Labor Day does not compare to major retail holidays in the US, like Black Friday or Valentine’s Day, but it still registers in the top 10 in most lists.  That said, while we have checked Samsung’s (005930.KS) recently, as we enter the Labor Day weekend, we thought another quick check might be in order.  In fact we find that while Samsung’s 2024 Mini-LED/QD models saw only modest declines (↓1.0% overall during the last 3 weeks) Samsung’s 2024 OLED models went in the opposite direction, with a 6.1% price increase..  That said, Samsung’s premium 2025 Mini-LED/QD line saw a substantial (↓9.1% ) decline, with all categories hitting new lows, while the 2025 OLED line saw even more aggressive discounting, down 15.6% during the last 3 weeks.  .

We compare the point at which the 2025 Mini-LED and OLED lines are currently aghainst the same point last year and we find that while OLED TV set pricing (aggregate of all models) is essentially the same as last year, aggregate pricing for Samsung’s Mini-LED/QD line (2025) is almost 3% lower than the same point last year.  In 2024 Samsung’s Mini-LED/QD composite TV pricing declined by 11.8% from this point last year (at the lowest point of the year, so we assume that the same will occur with the 2025 line, already 3% lower than last year at this point, for a total decline of ~15% from this point.  So, while the 2025 Mini-LED/QD models look attractive for the Labor Day holiday, if consumers are willing to wait 90 days, the can save another 12% and possibly 15%.  As we track all models and sizes, more specific data is available on request.

With Apple (AAPL) confirming the date of its next event (September 9) the rhetoric about the iPhone 17 line continues to grow.  We see, on a daily basis, “The 5 Improvements Apple will make for the iPhone 17..” or similar headlines with the same leaked or speculated changes that have been detailed hundreds of times over.  Some of these change reports are supported by ‘industry professionals’, or ‘those with knowledge of the matter”, or even ‘conversations with suppliers’, although we expect that Apple has control over most of these sources and uses them strategically to generate excitement leading up to the event. 
Rather than dwell on these changes, most of which seem rather mundane,  we look more toward the effect the iPhone 17 series will have on the smartphone ecosystem.  The Korean research firm UBI has been kind enough to share its thought on the production of displays for Apple smartphones this year and next, which we highlight, rather than dwell on whether the iPhone 17 will come in 3 colors or 4.  As we have previously noted, iPhone displays ae typically produced by three suppliers, Samsung Display (pvt), LG Display (LPL) and BOE (200725.CH).  There have been other in the past (Sharp (6753.JP), Japan Display (6740.JP), etc.) but when it comes to OLED displays, the three named above are the key suppliers.  Unfortunately not all sources agree on the percentages for each, but we have pulled together an amalgam of data that is a composite of the industry estimates for the past six years along with UBI’s projections for this year and next.  We note that due to averaging not all years equal 100%..

​While Samsung Display’s share has been declining, we note that the sales value of each iPhone model are different and in most years SDC was the dominant or sole display supplier of the higher value models while BOE tends to provide the more generic lower value models, skewing the sales value in SDC’s favor each year.  In most years SDC maintains its current year share, while ceding some earlier year production to others.  BOE has had a difficult time competing with its South Korean rivals as it has had qualification issues that have limited its production in the past, particularly on high-end models.   In 2022 it was found that BOE made a change in the design of the iPhone 13 it was producing without Apple’s authorization, for which it was punished with reduced shipment targets that year.  But Apple needs BOE to keep display costs down and rein in Samsung’s premium pricing, and based on UBI’s forecasts, despite the issues, BOE seem to be making progress toward establishing a firmer relationship with Apple, barring an legal issues[1].
One additional point.  Samsung recently won the exclusive contract to supply both the display and the hinge to Apple’s foldable phone in 2026, although LGD and BOE are expected to supply the non-foldable display on the device, although the proportions have yet to be decided.  We expect this will improve SDC’s production share with Apple despite the relatively small number of units to be produced, as the foldable display will carry a significant premium.


[1] Samsung recently won an ITC ruling that could make it more difficult for BOE to supply certain displays in the US.

