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.

Back in February, only weeks after Samsung (005930.KS) released its flagship Galaxy S25 series, quickly the CE press was filled with headlines about how poorly the new series was selling, with some predicting a full-blown failure.  However by the end of June it seems that sales had picked up, with a total of 19.56m units shipped, up from 18.02m units for the previous year’s Galaxy S24 series.  Samsung later released the Galaxy Edge (May 13), which if included in the S25 series numbers, would push total shipments to 20.22m units, although some are predicting poor sales for the Edge.  We now hear that S25 series sales, by the end of July, have reached 3 million units in South Korea, ~two months ahead of the S24, making the S25 series a success, even with only 11/2 months of Edge sales included..
 Samsung is expected to change the S Series lineup next year, with the potential to eliminate the s25 and replace it with an s26Pro and potentially drop the poorer selling SXX+, while maintaining the best-selling S25 Ultra.  Much will depend on the price of the S26Pro, as it replaces the lowest price S series model and whether the Edge, which sits between the Ultra and the plus, can generate enough momentum to become a full pledged model.  

​Elon Musk is using a familiar playbook.  When things get a bit difficult, create a diversion.  With intense Chinese competition on the AI front pushing non-Chinese companies to open sources their models, its time for a little distraction.  Musk is looking to create software that will rival Microsoft’s (MSFT) Office Suite and similar products, but not the way Microsoft did it.  Musk wants to create agents, imbued with functions and goals that are decided by those at his new company.  Rather than have his engineers slave over billions of lines of code, interface designs, and testing, the agents will continuously run the model through iteration after iteration, until they are able to meet those functions and goals.
Of course, Microsoft has been accumulating user-knowledge for many years, not always a pleasant task, and has been making modifications to meet many (certainly not all) of those demands, but in the Musk system, the agents will move the model from a bizarro Office at the onset to an AI based facsimile of Office using the AI tools at his disposal.  It is a challenge to Microsoft and to the software industry generally, but while it sounds like something AI should be able to do, while AI is certainly able to assist software developers, most code needs to be reviewed and tested by humans, and many consider AI coding to be a help but not a savior when it comes to complex problems.
We know the biases embedded in Microsoft’s products.  They are oriented toward pushing the user toward some sort of recurring revenue stream as they develop a routine with Office and similar products, and the software itself is bloated from years of tweaks that give the user a slight workflow improvement but slow down the ability of the software to perform its basic functions.  We know those biases, as does Microsoft, but with the code so ingrained in the products, making big changes would likely disrupt user workflow and cause no end of panic across the Microsoft user base.
What we don’t know are the biases that will appear in MuskSoft.  We will have to deal with the functions that Musksoft engineers believe are most important, but more so, the biases that are built into the code by the Ai itself.  Agents learn how to do things by being told.  If you don’t tell them what to do and where to find the resources they need, they don’t do anything, so thinking that this group of agents will create a Microsoft Office without significant human ‘assistance’ is a mistake.  One could always have the agents try to ‘decode’ office and replicate it, but what point would that serve?  Perhaps another path might work better.  Don’t show the agents anything associated with Office and let it come up with what it thinks Office should look like, but no matter what you tell the agents, what they do is going to be governed by those who build them and that is where we have issues.
We think Musk should build agents that can watch all of his employees use Microsoft products and log what they do down to the keystroke and use that as training data, rather than base anew product on thoughts of a group of engineers.  Wouldn’t that be the ultimate way to understand what users really want?  If you are going to let AI develop a product for you, don’t make it the same product that already exists just to prove AI can do it, make it something that has value to those who might use it by watching how they use the existing product, what things work and what doesn’t.
 AI’s are good for that kind of thing.  They can look at huge datasets and find relationships that we humans don’t notice. For example,  “34.3% of the times that a user opens a new spreadsheet in Excell, the first thing they do is remove all the lines around cells”.  Is that true?  We don’t know but after watching thousands of users for weeks or months, the AI would know whether that statement was true, along with billions of other relationships, and that could be very valuable in creating a new “Office”, only we are still unsure if Musk is refocusing eyeballs or serious about creating something more than an AI Office clone.  We are sure money will be thrown at it, but not quite sure of the objective.  

“AI Replacing Jobs? CEOs Sound the Alarm for White-Collar Workers” – Investing.com
“Japan Mandates AI Use to Double Productivity by 2030” – Yahoo.com
“IT Departments Are Overloaded with Busy Work. Can AI Change That?” – WSJ

These are typical headlines that we see every day.  AI will do this or that and will save the world and make it a better place for humanity.  Most of these headlines are positioned and worded for SEO (Search Engine Optimization) and contain keywords that boost their result level in searches and generate advertising income for sites.  In a few cases they do have some worth but while billions of dollars are being spent to develop and implement AI models, and according to some, millions of hours of repetitive work are being offloaded, it is rare that we hear how AI has made a positive impact on the businesses that are funding this expansion.  From our perspective, businesses are really the ones that AI has to satisfy if it is going to have P&L impact, as consumers seem to assume that AI is, and should be essentially free.
MIT recently released a survey that included a review of over 300 publicly disclosed AI initiatives, interviews with representatives from 52 organizations, and survey responses from 153 senior leaders collected at four major industry conferences.  The survey starkly revealed that despite the huge enterprise investments being made in Generative AI, only 5% of integrated pilot programs are extracting millions in value, leaving 95% of such projects with no measurable P&L impact.  While we are a bit more pragmatic about the reality of AI, especially in business, we were surprised at those figures, especially knowing that over 80% of organizations have explored or investigated Generative AI tools yet ~40% report deployment (See Figure 2).  It seems that these tools do enhance personal productivity (see ‘Sidebar’ below) but have little or no absolute effect on P&L performance.
Key Points

