unit expectations.
We now expect the PC TAM to decline roughly 10% in calendar year '22, characterized by broadening consumer weakness and relative strength in enterprise and higher-end SKUs
Meanwhile, AMD pre-announced, and the entire delta between consensus is PC. So I presume this implies PC shipments will be down more than 20% this year compared to their previous mid-teens estimate. Intel’s guide does not account for this and it was “10%” in Q2.
PC is doing horrid, but the knock-on effect is this simple. Who has the largest client business in the world? Intel. I think their revenue miss this next quarter is going to be horrid. Meanwhile, PC shipments are plunging.
I think it is impossible that Intel doesn’t put up a worse than 20% revenue decline in PC. Current estimates are -20% for next quarter, but that was their guide; clearly, things have gotten worse. How this flows through to operating income is unknown, but it’s no surprise that Intel will be doing a meaningful round of layoffs, likely announced during earnings.
Intel will come in below or at the low end of their range and proceed to have one of the ugliest guides we’ve ever seen. Pat already confirmed at Evercore TMT in September they would be at the low end.
I think things have materialized pretty much as we expected, but even a little bit worse. And so we gave a range for our outlook on the call, something Intel never does. We always give you a number. This time, we gave a range, right, given that overall economic uncertainty at the time. And I'll say we're within the range for the quarter and for the year, but trending towards the lower end, right?
Intel is still the world's largest PC company, and since PC shipments are getting worse, we should expect carnage. I do.
TSMC handily beat earnings. Below is a graphic from @ConsensusGurus.
This was not surprising, given they report monthly revenue. The favorable FX translations are part of the revenue beat, as gross margins are massively helped by completely Taiwanese costs.
7nm and 5nm now comprise over 50% of revenue, but 7nm, in particular, looks like a cyclical node as utilization is dropping much faster than 5nm. They still expect a 3nm ramp in the 2H of 2023.
What I thought was much more interesting was how much HPC slowed down and how much smartphones accelerated.
A reminder that this is QoQ revenue, and a meaningful amount of this is Apple, the only company that doesn’t seem to be feeling the global malaise. This is partially seasonal, but the weakness in HPC likely has something to do with Nvidia’s recent weakness.
IoT is the part that makes me wonder. IoT is now the fastest-growing platform and was closer to a single-digit percentage of revenue last year. At this rate, we should expect it to become teens’ revenue next year, yet there are hardly any questions on the call. The call broadly wasn’t that interesting to me — much of the same.
Some other housekeeping items. They didn’t guide CY23 capex but reduced their capex spending plans for 2022 to a more modest 20% YoY growth. I believe that the WFE spending consensus might look closer to just ~10-12% YoY growth when it all shakes out, especially given the rapid cuts by memory companies.
They expect inventory in the supply chain to peak in Q3, reduce in Q4, decrease utilization, and reduce gross margin headwinds from depreciation and N3 ramps to be at least 500 bps of gross margin erosion. Despite all that — they expect the industry to drop and TSMC to grow in 2023. Talk about confidence!
They did get a 1-year license for their 16/28nm factories in Nanjing, but obviously, that’s just a year. How much they can spend in Nanjing on an ongoing basis seems up to debate. Let’s talk about how this impacts Applied materials for a case study of semicap.
Last night Applied revised its fiscal fourth quarter ending October 30, 2022. This is the net impact of the bans, according to them.
On Oct. 7, 2022, the United States government announced new export regulations for U.S. semiconductor technology sold in China, including wafer fabrication equipment and related parts and services. Applied currently estimates that the new regulations will reduce its fourth-quarter net sales by approximately $400 million, plus or minus $150 million.
The truly odd part of the pre-release was this:
Applied is pursuing additional export licenses and authorizations where needed. The company currently expects the new regulations will impact net sales in the first quarter of fiscal 2023 by a similar amount as in the current quarter. Additional information will be provided during the company’s next earnings webcast.
Last quarter Chinese revenue at Applied was ~27% of revenue, so the initial estimate of $400 million, given that we are deep into the quarter, makes sense. Below is my proforma.
The weird part is the “similar amount” for the next quarter because, given their revenue, it should impact them to the tune of ~20%, yet they expect a similar amount. If I had to guess, that’s partially driven by the temporary licenses for companies like TSMC, SK Hynix, and Samsung fabs in China will likely continue. The implied net here is that only 30% of revenue will be outright blocked in the 1Q23 timeframe. This seems massively bullish, given historical public comments of 50% of revenue in China being Chinese domestic companies.
Another thought process is that these backlogs are fungible and can be re-diverted elsewhere to other companies hungry for shipments sooner. It’s hard to know more than that, given all the information we are given, but that’s a helpful upper bound for how much Chinese WFE gets 0’d out in the coming periods. ~30% of their Chinese business.
Anyways back to WFE.
There are so many pieces here that are moving, it’s pretty tough to estimate and know what WFE will look like for 2022, let alone 2023. I want to note that Micron’s drastic capex cut looks like the outlier now. Nanya reported and moved capex meaningfully down for the CY 2022 and CY 2023 numbers.
But the 2023 number looks less than expected, down 20% YoY.
My guess (seriously, this is a guess) is that total WFE for 2022 looks like 12% YoY growth, driven by ~20%+ in foundry and logic and mid-single-digit increases in Memory. That is mostly driven by the intense cuts that are happening in Memory right now. I would say that’s ~90-92 billion. The hard part is where things end up next year.
I haven’t done the full analysis quite yet, I am waiting on the Semicap and other semi-players to report before I take a real crack at 2023. The biggest unknown is China, which could move WFE much lower. But my guess is that the new consensus is in the high 60s range. Somewhere between $65-72 billion. I’m going to steal Dylan’s estimate - I think there’s nothing wrong with it quite yet, and I haven’t pried deeper. (Below is from Semianalysis)
I believe that puts most of the semicap names at mid to high teens P/E. The problem here is now the next leg is how much of WFE is China (my estimate is ~10-15% of global WFE). Is that going to zero instantly? Time will tell - the earnings calls will be the great color this quarter.
Stay sane, my friends. It’s a topsy-turvy time in semiconductor land. I’m looking forward to earnings season in earnest. As a reminder, I never put earnings content in front of my paywall, so if you want to keep up to date, I think you should subscribe. :)
Talk soon. Next week is ASML, LRCX, BESI, ENTG, Nordic, and more earnings. ASML is usually the first semicap company and sets the tone for the quarter. My guess is that the supply chain massively improved (AMAT pre-announce) but against that are the bans. Where that nets us is going to be interesting.
from Hacker News https://ift.tt/I0rO415
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.