The Premium Battleground: Margin Squeezes and Upmarket Shifts
The TV set market is among the most competitive in the Consumer Electronics space, with over a dozen brands maintaining a market share over 1%.  That said, almost 50% of the market is controlled by the top three TV set brands, Samsung Electronics (005930.KS), TCL (000100.CH), and Hisense (600060.CH), with Samsung remaining the dominant brand for the last 20 years.  Both Chinese brands became successful both in China and then globally, by offering lower priced sets that have features similar to higher priced brands originally achieved immense success, first domestically and then globally by offering lower-priced sets featuring specifications similar to higher priced legacy rivals.  This tactic was an  effective while the Chinese economy was in high growth mode, but slower domestic growth in China combined with a relatively stagnant global TV market has continued to squeeze brand margins across the board.
While Chinese brands ruled the lower-priced tiers, Samsung has dominated the premium TV set category, where it faces less intense competition from Chinese brands.  However, over the last few years, driven by local market saturation and margin compression, Chinese manufacturers have aggressively moved up the value chain to capture a slice of this high-margin premium territory.
The OLED Misstep vs. The Mini-LED Leverage Strategy
China was late to embrace the true potential of OLED displays, a strategic mistake that allowed South Korean panel producers to rule the large-format space for years.  The root of this error lay in state infrastructure planning: the Chinese government had heavily subsidized the massive build-out of legacy LCD manufacturing infrastructure. To protect the production value of those massive state assets.  A disruptive technology like OLED was initially suppressed because it threatened to devalue existing LCD resources
That said, Chinese panel producers did not make that mistake when it came to more recent mini-LED technology because Mini-LED is a direct architectural outgrowth of LCD—improving the backlight array rather than replacing the panel entirely.    Mini-LED technology was rapidly adopted by large Chinese panel manufacturers who weaponized it as a defensive moat against encroaching OLED market share.  By adapting existing LCD backlight lines without incurring the multi-billion-dollar capital expenditure of building fresh OLED fabs from scratch, Chinese producers rapidly matured their processes. This gave them a high-performing entry ticket into the premium TV market without devaluing their core LCD infrastructure, allowing them to develop large-panel OLED capabilities at a much slower, less risky pace.
Squeezing the Leader: TCL Overtakes Samsung in Shipped Units
Consequently, while South Korean firms remain the sole suppliers of large-panel TV OLEDs, China has successfully scaled its local small-panel OLED supply.  On the other side of the equation, Samsung has become the largest Mini-LED TV supplier, despite the fact that they do not produce any LCD TV panels themselves.  They have used Mini-LED TV to expand their premium TV offerings and fill in gaps in their high-end TV set lines consisting of OLED and ultra-premium Micro-LED lines.
Samsung, in  typical brand fashion, makes improvements in technology each yearly cycle, adding features to stay one step ahead of the competition. However, they have found that Chinese brands are closing the feature gap rapidly, keeping intense pricing pressure on Samsung’s premium tiers.  This year Samsung has taken a more radical approach and made substantial changes to its TV product line structure to offset incursions by Chinese brands.  We note that in the month of December last year TCL surpassed Samsung in terms of Mini-LED units shipping at a 16% share rate to Samsung’s 13%.  Below, we break down the structural evolution from the 2025 lineup to the newly overhauled 2026 strategy designed to neutralize TCL and Hisense.
Retrospective: Inconsistencies in the 2025 Samsung Premium Lineup
In 2025 Samsung’s premium Mini-LED TV set line was bifurcated into an 8K segment and a 4K segment, both of which consisted of Mini-LED TV sets with Quantum Dot enhancement.  There were no sets with only Quantum Dots last year, which in previous years had made up the lowest price tier of the premium line  The 8K line  consisted of two price tiers, while the 4K line consisted of 4 price tiers.

As the data shows, the 2025 product architecture suffered from clear operational friction, particularly in the 4K mid-price tiers where MSRP bands overlapped heavily.  While underlying features did vary by tier, the subtlety of those hardware changes was effectively invisible to the average consumer, whose purchasing criteria remains strictly ordered by price, then size.  Last year Samsung also eliminated the line of Quantum Dot (No Mini-LED) TVs that represented the lowest tier of the premium set line in previous years.
Note:  In this note we discuss only Samsung’s Mini and Micro-LED TV set products.  Samsung’s OLED TV sets will be covered in a separate note.
The 2026 Structural Pivot: Streamlining the Technology Matrix
To clean up consumer confusion and protect value capture, Samsung’s 2026 strategy redefines the premium tier into four highly distinct technology categories.

• Micro-RGB –  (14 models) While Samsung previewed this technology with one model last year, Samsung has expanded this technology to 15 models and made it the crown jewel of the premium TV set line.
• Mini-LED – (23 Models) In terms of premium TV set unit volume, Mini-LED TVs are Samsung’s largest category.  This tear it is broken down into three categories.  A single 98” 8K Mini-LED set with Quantum Dots, two categories of 4K Mini-LED TVs with Quantum Dots, and two categories of 4K Mini-LED Tvs without Quantum Dots, with the latter the lower priced category.
• OLED – (15 Models) Samsung uses two technologies for its OLED TVs.  QD/OLED, a quantum dot based OLED that is produced by Samsung Display (pvt), and WOLED (White OLED) which it buys from LG Display (LPL).  While the detail of which technology is particular to each model, on a general basis the high-end price category is QD/OLED and the Low-end price category is WOLED.
• Quantum Dot – There are no Quantum Dot only TV sets in Samsung’s premium TV set line this year.  Samsung does provide a single standard Quantum Dot line (QN60H) that is quantum dot enhanced but is not part of the premium lines.  Quantum dots are used in some of Samsung’s ‘Specialty’ sets, such as the Frame and The Serif.