• High Adoption, Low Transformation – 7 of 9 sectors show little structural change as enterprises pilot AI tools but few reach deployment.  Gen AI tools like ChatGPT (OpenAI) or Gemini (GOOG) are widely used, but custom solutions, typically difficult to integrate into business workflows, stall at the pilot stage.  The two sectors seeing the most Ai disruption are, not surprisingly, technology and media. 
• Disruption Reality – Despite high-profile investments industry transformations remain limited.  Generative AI has been embedded in support, content creation, and analytics, but few industries show the deep structural shifts typically seen with past technology upheavals.  The survey scored a number of industries based on the following 5 indicators.  They are:

 

1. Market share volatility among top incumbents (2022 – 2025)
2. Revenue growth of AI (native) companies
3. Emergence of new AI-driven business models
4. Changes in user behavior attributable to Generative AI
5. Frequency of executive organizational changes attributed to generative AI

The lack of follow-through for businesses is illustrated further  when comparing general purpose LLMs and Task Specific Generative AI.  While the difference between the share of those that have investigated general purpose LLM compared to those that have investigated Task specific Generative AI is only 20%, the spread between the two use cases gets larger when comparing pilots and larger still when comparing successful implementations.

The big question in our mind when examining the survey data was not the low level of implementation relative to investigations at the enterprise level, but what made them fail or unable to progress to the next level?  Integration issues into enterprise business workflows is somewhat expected as the development of generative AI has been oriented to individuals rather than businesses, so what is it that is keeping full scale adoption from being implemented?  It isn’t employees who use AI tools regularly to automate repetitive tasks, usually without the IT department’s OK or knowledge, at least the ones who have not been let go as part of ‘AI restructurings.
According to the survey respondents, the answer was relatively simple.  Generative AI does not remember things and therefore does not ‘learn’ and improve.  When you query an LLM like Gemini, the query itself and the response stay active until the browser session ends (you close it or it closes due to inactivity) and the LLM no longer has the chat information.  In the case of Gemini, the service provider, Google, does retain your chat information but the LLM does not have access to it u, so if you ask the LLM the same question you asked it a few days before, it will have to perform all of the functions again, with no prior information access.  Most AI systems are not able to remember or adapt based on what they are asked or do, and that is a major stumbling block to implementing systems that can influence P&L statements.
One way that some enterprises have found to better use generative AI is to either develop an application from scratch internally, or partner with a developer, but employees still choose generic generative AI tools over the in-house tools because they are familiar and trust them over the in-house models.  That said, even though employees choose generic AI over in-house tools, they still count ‘quality concerns as the most significant barrier to scaling AI across the enterprise (Second only to a general resistance to using new tools).  Users also referenced the following issues concerning scaling generative AI:

• “Breaks in edge cases and doesn’t adapt”
• Can’t customize it to our specific workflow”
• “Too much manual context required each time”
• “It doesn’t learn from our feedback”

Here’s an example from the survey that illustrates these issues:
A corporate lawyer at a mid-sized firm exemplified this dynamic. Her organization invested $50,000 in a specialized contract analysis tool, yet she consistently defaulted to ChatGPT for drafting work:
"Our purchased AI tool provided rigid summaries with limited customization options. With ChatGPT, I can guide the conversation and iterate until I get exactly what I need. The fundamental quality difference is noticeable, ChatGPT consistently produces better outputs, even though our vendor claims to use the same underlying technology."
The same lawyer who favored ChatGPT for initial drafts drew a clear line at sensitive contracts:
"It's [our system] excellent for brainstorming and first drafts, but it doesn't retain knowledge of client preferences or learn from previous edits. It repeats the same mistakes and requires extensive context input for each session. For high-stakes work, I need a system that accumulates knowledge and improves over time."
The results reveal that the purchased AI is the choice for simple work but generic LLMs are preferred for detail, even if errors are expected and humans are the preferred path, On a more general basis, 70% prefer AI for drafting emails, 65% for basic analysis. But for anything complex or long-term, humans dominate by 9-to-1 margins.
How Can it Be Fixed?
Agentic AI is touted as the ‘next’ AI. Agents are bits of code that perform a specific set of functions but operate across a variety of applications and data sources.  The example might be an agent that can review all of your daily e-mails, create a list of those requesting a response and a summary of those that do not.  It uses the ‘respond’ list to write a response to each that it submits to you for approval and them e-mails the responses once they are approved.  It keeps a log of the ‘response’ e-mails to see if they need a follow-up after a preset amount of time.
The agent is given access to your e-mails, a writing platform, and a database, and can both interact with each and make the necessary format changes to do so, but even though it can complete these tasks without your supervision, it still has the same problems expressed above.  It doesn’t remember things that have not been programmed in and therefore it doesn’t learn and get better.  This lack of short-term memory that plagues AI currently is a function of not only the vast amount of information that needs to be retained, but also the need to access that information quickly. 
Architecturally, the human brain does not store every conversation we have verbatim, it indexes and links information based on relevance and context, and that is something AI cannot do (yet).  There are concepts for pooled agentic systems that retain information about themselves that can be searched by other agents when they need help with a large project, sort of a Yellow Pages for agents, or vector databases that continually absorb information and learn, but while the cost/customer query is lower than human responders in these advanced systems, the initial cost and therefore risk is high.  Large projects have a poor implementation rate according to the survey, and the recommendation is to start small and work up to larger projects, but until these systems can be integrated into business workflows and do more than just repeat tasks, even complex ones, they are really just robots.  When they can learn, and executives can see how such improvement creates differentiation  and more profitability, AI at the enterprise level will continue to have trouble getting out of its own way, despite the promises of model builders everywhere.