When 5G was new there was considerable promise for the technology’s two tier structure.  Sub-6 was the lower frequency portion of the 5G spectrum while MMWave was the higher frequency 5G spectrum.  MMWave offered much wider bandwidth and higher speeds but is far more susceptible to signal loss as the distance between source and receiver increases.  This forces operators to deploy repeaters at relatively short distances, especially as compared to cellular 4G, which has limited MMWave use cases relative to Sub-6 5G, despite it higher performance characteristics.

Looking at the characteristic table above, it points to the trade-offs for each frequency band, which also points to their practical applications.  Early 5G networks were low band, with most built on existing 4G towers.  Given the signal distance for low band 5G this was the least expensive way to get a 5G system up and running even though the metrics were only marginally above 4G.  As 5G deployment expanded, Sub-6 Mid band became the point of focus for operators, as it provided a balance between performance and deployment cost, and continues to be the ‘sweet spot’ for 5G.  This leaves MMWave in the position of being the highest performing 5G service, but also the most limited in terms of deployment (distance between sources).  So does this mean operators are just ignoring MMWave?
MMWave is ideal in circumstances where ultra-wide bandwidth is important.  Think of a football stadium with thousands of folks watching an exciting game.  The quarterback breaks away from the pack and heads down field.  What does everyone do?  They grab their phones and record the touchdown, and then they either post it on social media or send it to friends.    College stadiums can hold up to 100,000 spectators, which means there are a lot of phones pulling down bandwidth after that run.  A quick calculation for a 30 second ‘good quality’ 2K video would allow for between 1,000 and 2,400 videos to be sent before Sub-6 5G bandwidth is used up, leaving a lot of spectators watching the circle spin.  MMWave would allow up to 10,000 such videos to be sent every 30 seconds, and while not satisfying everyone in the stadium, its almost 5 times the best Sub-6 metric.
This kind of relatively small space is ideal for MMWave, as the coverage distances are relatively small and repeaters can easily be placed near  spectators.  As it turns out all 30 NFL stadiums have a combination of Sub-6 (Mid) and MMWave coverage. A number of MLB stadiums have similar MMWave coverage including Citi Field (NY) and Progressive Field (Cleveland), along with a number of NBA venues, including MSG in NY and the United Center in Chicago and a number of college stadiums.  Arenas and halls like the Javits Center in NY, the Las Vegas Convention Center, or McCormick Place in Chicago, are all MMWave equipped and able to handle high usage volumes.
But what other environments might benefit from MMWave’s high bandwidth, high speed and low latency?  What about private networks?  Here’s an example:
Bosch (BOSH.IN) in Stuttgart – Bosch decided to team with Nokia (NOK) to create a private network for its 10,000 m2 facility using eight small cells (both MMWave and Sub-6) to cover the entire campus.  The network supports a fleet of autonomous ‘Active Shuttles” on the factory floor that are able to avoid obstacles and people while delivering parts.  The low latency allows  the system to control, in real time, robotic arms and similar automated machinery used in complex assembly, and the system allows for mobile devices to be connected to the network, removing the wired connections that would normally exist at many point on the line. 
At the same time the network collects a constant stream of data from hundreds of machines and sensors which is fed into an edge computing system where an AI algorithm analyzes the data to generate predictive maintenance, process optimization, and quality control, all with the goal of improving production efficiency and reducing equipment failures.  Because the network is private and separate from the public networks, all of the data remains in the system, giving Bosch heightened security control.  None of this could be done with Wi-Fi or public cellular as the ultra-low latency and high bandwidth allows for large amounts of data to be moved with six 9’s reliability, and a wireless system, reconfiguring the factory floor no longer means removing cables and placing new ones to connect to tools, a major cost savings and flexibility improvement.
This is just one example, but the benefits of MMWave in these environments are obvious, without the issues that make MMWave less than ideal for consumer to consumer communications.  Strangely, while North America (Primarily US and Canada) lead in operator investments in MMWave, Europe leads in launches and deployments, with NA only slightly ahead of Asia.  If the US is serious about bringing manufacturing back to the US, MMWave 5G has to be a part of the infrastructure, and adding AI to the mix just increases the need for high speed data movement and massive bandwidth capabilities.  MMWave makes the reality of highly efficient private networks attainable and gives us the ability to compete on a more level playing field with countries that are manpower based.  The technology is here, and the equipment exists, but moving the focus away from consumer oriented 5G products toward MMWave for business is the challenge.
Note: SCMR LLC receives no compensation from any company mentioned in these notes and has no commitment to publicize or comment on any company or product.
​

