​Conducting business in China has many nuances that would seem foreign to many, but it seems that taking advantage of imbalances in supply and demand is one that is common to almost all regions of the globe.  In some cases it is a function of double ordering or inflating potential targets to increase leverage with producers, while in other cases it is just buying whatever is in short supply and hiking prices until buyer’s eye begin to bleed.  In many cases governments look the other way, but when it comes down to how such unsavory practices affect state or other high profile businesses, the Chinese government has a tendency to take action.
It seems that three Chinese electronics distributors took advantage of the supply/demand imbalance facing vehicle manufacturers, particularly when foundries were also facing COVID-19 staffing issues, earthquakes in Japan, and a blizzard in Texas that exacerbated the imbalance that foundries were already facing from increased semiconductor demand from increasing electric vehicle production.  According to a briefing by the State Administration of Market Supervision, the shortage of chips, particularly those needed for electric vehicles, has reduced such vehicle production in China, and when tracing back the cause for such issues, it was found that the three chip distributors had ‘maliciously snapped up chips in China, hoarding them and driving up prices. Causing the prices of some chips to continue to rise, some by 3 – 10 times, and some by 30 – 40 times, which seriously affected the orderly production and healthy development of the industry.’
In August a task force set out to find why overall auto related chip prices had increased by 10% -15%, with some individual chips increasing as much as 50% and when it was discovered that three chip distributors had purchased certain chips at 10 yuan ($1.55) and sold them at more than 400 yuan ($62.14), with a standard markup being between 7% and 10%, a full investigation was held and each of the three distributors was fined 2.5m yuan ($338,373 US), and were put on a watch list that will continue to be monitored by the state.  Milton Friedman said,” What is greed? Of course, none of us are greedy, it’s only the other fellow who’s greedy. The world runs on individuals pursuing their separate interests”, but it was better said by Veruca in Charlie & the Chocolate Factory, “I want an Oompa-Loompa!  I want you to get me an Oompa Loompa right away!”

Not to be left out of the semiconductor industry’s rush to add capacity that we have been witnessing for the last 6 months, China’s largest foundry SMIC (688981.CH) has signed a Cooperation Framework with the government of Shanghai to build a 12” foundry line that will operate at 28nm and larger nodes.  The project is expected to cost $8.87b US with SMIC contributing at least 51% of the capital and the Shanghai government no more than 25%.  The balance will be raised from 3rd party investors.  There is no start date on the project as this is just a framework rather than a completed agreement, but we assume that the Shanghai government has made inquiries about potential 3rd party investors and is confident that they will be able to raise the additional capital necessary.
At the end of June, according to SEMI, 19 high volume fabs will have started new construction before the end of the year and another 10 in 2022.  15 of the 29 fabs (excluding SMIC) will be foundries, with the largest producing ~220,000 wpm (200mm equivalents).  22 of the 29 fabs will be 300mm lines, with the rest between 100mm and 200mm.  The combined output of the 29 fabs is expected to be ~2.6m wafers/month[1] and spending for 300mm fabs in 2022 is expected to reach $17b, an all-time high, but will outpace 300mm wafer production if no additional wafer capacity is announced this year.  Hopefully new chip production capacity will not outpace demand in 2023 – 2024, but if wafer capacity expansion does not match capacity, it will be a moot point.


[1] 200mm equivalents

And while we are on the subject, documents have revealed that the city of Taylor, Texas, about 25 miles from where Samsung Electronics (005930.KS) has its only semiconductor production fab in the US, has offered the company some very attractive tax breaks to establish its new $17b fab in their quaint town.  Taylor, a town of almost 16,300 residents, has been the site of a number of famous and infamous movies and TV shows including Varsity Blues, Friday Night Lights, and The Texas Chain Saw Massacre, and such notables as Tex Avery, animator and creator of Daffy Duck and Droopy, and actor Rip Torn, and while the town does have a crime rate a bit above the US average, one of the town’s biggest problems is the annual migration of the Cattle Egret, a heron that is noisy can damage vegetation. 