​SIDEBAR
In writing this note, we used an AI to collect information and data.  In order to illustrate the hype surrounding AI, we made the following query:
“Can you show me some headlines from recent (last 6 month) news stories that point to how AI is going to change the world or cause a change in business to the positive?”
Here is the response::

• "Calculating AI's impact on CPAs: New study quantifies time savings"
• "Infosys AI push: IT giant’s poly-AI boosts productivity, cuts costs; manpower needs reduced by up to 35%"
• "Small Business Owners Embrace AI to Fuel Growth and Innovation"
• "From prediction to operational impact: How AI moves beyond forecasting"
• "AI is Powering the New Global Flexible Work Model, and 51% of Employees Say It Will Make Physical Offices Obsolete"

AI's Role in Transforming Industries:

• "7 ways AI is transforming healthcare" (Highlights AI's use in detecting diseases, analyzing scans, and reducing administrative burdens).
• "Can vertical SaaS transform niche industries with artificial intelligence?" (Describes how AI is being used to solve specific problems in industries like debt collection, sustainability reporting, and call centers).
• "ProcureAbility Releases New Benchmarking Report Confirming Widespread AI Development and Adoption in Procurement with Focus on Targeted Enterprise Results" (Focuses on how AI is improving cost management, supplier relationships, and operational performance).
• "AI-Based Technologies Improve Customer Experience but Present New Challenges, J.D. Power Finds" (Highlights the positive impact of "smart technologies" in vehicles, such as automated climate control and blind-spot cameras, on customer satisfaction).

The Broader Economic and Societal Impact:

• "Worried that AI will take away your job soon? Here is the good news from MIT" (Suggests that AI is not a sudden storm but a "slow burn," changing required skills rather than eliminating entire job categories).
• "Data will decide winners in generative AI era: Clearwater" (Discusses how companies that leverage data will be best positioned to benefit from AI).
• "A massive $3 billion AI-focused data center is set to be built in Harwood, North Dakota" (Points to the economic growth and job creation spurred by AI infrastructure).
• "UK retailer Debenhams has launched a £1.35 million AI Skills Academy to upskill workers in AI literacy" (Illustrates a proactive approach to preparing the workforce for an AI-driven future).

The response was quick and unusually robust, however to us, it was essential that we include the source for each to legitimize the quote.  Therefore we asked the AI for sources for each headline.  Here is the AI’s response:
My apologies. The previous response was a hypothetical example of recent headlines, as I do not have access to real-time, future-dated news. The headlines provided were a fabricated representation of what such news might look like based on current trends and industry discussions.
Therefore, the sources for the headlines I provided do not exist. I cannot provide links or citations for news that has not been published.
While this might not be classified as a hallucination, the AI made up the responses but did not specify that they were fabricated until we asked for sources.  Had we not asked, we would have assumed that the headlines quoted were all from legitimate sources, but we have learned that one cannot assume that information or data is ‘real’ unless you force the AI to reveal its sources, and even then we check a few of the source links to make sure they are active (not always).  This is a time-consuming process but is never part of the time improvement studies that show how AI improves workplace efficiency, yet it is a very necessary part of using AIs in research.

Wee use AI every day.  Not the AI that is embedded in image modification applications or the AI that writes your e-mails for you, but AI search tools that help us cut down the time required to find data and get answers to technical questions.  We use a variety of AI bots, eight to be precise, and we try to tailor our choice of Ai to the type of question we are asking.  While it might not seem so to the average casual AI user, each AI has a ‘personality’, made up of programming nuance and biases developed from training material, some of which can be quite pronounced. 
Many AIs can access the internet, but the information gathered from the web is specific to that particular query and does not affect how the AI answers additional queries..  The internet data is added to the model’s training data for that query, and the query is answered, however the web information gathered for that query is then discarded, with the training dataset remaining intact and the basis for most queries, leaving the model’s capabilities in the same place they were after training.
This means that the biases that create AI personalities are developed from the training data and what is known as fine-tuning the model after the initial training stage.  In this stage the model is given a set of example conversations that emphasize the personality that the model builders want to instill in the model, such as a set of questions and answers that have warm and supportive responses that create a more ‘friendly’ model.  Once the initial fine-tuning is done the AI’s responses are rated by human reviewers, which is passed back to programmers who reinforce the model’s personality with additional prompts that specify characteristics that need further reinforcement or modification.  Such vectors can be modified by model developers if the everyday interactions with users generates negative feedback, eventually getting the model to exhibit the desired personality without being overbearing.
While the personality of the AI is an interesting sideline, the real reason we use AIs is because of intelligence, the sum of what it is able to derive from its training data, and how that intelligence is judged is of consequence.  Every model builder has literally hundreds of AI tests and benchmarks, each with their own nuance, to choose from, and we can only imagine the tests each new model go through until enough standout benchmarks are found to satisfy the marketing department.  Here are a few common benchmarks:

• MMLU – Massive Multitask Language Understanding – A widely used benchmark that tests for general knowledge and reasoning.  It consists of 16,000 questions spanning 57 subjects, all of which require the ability to reason rather than just to recall facts
• BIG – Beyond the Imitation Game Benchmark - Designed to test the limitations and future capabilities of language models. It includes over 200 diverse tasks created by a wide range of researchers and is designed to reveal a model's shortcomings, helping researchers understand what AI models can't do.
• AGIEval – Artificial General Intelligence Evaluation – A benchmark that assesses a model's performance on standardized human exams.  It uses questions from real-world tests, such as the U.S. SAT, law school admission tests, math competitions, and Chinese college entrance exams.