It seems that according to informed sources in Japan, Sony (SNE) has decided to adopt the latest advancement in LCD technology in a big way.  The system, known as RGB (mini) LED is another push against OLED technology, which has been gaining recognition in IT devices.  Although this technology has been shown at shows over the last few years, it was officially introduced at CES this year, with Hisense (600060.CH) showing a 116” model and Samsung (005930.KS) a 115” model.  TCL (000100.CH), an early supporter of mini-LED/QD backlighting technology is thought to be developing an RGB mini-LED product for consumer release as part of its high-end line.
The purpose of RGB mini-LED backlight technology is to create a brighter set with more vivid colors while still using LCD technology, as opposed to the more expensive OLED process, which is known for its deep blacks and rich colors.  Given the vast LCD infrastructure that has been developed over the last decade, LCD display producers are always looking to find better ways to compete with new or emerging display technologies and RGB LED is that next step.
In a typical mini-LED LCD TV, the backlight is made up of white or blue LEDs arranged in zones. As the light from these LEDs determines the maximum brightness of the set, one would think that the LEDs for such a backlight should be big and bright, but that is not always the case. 4K TV sets have over 8 million pixels, each able to generate a small part of the complete video frame, with their brightness theoretically based on the amount of light behind them (the LED backlight array).  However, when the scene has both dark and light images, large LEDs leak light into dark pixels and turn them gray instead of black.  In order to reduce this effect, mini-LED backlights use small LEDs grouped in zones that can be turned on or off zone by zone.  While the zones don’t line up against the light and dark parts of the image, the fact that there can be many individually controlled zones, improves the light leakage issue.  Further, each zone is comprised of multiple strings of LEDs (strings are a group of LEDs that work as one) that can be controlled string by string. 
Logic holds that as the number of zones increases, the preciseness that the backlights can provide to the display’s millions of pixels improves, but more zones usually means smaller and more LEDs overall, which is why mini-LED sets tend to be more expensive than generic LED backlight sets.  Behind all of these LEDs that are constantly being turned on and off (and many levels between) is the system that controls them.  The basis for this system is the video processor, a specially designed chip that accepts video information before it hits the screen and reformats it to to the zone closest to that pixel set.  Typically video processors look at this information on a frame-by-frame basis, which means they see brightness  information on every pixel 30 times each second.  Just one data point for each pixel would generate ~248 million data points each second so these processors are specially designed to provide a high-speed data stream.  All of that data is passed on to the LED drivers whose channels are connected to the LED springs.  Based on that massive amount of information, the LED zones are lit to try to correspond to the light and dark parts of every frame.
It’s a complex system that moves at millions of bits/second, but it’s all black and white.  As the light from the LEDs passes through the liquid crystal pixels that open or close to create an image, on top of the pixels is a color filter, essentially a sheet of red, green, and blue dots.  Each LCD pixel is made up of three sub-pixels, each of which correspond to a red, green, or blue dot on the color filter.  By adjusting the amount of light that passes through the three LCD sub-pixels, each pixel can create any color.  There is a problem however, in that the color filter is subtractive and reduces the overall brightness of the display as it subtracts two primary colors to create one from white LED light. 
This system is the basis for every LCD TV, with Mini-LED backlight systems adding more LEDs to the mix, but what about the RGB LED backlight systems mentioned earlier?  In order to reduce the subtractive effects of the color filter, display engineers decided that using colored LED backlights would reduce the amount of ‘filtering’ the color filter needed to do to create colors, and RGB LED backlights were born.  But it is not as simple as it sounds.  Now the video processor not only has to figure out what light level each string much have (30 times/sec.) but now has to figure out how to mix the three color strings in each zone to get closest to the color of that particular section of the image frame.  Once it does those calculations for each zone and string for each frame, it sends the data to the drivers that generate the electrical charge.  This means that instead of one white string in a particular zone, there are now three strings (RGB) and one driver now becomes three drivers.
  