​Competing with Austin, the state capital and city of just under 1m, can be daunting but with Taylor offering Samsung a grant of 92.5% of assessed property tax for 10 years, 90% for the next 10, and 85% for the last 10, and that same 92.5% waiver on any new property built on the site, along with the repayment of any development review costs, and a 1,188 acre site (the existing Samsung site in Austin is 250 acres plus 350 that Samsung recently purchased).  If Taylor were to win the Samsung semiconductor fab bake-off, which also includes sites in Arizona and New York, it would create ~1,800 new jobs with construction starting in 1Q 2022 and production in 2024.

Both DRAM and NAND prices remained flat in August according to Trendforce, after a big jump last month.  On a YTD basis DRAM (DDR4 8Gb 1Gx8) is still up 43.9% and NAND (128Gb 16Gx8) is up 14.5% YTD.  That said recent spot DRAM prices have been declining, at least for some segments, and there seems to be considerable disagreement between chip producers and industry research over the direction for the remainder of the year, with producers expect server DRAM demand to increase in 2H and Trendforce expecting server DRAM pricing to fall.  PC DRAM was also expected to see some price weakness in 2H as the COVID-19 vaccine became more widespread, but recent variant breakouts put that view into question, at least in the near-term.

​The UK’s Competition & Market Authority has issued a report that suggests a Phase 2 investigation into the proposed merger is warranted on competitive grounds.  The CMA stated, “We’re concerned that Nvidia (NVDA) controlling ARM (9984.JP) could create real problems for Nvidia’s rivals by limiting their access to key technologies, and ultimately stifling innovation across a number of important and growing markets.”  The report summary goes a bit deeper in that it uses ARM’s estimation that 70% of the world’s population engages with ARM-based technology, which is a bit like saying 100% of the global population walks on silicon based products (sand).
That said, the report did note that it saw the possibility for SLCs (Substantial Lessening of Competition) should the merger proceed concerning CPUs and GPUs for data centers, network interface controllers, SoC’s for IoT, automotive, autonomous vehicle and driver assistance systems, and most important to the ‘real world’, the supply of SoC’s for gaming consoles.  This covers quite a wide swath of silicon production and while the wording might be a bit melodramatic, the conclusion that the CMA came to was quite specific. 

• The CMA does not consider that the conduct required to address the competition concerns identified can be specified with sufficient clarity, to provide a lasting remedy that is capable of effective monitoring and enforcement.
• This risk is significant in this case, having regard to the complexity and evolving nature of the contracts and markets, the magnitude of the concerns identified, and the breadth and technically specialist nature of the offer.  The CMA found that such a behavioral remedy would carry material specifications, circumvention and monitoring risks.  Therefore the CMA does not believe any form of behavioral remedy would address the competition concerns identified.
• It also does not believe and partial divestment of ARM’s IP business(es) would be sufficiently clear-cut and comprehensive for phase 1. 

Simply, the CMA does not like the merger and recommends that the Secretary of State initiate a deeper investigation into the impact of the proposed merger.  Nvidia says ARM would be kept neutral if the acquisition takes place, but actress Shirley MacLaine said it better, “It’s useless to hold a person to anything he says while he’s in love, drunk, or running for office.”  Just add ‘or making an acquisition…’
 

​On Tuesday we noted that UMC (UMC) had notified customers that it would be instituting another price increase for products developed on its 22nm and 28nm nodes starting in September and November, with another expected in January.  This followed a ~13% price increase at UMC in May, again for their 28nm node.  According to local press in Taiwan, Taiwan Semiconductor (TSM) has also given notice to customers that it will be raising prices between 10% and 20% for 7nm, 12nm, 28nm, and 40nm nodes in 1Q 2022, although some customers have indicated that TSM has already put those increases into place.  TSM increased prices once in 3Q last year and in 2Q this year.  The basis for the increases are said to be to cover the higher operating costs of the new fab being built in the US, although it will not be in production until 2024, though more likely to keep margins at current levels or higher.
Rumors have already surfaced that the price increases will affect the retail price of the iPhone 13 series, whose A15 processor is produced on TSM’s 5nm node.  Until a real teardown is done it will be difficult to define the effects of a TSM price increase as Apple (AAPL) is expected to add a number of photo and video features that could also add to cost, but the iPhone 12, which was the first to use a 5nm processor, saw an estimated cost increase of over $17.  With wafer costs rising this year, even with the higher volumes predicted for the iPhone 13, the BOM will likely see another increase.  We note that TSM’s price increases for 5nm and 7nm are expected to be between 7% and 9% while more mature nodes could top out at ~20%.