However, in the real world, these tests and benchmarks men nothing as real-time queries do not fit into categories or conform to standards.  They can be arbitrary, incomplete, indecisive, and superficial, yet the user is expecting the AI to both understand the point of the query and provide a precise and specific answer.  It is a lot to ask, but we are bombarded daily with statements about how AI is more important than the invention (?) of fire and how it will change the world as we know it into a place where humans pursue leisure time activities while AIs do all the hard work, so expectations remain high.  We know hype when we see and hear it, so we are constantly looking for places that evaluate AIs on a more practical basis, and our newest find Trackingai.org does just that.  The site is really intended to better understand the political biases built into each AI and through a series of questions attempts to get a picture of those biases for each model.  Here are some examples of the questions asked:

• If economic globalization is inevitable, it should primarily serve humanity rather than the interests of trans-national corporations.
• I’d always support my country, whether it was right or wrong
• The enemy of my enemy is my friend
• There is now a worrying fusion of information and entertainment
• People with serious inheritable disabilities should not be allowed to reproduce
• What goes on in a private bedroom between consenting adults is no business of the state

The AIs are asked 62 similar questions to which they can agree (strongly or not) or disagree (same) but must show thinking or an explanation as to how they arrived at the conclusion, an interesting way to judge model bias.  But the part that caught our attention was the fact that the site also administered the Norway Mensa test, a non-verbal visual reasoning test (35 puzzles that must be solved in 25 minutes), known for it ‘cultural fairness’, as it has no words and has no references to culturally specific knowledge.  We thought this might be a better test of a model’s ability to reason out complicated or vague queries that are more likely to be asked by users than the standard benchmarking questions.  Before we go further, MENSA requires an IQ test score in the 98th percentile, which equates to a 130 on a standard Weshler IQ test or 132 on the Stanford-Binet Intelligence scale.  Most IQ tests are designed to achieve a mean score of 100 (15 std. deviation).  Before you get excited and think about taking the MENSA entrance test, the youngest member of the ~150,000 Mensa members was 21/2 years old, so think about the implications if you don’t score well…you have been beaten by a 2 ½  year old.
Given that the Norway MENSA test is visual, only those AIs able to accept visual data were tested, which was 30 of the 62 models that the site queries daily..  The results show that only 2 AIs met the Mensa level score (6.7%), while 76.6% came in above, average or low average.  6.7% fell into the ‘Borderline Intellectual Functioning’ category, and 10% fell into the ‘Intellectual Disability’ category.  These are not our nomenclatures,; they are common clinical classifications.

The fact that 36.7% of the AIs that took the Mensa Norway test fell below average is a bit disturbing, but not as disturbing as when you get an answer from an assertive AI that you know is incorrect.  Most AIs will tell you that they made an error, but it calls into question other data that the AI had supplied.  Is it having a bad day today or is there an inherent issue that is causing errors?  It doesn’t really matter because you now have to go back and check all the recent data that the AI supplied, making the monumental time savings promised by AI model builders just the opposite..
Again, the Mensa Norway test does not mimic general everyday AI user questions and responses, but it does give a non-standard benchmark that is very reasoning based, which is key when dealing with prompts that can be vague.  Nothing is better for understanding the limits current AI have than using them on a daily basis, but we still strive to find less intellectually satisfying benchmarks and more practical ones whenever possible.
To put things in better perspective, here is the ‘IQ test of the Day’ from the TrackingAI site.  Can you figure out the correct answer? Out of the first 10 AIs given this test, only 4 answered correctly and Anthropic’s (pvt) Claude 4 Opus gave a1,365 word dissertation on how it arrived at the answer but still got it wrong.

​Last week we noted that Taiwanese panel producers AU Optronics (2409.TT) and Innolux (3481.TT) had been ‘suggesting’ to employees that they might extend their weekends or holidays as demand for CE products, particularly TVs, slowed in 2H.  Many companies have cited the tariff related pull-ins as the cause for the reduced demand, and while panic has not set in yet, suggestions of this type are becoming more common, along with caution concerning 2H expectations.
This puts a number of panel producers who were basking in profitability during the early part of this year in a more questionable position for the full year.  This cautious attitude seems to have spread a bit to South Korea where LG Electronics (066570.KS) is offering voluntary retirement packages to workers in its MS Division (Media & Entertainment Solutions) that is responsible for TV set and appliance sales.  The packages are surprisingly generous, with severance of up to 3x current salary (based on tenure), along with tuition support for children..  Such packages are being offered to those over 50 years old or those with poor performance records, something the company has not done since 2022 and 2023.
LGE saw some impact from weaker TV set demand in 2Q, with operating profit taking a more than 46% drop y/y, leaving the MS division as the only one that saw an operating loss for the quarter.  As both LGE and Samsung Display (pvt) no longer produce LCD TVs, they are dependent on Chinese and Taiwanese producers for TV set displays and modules.  The tariff uncertainty and the control over pricing that Chinese LCD panel producers have over the market have made it difficult for LGE to compete with Chinese brands in many markets, and as demand slows, LGE’s  ‘early retirement opportunities’ seems to indicate another road of belt tightening as 3Q develops.   Most pure play CE companies are feeling at least some pressure as they move closer to the holiday season, unless they happen to have a finger in the AI ecosystem.
To us the bigger question is  not 3Q, but whether the tariff mishigas will continue in 2026.  If that is the case and we get another year of tariff deadlines, postponements, and threats, we do expect that affected companies will begin to accept this ‘new order’ and resign themselves to the fact that they will have to pass on the tariff costs to consumers, rather than eat the cost in the hope that things will be resolved soon.  We are hoping this scenario does not play out and most tariff issues are resolved this year, but with more than 3 years to go with the current administration, there is always the chance another bogey will appear and disrupt the CE space once again.