​This is where the trade-off of cost vs. quality becomes important.  Driver ICs represent ~20% of the cost of the backlight system, so increasing the number of drivers increases the system cost and while the total number of LEDs can remain the same; in order to maintain the same number of zones in an RGB system, the number of LEDs could triple.  The processor, which now handles considerably more information processing would likely need an upgrade, so the bottom line is that the cost of an RGB LED backlight system for LCD TVs will be higher than those currently used.
Sony does have an advantage over other brands, making it more obvious why they seem attracted to this new technology.  They are the top of the premium TV price tiers and can absorb the higher cost more easily than some.  As of now, with the only available RGB LED sets being over 100”, the additional cost means little, but it will when the technology gets down to smaller size TVs.  Since the point is to compete with OLED, RGB LED TVs need to be at or below OLED premium levels, which will be difficult at the onset, so this new technology will need to prove itself in the quality arena  which is Sony’s wheelhouse.  More than likely Samsung and TCL will have the largest number of RGB LED backlit TV sets in terms of models, while Sony will have the premium (most expensive) sets, once this technology enters mass production (2026), but there has to be a meaningful difference in visual quality to measure up to the higher price, and that will be up to the consumer to decide.  The current price of the Hisense 116” model in the US is $30,000 as is the 115” Samsung model.  A ‘regular’ 115” Mini-LED TV is ~$20,000 and the largest LG (066570.KS) OLED TV (97”) is $18,000.  Sony has a 98” LCD set that sells for ~$6,200 (MSRP).

The US wins this race!  It’s the race to the fastest broadband, although in this case it is not information reported by those who are being tested (carriers) it is actual recorded speeds from the content source to the end-point, meaning not only is your carrier broadband included, but your gateway, and home Wi-Fi are also, making this data far more practical than the iffy data reported by carriers.  While more than 50% of broadband users in all countries listed receive at least 30Mbps, at 100 Mbps things change quickly.  At 30Mbps 86% of users reached that level, but at 100 Mbps that average dropped to a mere 40%, and at 250M bps the average fell to 21%, meaning at 250 Mbps (despite carrier promises) only 21% of broadband users realized that target speed.  The US was the leader in all three speed categories but only 55% of those broadband users in the US who were promised 250 Mbps actually received it.  We don’t blame carriers as much as usual  with this data as home Wi-Fi can do some serious damage to broadband speeds if the equipment is old or set-up improperly.  So if you find you are not close to what you carrier promised (and you are paying for), call them and tell them to pay a visit because they will either fix their equipment or tell you what’s wrong with yours, usually for free.

​AI model builders are not known for their attributions.  Model datasets are notoriously built on content that has been used without permission or without acknowledgement, and model builders have spent millions in court defending their right to ‘Fair use’, but in an unusual turn of events, Perplexity (pvt) has decided to level the playing field a bit.  Perplexity has initiated Comet Plus, a subscription-based program that shares revenue with publishers who decide to participate.  In order to facilitate  the program Perplexity has funded a $42.6m pool using Comet+ subscription fees ($5/month) and will split subscription fees, with 80% going to publishers and 20% to Perplexity.
But…publishers only get their share of the subscription revenue if their content is actively used in a search answer (cited), the site is visited directly through the Comet+ browser, or the content gets used by an AI agent to complete a task.   But…each publisher’s share of the pool depends on how much their share is accessed or cited by Comet+ and by how many times the content is indexed, so it is not quite as simple as it sounds.  That said it certainly is both more transparent than one would expect and addresses one of the big issues that content providers complain about, the loss of direct web traffic and potential ad revenue.  While others (OpenAI (pvt)/Google (GOOG)) pay a lump sum to content providers, the Perplexity model is a usage model that pays content providers based on the value (usage) of their material, a more realistic way to value content.  While it is certainly going to be more complicated to administer than a lump sum payment, it could change the way content providers are compensated and give the courts a way to better judge ‘fair use’.