​According to Taiwan press, UMC (UMC) has notified customers that it will be increasing prices for product produced on 22nm and 28nm nodes in September and November, with another hike expected in January.  This comes after UMC announced a ~13% price hike in May, again for its 28nm process. Taiwan Semiconductor (TSM) has stated that they will hold 28nm prices steady for the remainder of the year, after raising prices earlier this year.

​Samsung Electronics (005930.KS) is a bit sensitive about its semiconductor IP and process technology and its suppliers are getting a taste of what that can mean if somethings leaks out.  It seems that Samsung has sent a new confidentiality agreement to a number of its component suppliers last month that outlined stricter security measures concerning its more ‘delicate’ technology and potential fines (>$85,000 for each violation).  While there were two ‘levels’ of contracts, the more stringent went as far as asking partners to notify Samsung if an employee leaves the supplier and winds up working for a Samsung competitor within a year.  While this is something usually reserved for Samsung employees, the company is looking to track partner employees if they leave after working on a sensitive Samsung project, and is also asking that partner employees that previously worked for a Samsung competitor be excluded from sensitive Samsung projects.
While these requests might seem a bit overbearing, they are based in a healthy paranoia that Samsung developed after it was discovered that SK Hynix (000660.KS), the 3rd largest semiconductor foundry, discovered that its wafer cleaning tool supplier Mujin Electronics (pvt), from whom it had bought between 350 and 400 such tools, had decided to sell that same tool, which had been developed in conjunction with Samsung Electr0-Mechanics (009150.KS), to a Chinese semiconductor competitor, Changxin Memory Technologies (pvt), which did not sit well with SK Hynix or the Korean courts who jailed the Mujian executives that made the sale.   
SK Hynix was forced to ask other suppliers to develop wafer cleaning tools to replace Mujian, but was forced to continue to buy from Mujian given the long development time for alternative suppliers.  Samsung, SK Hynix’s major competitor, forced Hynix to remain with the vendor who sold their co-developed technology to a competitor, withholding the SEMES tool after the leak, and also ended its distribution contract with Mujian (see our 01-29-21 note for details). 
This obviously made an impact on Samsung, who ended its who seems to be reminding its suppliers that proprietary technology is exactly that, and risking a lucrative contract with the 2nd largest semiconductor foundry by allowing an employee to leak information to a competitor is a big mistake.  While we expect competitors would pay premium prices for a tool that would give them an edge over Samsung, the cost of that sale would likely be far more than a fine of $85,000 and its discovery would end its contract with Samsung.  Perhaps the new confidentiality agreements are a bit over the top, but it is easy to see that Samsung’s concerns are real and that suppliers either comply or risk losing what could be their largest customer.

Semiconductor shortages have been a problem for the CE space for a year, with the situation worsening since the beginning of this year.  Semiconductor fabs are running at full capacity, and while conversations about fabs used to be centered around who would be producing at below 5nm, the focus these days is who has the most available capacity at 28nm and above, as these nodes are the workhorses of the semiconductor industry and produce much of what goes into most CE products.  Application processors for smartphones and similar products get quite a bit of attention as they lead the industry in transistor count and processing metrics, but things like display drivers, FPGAs, and lots of gut level silicon is what makes up most of the CE products we use on a daily basis, and many are really in short supply.
Building out semiconductor capacity takes lots of money and time, so a quick fix for the situation, despite the headlines, is not new fab construction as it will do little to change near-term capacity, and the capital investment cost of building smaller node fabs is easily twice that of building 28nm fabs, where the real capacity is needed.  With all semiconductor buyers competing for limited fab capacity and some of that fab capacity allocated to very specific products, prices continue to rise and while semiconductor foundries are making big plans for adding capacity, there is a Catch-22 that will slow those plans and extend the time it will take to bring that new capacity into the market.