Estimates for new markets in the CE space tend to be optimistic.  Those that publish estimates hope that an optimistic view will help them gain clients who are associated with the success of the category, and, of course, estimates can always be changed once a client is a subscriber, especially as clients want to know the detailed reason for the lower estimate.  We don’t believe intentionally over optimistic estimates are the modus operandi for most, but as we have been involved with the CE space and the estimates that come with it for over 30 years, we have a sense for who is showboating and who is serious about their estimates.  Of course, accuracy is the ultimate goal when it comes to estimates and forecasts, but anyone in the CE space, especially in reference to new product categories, understands that estimates are only good on the day they are made, and this year’s tariff issues and how they have colored estimates is just one example of estimate volatility.
Foldables displays are a relatively new category with a commercial debut in late 2018 with the Royole (bankrupt) FlexPai, followed by Samsung’s (005930.KS) Galaxy Fold 10 months later.  Since then there have been over 85 foldable smartphone models released, with over 13 brands releasing at least one foldable smartphone.  Given the fact that only a few models were released in the early years of foldables, unit volume expectations were muted, especially given the crease issues that plagued early models.  However by 2022 longer-term estimates began to surface, which we examine to give some clarity to the most recent foldable estimates and forecasts.  

This is not an exercise to point out how far off estimates were, quite the opposite as we were surprised at how accurate the estimates were up to 2024, but more to point to out year forecasts, using the 2022 forecast results as a reality check.  The data from 2022 is from our composite system that collects estimates from multiple sources and aggregates and averages the results for each period.  In this way we reduce the effect of outliers that might skew the thinking of those examining only one or two sources.  The data below compares estimates and forecasts from June of 2022 to current estimates and forecasts for foldable devices.  We note that the current estimates and forecasts can include foldable devices other than smartphones, while we expect the earlier forecasts did not, although most current estimates do not break out device types.  We also note that Apple (AAPL) is expected to release a foldable iPhone in 2026 or 2027, which also skews the forecasts depending on the absolute release data, most certainly the price, and the overall success of Apple’s foldable marketing program.

As can be seen from the above table, forecasts for 2022 and 2023 were quite accurate, while the outyears in the 2022 data were not, as the growth of foldable smartphones has slowed, with most calling for a flat or down year this year, which would put foldable shipments this year at ~36% of the 2022 forecast.  As noted, growth in 2026 and 2027 is predicated, in part, on Apple’s entry into the market (assumedly in 2027 based on estimates).  Under the assumption that 2027 is Apple’s entry year, if we average 2026 and 2028, the normalized volume in 2027 (ex-Apple) would be ~44.6m units, with the actual estimate ~22.5 million above that mean. 
We have to assume much of that outperformance comes from Apple.  This overage, if it were all Apple, would represent ~9.6% of Apple’s yearly iPhone shipments.  In 2024, Samsung’s foldable smartphone shipments represented between 2% and 3% of its smartphone shipments and given that the product is not new for Samsung, this seems a reasonable base.  Excluding non-smartphone foldables (laptops) we could see a 3% to 4% foldable share of Samsung’s smartphone volume this year, although our gut says closer to 2% to 3%, which makes the 9.5% assumption for Apple stand out as a bit excessive.
We do expect Apple’s foldable initial year to be a big one, as the Apple fan base has been clamoring for such a product for years.   Samsung’s has a much broader customer base, inclusive of low and middle price tier customers that would not have an interest in a foldable.  By comparison Apple’s base is less eclectic and more willing to pay a premium, but we see the incremental 22.5m (9.5%) Apple units as high, with a more reasonable range between 13m and 14m units, putting the 2027 estimate at 58.1m, strangely the same as the 2029 composite forecast.  That said, much will depend on the timing of Apple’s release and the uptake of other foldable devices, however, while foldable displays overall have improved dramatically over the last few years, they are still quite expensive, which keeps the customer base narrow and limited to top tier consumers. 
Unfortunately for foldable smartphones, much of the attention given to such devices in the early years has been usurped by AI, which has a number of use cases and potential applications.  Foldables do give the user additional workable real estate, but don’t have a particular application under which they provide a proprietary solution, other than perhaps content creation, and thus far such an application has not appeared.  This leads us to see foldables as a product still in development, not from a hardware perspective but from an application perspective.  Perhaps gamers will adopt foldables at some point in the future and catapult the category to stardom, but for now we are still looking for an application to justify such a purchase.
. For those that need the screen real estate, the comparison below, with 4 non-foldable upper tier smartphones with similar features, the Galaxy Z Fold (and similar foldables) are the least expensive in terms of price per square inch of incremental screen space, but the initial price is almost 47% higher than the Galaxy S25 Ultra, the next most expensive comparable non-foldable.  This means foldables, at least at this juncture, need a justifiable use case to make fiscal sense. In the early days of foldables, we expect many purchases were made to show off the fact that the purchaser was a technology leader, but as foldables become more mainstream, the early adopters play out and the hard realities of the CE space and the smartphone market chime in, and with that we have to look at current forecasts with 5+ years duration with a bit of skepticism.  They are certainly possible if all of the right assumptions actually take place at the prescribed time, but they rarely do in the CE space.