Before social media and AI, MAUs (Monthly Active Users) were metrics that measured the success of mobile applications, web applications, and online gaming, and even further back, the performance of early internet companies.  Since those days, MAUs have become the yardstick for AI chatbots and similar AI tools, with an emphasis on the need for MAU base growth month after month or quarter after quarter, to sustain ‘leadership’ roles in the AI space.  Much of what we regularly see is MAU data for the US, but less so MAU data from China, the world’s largest market by the number of potential users.  Facebook (FB) has the largest number of MAUs globally (~3bil.), but from a country standpoint Facebook has between 250 and 280m MAUs in the US, while WeChat (700.HK) has between 1.2 and 1.3 billion MAUs in China.  The China/US population ratio is a bit over 4x, so based on population alone, WeChat has a higher country specific MAU than Facebook and gives significant value to MAU data from China.

As some of these companies might be unfamiliar to US investors, we summarize  the primary business of each and provide the closest US equivalent:

• WeChat: A multi-purpose messaging, social media, and mobile payment app.   The closest equivalent (there is really no equivalent in the US) would be a combination of Facebook Messenger, PayPal (PYPL), and Yelp (YELP).
• Taobao (BABA): An e-commerce platform  for C2C.  The closest US equivalent would be eBay.
• Alipay (BABA): A third-party mobile and online payment platform and digital wallet service.  PayPal would be the closest US equivalent.
• Douyin (pvt): A short-form video app.  TikTok (pvt) would be the equivalent..
• Amap (BABA): A digital mapping, navigation, and location-based service provider.  Google Maps would be the US equivalent.
• Baidu (BIDU): A Chinese technology company that primarily provides search engine services. Since it is often called “The Google of China”, Google would be the equivalent.
• Sogou Input (700.HK): A Chinese language input method utility.  This is an application that sits between the keyboard and the WP software that takes Pinyin text (Romanized Chinese) and offers suggestions for Chinese characters.  The closest US equivalent would be Google’s GBoard, which is a predictive utility (suggestions) but not a word processor.
• QQ Browser (700.HK): A web browser developed by Tencent. The equivalent would be Google Chrome.
• Tencent Video (700.HK): A video on-demand streaming website owned by Tencent.  US equivalents would be Hulu (DIS) or Paramount (PARA).
• Huawei Weather (pvt): A weather forecasting and information app provided by Huawei. The US equivalent would be weather.com or Accuweather.com.

 
Other than Huawei’s weather app, all of the top 10 applications with the highest MAUs are owned by either Tencent or Alibaba and e-commerce dominates the pack, along with digital payment applications.  However, with so many of the largest applications owned by two companies, we look toward those Chinese applications that are growing their MAUs, and some of these are growing quite rapidly.  Deepseek (pvt), the Chinese AI that challenged incumbents on cost at the training level did not have a comparative year to generate y/y MAU growth, but would have likely been the fastest growing application followed by AI assistant Doubau (pvt) at over 400% MAU growth, followed again by Honggou (pvt), a short-form video application with MAU growth of over 150%, and CloudPay (pvt), at almost 80% growth.  From there things quiet down to between 25% and 35% MAU growth for a number of service and ‘utility’ apps that continue to gain users.
While we in the US are certainly not immune to trends, fads, and the mania behind on-line  applications, China seems even more so, so when something gets going, the momentum behind such a large population base can be quite strong.  Its worth checking in on user growth on the mainland to see if anything new has moved into the top 10 position and we do it regularly.  While Deepseek will have to wait for a y/y comparison, we expect it will remain at or near the top of the list during the next MAU round.