​Based on data accumulated by The Elec, semiconductor equipment suppliers are rapidly increasing the lead times for a wide variety of equipment, not because they cannot produce such tools, but because they do not have the semiconductors they need for the equipment and are on allocation at fabs along with the rest of the industry.  As was the case with automotive semiconductors, shortages caused serious ripples down the supply chain causing automotive assembly plants to be shuttered temporarily until governments began to pressure semiconductor foundries to allocate more capacity to the chips necessary to reopen those plants and put workers back on the lines.  Given that there were other products that produce higher returns for silicon producers, this took a bit of cajoling until a more reasonable balance was found.
Unfortunately, the same seems to be happening to semiconductor equipment suppliers who face shortages of the components they need to produce the semiconductor equipment that will eventually add enough capacity that we won’t have any semiconductor shortages, but such issues for tool vendors have extended equipment lead times by some extraordinary amounts, leading to an extension of the shortage cycle, a perfect Catch-22.   Demand has also contributed to the lead time issue, as semiconductor foundries have been stepping up orders for equipment as they face maxed out utilization rates and have to turn away business, but they do have the option to raise prices for what they are able to produce, which they have.  While equipment suppliers technically have the same option, higher prices make little difference if you cannot deliver product, so the supply chain slows and the crisis remains.
Based on the data we have seen (see below), average semiconductor equipment lead times have increased from between 3 to 6 months last year to ~10 months in the 1st quarter of this year and have continued to increase to between 11.3 months and 12.3 months on average, with some extending out as long as 24 months.  Based on the mid-point of those suppliers that gave a range, the average equipment lead time is 11.8 months, with both US and Japanese companies averaging 12.1 months and 12.2 months respectively.  ASM Pacific (522.HK), a supplier of dicing and bonding tools and Korea’s PSK (319660.KS), a supplier of dry stripping tools, being the lowest at 6 months and 8 months respectively, while ASML (ASML), Ebara (6361.JP), and KLA (KLAC) are the highest among those that were polled, at 19 months (average) and 14 months respectively, with the latter two the same.

​Last month we noted that India had been again working toward attracting a leading LCD producer to build a fab in the country, offering considerable financial incentives, but has had little luck, making it the second attempt over the last few years to develop a display ecosystem in India.  It seems that the Indian government has also been doing the same with semiconductor companies, having instituted a more aggressive policy last year and taking EOI’s starting at the end of last year.  The EOI deadline however has been extended twice and local headlines have indicated that various global semiconductor producers have expressed interest in the plan (and incentives) but the 20 entities having submitted documentation, are mostly financial institutions or private equity firms rather than actual semiconductor producers.
Among the applicants are Foxconn (2354.TT), who has considerable assembly capacity in the country, but not semiconductor production expertise, and Next Orbit Ventures (pvt), an investor group based in Abu Dhabi, but large foundries are also evaluating incentive plans from the US, China, Taiwan, Japan Malaysia and Europe, many of which are more lucrative than the 25% incentive on capex that India is offering, and the Indian government’s highly bureaucratic path toward narrowing the field is a turn-off to foundries that are used to dealing directly with the most senior government officials.  The real problem however is that India has no real semiconductor ecosystem, which is an absolute necessity for the top semiconductor producers and while the Indian government believes they can ‘buy’ their way into the semiconductor business, that approach has failed in the past.
The semiconductor market in India is still small, less than 5% of the global total so the attraction of localized semiconductor production is not present, leaving the only real attraction for large semiconductor companies as India’s commitment to building out an infrastructure that would support a real semiconductor production campus, and while the top few foundries would not want to anger the Indian government with outright refusals, they would likely be more interested in what the Indian government plans to do to develop that infrastructure over the next few years than the incentives offered today.  India would likely be better served by attracting smaller producers using more mature processes in order to develop the suppliers and back-end providers that would eventually attract the primaries but the Indian government seems to be more interested in a big win than a longer term project.  With the increased demand for semiconductors, everyone wants a fab in their neighborhood, and producers will use that advantage to generate the best possible deals.  Taking a shot on a new and unproven country is likely not in the cards.