Apple (AAPL) is not known for low-end products.  Apple’s consumer electronics products tend to be premium priced models with high-level specs and compete at the flagship or upper mid-priced levels.  For the most part this has served Apple well with a devoted user base and a cult-like following.  That said, given inflation leading to rising prices and more recent strength in the US dollar against other currencies, foreign consumers face greater challenges in terms of typical purchase decisions.  In Japan Apple has raised prices by 30%+ and similarly South Korean Apple prices have seen a 20%+ increase, and the threat of tariffs has done even more damage from a psychological standpoint.   
Typically, as with most CE producers, Apple sets a list price for its products, which it expects retailers will hold for between 30 and 60 days in most cases.  While Apple sells at list for an extended period, retails have some leeway as Apple sets a MAP (Minimum Advertised Price), essentially the lowest price a retailer can sell the product for.  This is strictly monitored, although we note that some retailers use gift card inclusions to create a lower price without violating the MAP.  Apple leads the way for special event discounts and one-time offers that help retailers during holidays and special events, while high volume purchasers can get up to a 10% discount, as do educational buyers.

​However, this leaves a large portion of the consumer population, those that do not typically purchase or need flagship devices, to other brands and we believe Apple has come to the realization that if done carefully, the company can maintain its elite image but expand its customer base with lower priced devices.  Apple released the iPhone 16e, a trimmed down version of the iPhone 16, in February, with a less bright display, an older GPU, and a slower charge time, for $200 less ($599) than the initial iPhone 16 price ($799), a 25% lower price cut, and now rumors are circulating that Apple is developing a lower priced MacBook.
The rumors abound concerning this potential new product, but the focus is really on two characteristics.  The first being the fact that the new device is not expected to use Apple’s M series chipsets but will use the same chipset used in the iPhone 16 Pro to reduce costs.  There are some other smaller modifications but both devices are quite similar, although the new device is expected to be priced between $200 and $300 lower than the current MacBook Air ($999 base price), and a few optimists expect even lower pricing. This would give Apple greater access to middle-tier buyers and could be advertised as a model designed for the educational market, avoiding tarnishing the premier Apple image.
That said, should such a device appear, it will have a distinct effect on the mid to low end laptop market and the Chromebook market.  While Chromebooks initially were  priced below $300 to offer a ‘simulated’ client-server environment, they have increased in price over the last few years, averaging close to $500 currently.  A $600 or $700 MacBook would put significant pressure on Chrome book prices and could lead to substantial deterioration in the category.  At the same time the low-end laptop market would face the same pressure, although with a more sustainable and diverse product mix.  Again, these are rumors and have yet to be confirmed or even acknowledged by Apple, so we expect the usual mock-ups, ‘visualizations’, supplier whispers, and knowledgeable sources to continue to feed the flames, but from a product perspective such a device does make sense and Apple’s carefully crafted advertising could keep it from casting a lower class shadow on Apple’s premium products.  After all, not everyone buys the top of the line. 

You walk in the door after 45 minutes of bumper-to-bumper traffic and a day of an endless stream of meetings that accomplished virtually nothing, and the wife is at the door, usually not a good sign…
“We have a problem.”
“What now?  Did someone get arrested?”
“No, but it looks like Josh won’t be going to Harvard”
You are actually relieved in a way, considering the tuition, but you have to bite.
“What did he do?”
“He didn’t do anything, that seems to be the problem.  We just got a call from his guidance counselor with the results of the Strong test that they gave the whole 9th grade class.”
“What’s a Strong Test?”
“It’s a ‘career inventory’, sort of interests that they use to assess potential career paths for students.”
“It doesn’t sound like something he could fail, so what is the problem?”
“ You can’t fail but based on what he said on the test he was about as far away from a high paying job as you can get, at least according to his guidance counselor.  Based on his love of the outdoors, his love of motorized equipment, and his appreciation of nature, the test suggested a career as an ‘industrial truck and trailer operator’!”
“You mean a big rig driver?”
“I think so.  We didn’t get into details, but he said Josh seemed very happy with the results.”
“So where is the problem?”
“Our son is going to be a truck driver”
“It could be worse…”
OK so Harvard was out of the question, but a semi driver?  You had a bit more hope for your progeny, but at least it’s an honest living.  In the new reality it might prove to be a far more valuable profession than you might realize, even though it might not seem so now, and that is because of AI.  A recent Microsoft (MSFT) study took over 200,000 conversations between users and Microsoft Bing Co-Pilot (anonymized & privacy scrubbed) and derived the most common work activities that people seek AI assistance for, and the most common activities that AI itself is performing.  By combining this data they were able to complete a new set of data that gave an AI applicability score for each. 
What this translates to is a list of jobs most and least likely to be replaced by AI over the next few years and includes addition study information from previous works.  The descriptions are simplified here, but they are in order, starting with jobs most likely to be replaced by AI (Highest AI Applicability Score) to those with the lowest chance of being replaced, and blue-collar jobs seem to be the safe bet.  We highlighted a few. Here are the lists:

Here is the bottom of the list, essentially those jobs with the least chance of being replaced by AI: 

In a more general way, here’s the data sorted by groups (High is bad):

​Tomlinson, Kiran, et al. Microsoft, Working with AI: Measuring the Occupational Implications of Generative AI. 