 

• DeepSeek: A Chinese artificial intelligence company that develops large language models.  US equivalent would be Gemini  (GOOG) or ChatGPT (pvt).
• Doubao: An AI-powered conversational assistant platform that provides a variety of services, including real-time video calls and a video generator.  The equivalent would be a combination of ChatGPT, Jasper.AI (pvt) and any of a number of creative AI tools.
• Hongguo : Short form videos.  US equivalent would be TikTok ().
• CloudPay : A global pay solutions provider focused on platform for payroll, payments, and pay-on-demand services.  US equivalents would be Workday () or Paycom ().
• ChinaMobile (600941.CH): A state-owned Chinese telecommunications company (World's largest mobile network operator).  US equivalents would be T-Mobile (TMUS), Verizon (VZ), and AT&T (T).
• Dianping (3690.HK): An online-to-offline (O2O) platform for consumer reviews and information on merchants, as well as services like group buying and online restaurant reservations. Yelp () would be the US equivalent.
• MiHome: (1810.HK) A Xiaomi application that gives users access to all of their mobile devices.  Apple (AAPL) would be a US equivalent.. .
• Quark (BABA):  Publishing software and content automation. US equivalent would be Adobe (ADBE).
• Didi (pvt): A Chinese mobility company that offers app-based transportation and related services, including ride-hailing, taxi services, and bike-sharing. US equivalent would be Uber ().
• ABC Bank (601288.CH): An international wholesale banking provider that offers corporate banking, trade finance, and treasury products.  US equivalents would be JP Morgan Chase (JPM) or Bank of America (BAC).

Panel pricing was down 0.2% in August, a modest move after July’s 2.4% decline.  TV panels were the standout, with no change for the month, far from our expectations of a 2.0% to 3.0% decline.  It seems that after the 4.7% decline in July for TV panels, brands are using the lower prices to get inventory levels closer to targets, which, as we noted previously, have been lowered for the year.  Seeing some demand strength (or perhaps a lack of pricing pressure), Chinese panel producers are pointing toward raising utilization rates from their recent upper 70’s% to 85% to 90% through September, to try to save 3Q after the weak July, but they run the risk of overbuilding into a 4th quarter that has little clarity for demand as to real demand.
While TV panels showed at least some price resilience, IT panels (Monitors, Notebooks, Tablets) did not, declining 0.1% for the 3rd month in a row.  If you remember, IT products were to be the 2025 ‘saviors’  at the start of the year, yet demand has fizzled out as the complexities of tariffs and other macro issues make consumers wary.  Small panel prices (Mobile) dropped by more than expected (down 2.4% m/m against expectations of down 1%) in August, making the 4th month in a row of negative mobile panel pricing.

While volatility, for the most part, has been low this year, Figure 1shows how actual panel pricing performed thus far, against the 5-year averages.  In January, panel prices grew more than the average, but February, March, and April saw pricing movements below average.  May was better than average, followed by June and July where pricing was below average.  In August and September (forecast), while negative, pricing outperformed the averages.  We note that the 5-year averages for the last three months of the year are negative, indicating weaker pricing potential.
All in, August was a weak pricing month, albeit one where TV panel pricing was flat, but better than expected.  We expect TV panel pricing in September to be flat on brand stocking demand, but concern over an inability to discount enough during the holidays to attract buyers continues to overhang the market.  Chinese panel producers are reacting on a very short-term basis and increasing utilization but run the risk of flooding the market and creating a weak 4th quarter.  They want to show profitability for the year in order to justify both short-term fab purchases and expansions, but are also focused on pointing out how they dominate South Korean producers concerning market share.  It is hard to balance both.

July panel shipments declined by 3.7% m/m but were up 7.4 y/y and on a cumulative basis (7 months), shipments are still up 7.9% for the period.  TV shipments increased 6.7% m/m, the only category to do so.  As we noted above, TV panel prices declined substantially in July, pushing brands to use the opportunity to bring order levels closer to full-year targets, while IT panel shipments saw little momentum as demand remained weak.

In terms of producers, of the major large panel producers, 3 of 8 saw a m/m increase in panel sales while the remainder saw decreases, with LG Display (LPL) the big m/m loser, down 30.6% m/m, after two relatively strong months.  Sharp (6753.JP) saw a m/m increase in sales of 4.9% and HKC (248.HK) saw a 5.0% m/m increase, while all others saw m/m decreases.