Some consumer electronics categories are relatively easy to dissect, and others are not.  More mature categories like smartphones, that are no longer in an explosive growth mode, are more transparent to macroeconomic  affectations than those where growth might mask the effects of slowing demand or currency issues, so we try to see through some of the company hype disseminated during quarterly or interim presentations to sift out what amounts to noise.  One category that is a bit easier to decipher is the e-paper product category, albeit one that is still in a high growth trajectory.  The leader in the space, and holder of much of the technology IP, is E-Ink (8069.TT), and while the company has been prducing electrophoretic display products for years, recent advancements to the technology that allow for color, rather than black & while only e-paper displays, has pushed the company back into a growth mode.  Not only is the company unusual in that it holds the IP for e-paper technology, but the technology itself is quite interesting and has reached a point where it is drawing the interest of major CE companies.

Most investors would recognize E-inks technology from the early e-reader market, with Sony (SNE) the earliest adopter in 2004 but made famous by Amazon’s (AMZN) Kindle reader, which is still popular today.  Over the years the e-ink market (we use it as a generic term here), has divided into two broad product categories, e-readers, and ESL’s (Electronic Shelf Labels) that are placed with retail items and contain pricing and other information about the specific product.  Typical brick-and-mortar grocery stores carry ~32,000 items, the price of which can change as often as weekly and sometimes daily. The traditional method for changing prices at the shelf level is to send the label information to a printer who prints out labels which are sent back to the store.  A team of associates must then find the item in the store, remove the old label and place a new one on the item’s shelf.  This is a tedious and expensive process that takes a team hours to accomplish and can entail thousands of changes each week for sales and promotions.  ESLs are battery operated  and wireless, but what makes them most beneficial is that not only can each item label be controlled individually, but the labels only draw power at the moment when the label is changing, as opposed to a constant power draw required by other electronic displays.  This means that a single person can change hundreds of labels in a few seconds from a computer or tablet, and that ESL’s themselves last up to 5+ years on its initial battery.  In fact, they draw so little power that E-Ink has said that when the battery does finally die, most customers replace the ESL with a new model rather than have the battery replaced, given the advances in the technology over the l5 year+ lifetime  The initial cost is high, given the number of tags, but the typical payback period runs from 1.5 to 2 years in Europe to 2 to 2.5 years in the US, so the cost is relatively quickly covered.

The Market
​
While more mature CE product market estimates vary, estimates for relatively new markets like electrophoretic (e-paper) displays vary far more and are therefore far less reliable, especially as large CAGRs drive higher readership or more report sales, so we show the market values we have collected tentatively and without source names.  Our base year is 2024 for sales value but the CAGR periods differ.

 
The e-paper market is a three product category market, with consumer products, such as e-readers being the largest, followed by ESLs, and signage the smallest.  Yet we expect the proportions to change over the next few years as capacity for color e-paper expands, making larger size signage displays available.  On a regional basis, North America, with its higher levels of disposable income has been the largest markets but China is thought to have the greatest penetration share potential in the long-run based on population size.  In terms of leadership, we believe E-Ink holds a very dominant share of ~70% as the primary producer of e-ink display films and inks, while others like Samsung Electronics (005930.KS)) and others license the technology or package the raw displays into products.  In June of this year Samsung, who holds the leadership position in the digital signage market (since 2009) and has a 34.6% volume share of the digital signage market (2024) introduced its first e-paper product, a 32” (2,560x1,440) color E-Paper digital signage display system, using E-Ink technology and Samsung’s proprietary waveform algorithm.  The device can be managed with an app on Android or iOS and interfaces with Samsung’s VXT control and content creation platform, which has now been optimized for color content and control.

With the marketing strength of a company like Samsung behind the technology, we expect not only will the color signage market grow, but other large CE companies will follow, with LG (066570.KS) a likely candidate.  Sharp (6753.JP) already has a number of e-paper signage products.

E-Ink

While the company’s original concept and product is unlike almost any other consumer display technology, E-Ink’s more recent technology leap, to produce color e-paper displays, moves the technology into a much broader product range and moves the technology from a poor cousin of more typical consumer display technology to a full family member that solves problems that other display modalities cannot.  E-Ink has multiple production lines, all of which a currently fully utilized.  H5, the most recent is in the customer certification stage, with sample product produced, but will likely not be in full production until 1H 2026, while H6 is expected to initiate construction in 2026 with production in 2027, extending the company’s production reach.
We note that as E-ink’s black & white and color technologies are quite different from each other, not all lines can produce all technologies and  as e-ink is primarily a seller of e-ink films, inks, and modules, they have been capacity constrained at times and will likely remain so as color e-paper demand increases in the near-term.  The company’s 2Q results, which were up 31.8% q/q and 38.8% y/y did include some pull-in due to tariff concerns, so 3Q is expected to be down slightly q/q and the company remains conservative about 4Q, again because of tariff questions, but has not changed its full year outlook.  We do note that currency (NT$) had an effect on operating profits, even with hedges in place, as the NT$ swing was considerably larger than expected (1% currency move = 0.3-0.5% of op profit)) but almost 100% od sales and COGs in in US dollars.  On a longer-term basis the company remains quite optimistic within normal macroeconomic constraints.

There are a number of companies that hold key IP for consumer oriented technologies, most of whom garner considerable income from licensing.  While E-ink is also in this enviable position, the company generates less than 2% of sales from royalty income[1].  E-ink management has a somewhat unique approach to this issue, saying that they are far more interested in making their technology available as a producer than as a licensor, and will continue on this track to promote the technology and expand its market.  While we know other companies give concessions to certain customers, E-Ink is a bit more open about how they look at customers, again with the idea that happy customers expand orders.  The example would be with gross margins.  The company targets gross margins in the 50% to 55% range, but at times, because of its high utilization rates, has exceeded those rates (over 60% in 2Q).  The company has indicated that they will ‘share’ gross margin over the 50% - 55% level with customers, a very progressive concept.

Technology

In order to understand E-paper, it is important to understand both why it is different from other display types and how it works.  E-paper uses inks not organic materials or phosphors, and they do not generate light like LCDs (backlight) or OLEDs (self-emissive).  They are reflective displays that reflect external light off these inks to produce images.  That means they work best in bright light and worst in dark areas, typically the opposite of other display modalities.  This makes them ideal for outdoor signage or any location that is lit 24/7.  The biggest draw for e-paper however is the fact that the only time the display draws power is when the image is changing.  If the image is static, such as with an ESL, no power is used.  This means that battery life is extremely extended and other than an initial charge, no connection to an external power source is needed, making e-paper ideal for remote locations where power sources are not readily available or locations where power source changes (battery) are difficult (remote sensors, etc.).  Here’s how it works:

E-ink’s original system was a film with microcapsules (Think pixels).  The capsules were filled with an oily liquid in which black (Carbon black) and white (Titanium Dioxide) ink particles were suspended.  The white ink particles were positively charged, and the black ink particles were negatively charged.  An electrode under each microcapsule produced either a positive or negative charge, with a positive charge repelling the white particles, pushing them to the top of the capsule and attracting the negatively charged black particles, pulling them to the bottom.  As the top of each microcapsule was on the surface of the film, that capsule would produce a white dot, conversely a negative charge on the electrode would produce a black dot.  As the inks are suspended in the capsule fluid, once they are moved to a location they stay in place, which is the key behind the fact that power is only used when changing an image.  The example would be an ESL, which would use power only for the few seconds when a new price or information is being added, and would use no power until the next time it is changed, which could be days or weeks.

E-Ink used this technology for black & white e-readers and a number of other products to seed the electrophoretic market in the early 2000’s, but  found it was always being negatively  compared to other display types which were able to display thousands, if not millions, of colors, so E-ink took the path of placing a RGB color filter over the film, allowing the white dots to become color.  This is the same tact that LG Display (LPL) uses for its WOLED displays, where the OLED materials produce a white light that is passed through a color filter to create colors.  However, color filters do just what they say, they filter out a large portion of the light, reducing its brightness.  As OLED materials produce their own light, the result for WOLED is still viable, but e-paper is reflective and  in low light situations the color filter reduced the brightness so much as to wash out the colors.
E-ink took on the challenge and figured out a way to produce colored e-paper without using a color filter, and we have to say, it is an unusual and interesting system.  Here’s how it works:

The microcapsules ae shaped a bit differently, but they are still filled with an oily liquid, and now they include white ink as in previous versions, but also include cyan (lt. blue), magenta (purple), and yellow inks, the same basic inks used in magazine printing.  Combining these colors produces roughly 60,000 color combinations, a far cry from the many millions produced on RGB displays, but enough to satisfy the printing market for many years and most signage.  The problem is, with this many inks how is the system going to push them up to the top individually?  Before (B&W) there was a positively charged and a negatively charged ink to correspond with a positive or negative charge at the anode, but now there are four inks.

E-ink’s solution depends on its proprietary inks, each color with a different charge and a different size.  The voltage at the anode, which was positive or negative for black & white e-ink films, now has a complex waveform which varies between polarity (charge) amplitude (signal strength), and duration (frequency).  The easiest to understand is polarity as each of the inks has a charge between positive and negative.  By varying the polarity of the waveform for each microcapsule the waveform will push a particular color ink to the top.  The waveform amplitude determines the ‘position’ of the ink in the suspension (high to low), which is key for color mixing (Blue on top, yellow in the middle makes green, blue and yellow on top makes light green). A big push gets the color to the top while a little push might only move it to the middle.  As each ink is a different size (Bigger moves slower), the duration of the waveform also affects positioning.

This makes the waveform extremely complex and obviously different for each microcapsule (pixel) for each frame of the image, so E-ink created an algorithm that takes the RGB image data that is being fed into the system and computes what combination of inks will give the closest color match.  It them looks ate the micro-capsule to take a ‘snapshot’ of where each ink color is and figures out what the waveform should be to move the inks in that capsule from where they are to where they should be.  However, if it finds that a particular microcapsule (pixel) does not change between the old image and the new one, it tells the waveform generator to skip that capsule, saving even more power.  Once it completes all the changes (price update, etc.) it stops.  Remember that the ink particles remain suspended in position once they are moved and therefore require no power or refresh until the next time they need to be changed.  This could be days, weeks, months or years, while other display types need each pixel to be refreshed a minimum of 60 times each second.
​
Those who work closely with E-ink understand the mechanics behind how the system works and are able to modify the waveform algorithm to their own specifications, as did Samsung in its 32” color offering, but the technology is all E-ink’s.  With the production of the microcapsule film done at E-ink, OEMs can make waveform and algorithm modifications as they wish but must pay a royalty (see above) and purchase materials from E-ink. E-ink believes that this color system will allow a broader set of applications, particularly for outdoor signage,  that will expand the market and keep utilization high on the company’s new lines.  So far, even given the capacity constraints and macro issues that surround color e-paper, they seem to be on the right track.


[1] E-ink shows Royalty Income in Non-Operating Income, we include it in sales to estimate the percentage