Here is the breakdown of Samsung’s Premium line in more detail:

Cross-Category Price Optimization: The 65” Index
Even with structural cleanups, pricing overlaps across different display technologies still exist. To map how Samsung is positioning these competing technologies against each other in the showroom, we can isolate and rank the standard 65” models directly by price.

Engineering Fixes for the "Micro-RGB" Backlight Mismatch
This table illustrates the three price tiers.  These sets are called “Micro-LED RGB”, meaning that each backlight LED no longer consists of a single blue LED that is converted into an RGB pixel with a phosphor or quantum dot color filter.  This hierarchy brings us to an important technical distinction regarding the top-tier "Micro-RGB" sets. In these sets, each backlight LED no longer relies on a basic blue emitter converted via phosphors or a traditional color filter sheet. Instead, Samsung deploys miniature red, green, and blue LEDs (roughly one-fifth the size of standard sub-pixels) to generate a localized backlight point matching the exact color value of the targeted image zone. 
This differs from true Micro-LED displays in that in the Samsung consumer product mentioned above uses traditional LCD technology, while true Micro-LED displays use only the light generated from the Micro-LEDs (It is critical to note that this is not a true emmissive Micro-LED display. These consumer models still utilize an active LCD panel, whereas true Micro-LED displays completely eliminate the LCD layer, using the light from the micro-LEDs directly.  
In a true Micro-LED TV, there are 8.2944 million pixels, each with a red, green, and blue micro-LED (24.8832 million Micro-LEDs).  Each pixel can represent any color and intensity.  In the Samsung consumer “Micro-LED RGB” TV there are 5,000 backlight zones to light up over 8 million pixels, which means there is a significant mismatch between the backlight and each pixel, essentially meaning that each backlight point must illuminate over 1600 pixels.  If in that 1600 pixel block there are blue pixels on one side and red pixels on the other, the backlight has to decide whether to be red or blue (or purple), which doesn’t always match the image.  This can lead to less precise colors than might be possible with a color filter.
Samsung has engineered a number of ways to compensate for this.  Rather than increase the number of ‘zones’, which comes at a steep cost, Samsung is thought to employ a simpler color filter, not at the level of those used in its standard LCD TVs, but one that helps to ‘corral’ the light a bit to lessen color bleed.  The real trick is Samsung’s AI engine, specifically designed to predict real-time luminance and color leakage across adjacent zones, proactively adjusting block intensity to minimize visible halo artifacts.  While this digital patch does not completely erase the architectural limitations of an LCD bottleneck, the resulting visual output easily supersedes standard Mini-LED performance, justifying its place at the apex of Samsung’s pricing ladder
Channel Clearance Trends: Deflation in Legacy 2025 Inventory
Most recent 30 day summary

​We expect the most recent price declines in the Samsung 2025 premium Mini-LED TV line were in anticipation of the release of the 2026 line, which is rather different than the 2025 line.  The changes in the 2026 line would require distributors and retailers to make room for new inventory, both at the high end (dollar value) and the low end (physical) in order to capitalize on the momentum and high margins behind the initial release.  Consequently, multiple 2025 SKUs have just plunged to all-time lifetime lows. 
For those looking to build a home theater, Samsung’s QN115QN90F, a 115”  Mini-LED/QD behemoth, a bit over 8.4 ft wide and 4.82 ft high, has dropped in price from $27,000 to a new low of $25,000, now almost 17% below its original MSRP of $30,000.  For those with less wall space and smaller budgets, both Samsung’s QN75QN80F (75”) and the QN65QN80F (65”) sets also hit new lifetime lows, with the 75” model dropping from $1,600 to $1,400 (↓12.5%) (Originally $2,300), and the 65” model dropping from $1,000 to $900, now at 50% of its original $1,800 MSRP.  If you are willing to forego 2026 bells and whistles the need for inventory space and capital typically creates value and for commercial buyers or custom installers unconcerned with the 2026 AI backlighting algorithms, the immediate cash-flow demands of the 2026 structural retail reset are opening up significant, high-yield buying opportunities
Strategic Outlook: The Cost of Market Leadership
Samsung’s aggressive 2026 restructuring is a textbook defensive maneuver against structural changes in the display industry. By splitting its Mini-LED line, expanding dual-sourced OLEDs, and introducing the premium "Micro-RGB" tier, the company is attempting to out-architect competitors it can no longer out-produce on a pure hardware level.
Protecting the premium tier comes with significant challenges:

• Engineering vs. Asset Dominance: Chinese brands like TCL and Hisense have successfully converted their subsidized LCD infrastructure into a high-yield Mini-LED advantage. Samsung is forced to rely on complex AI processing and algorithmic corrections to maintain a performance gap, compensating for its lack of in-house LCD manufacturing.
• The Squeezed Middle Tier: As the 65-inch pricing index demonstrates, Samsung’s mid-tier OLEDs, high-end Mini-LEDs, and entry-level Micro-RGB models are priced closely together. This overlapping layout risks confusing consumers and cannibalizing sales within the company's own lineup.
• Inventory Demands: The immediate downward pressure on remaining 2025 inventory shows how quickly value drops when a product line is reorganized. Retailers are slashing prices on larger models like the 115-inch flagship to clear warehouse space, demonstrating the high financial cost of shifting to a new hardware cycle.

Ultimately, Samsung's strategy shows that maintaining market leadership in consumer electronics now depends as much on software optimization and clever product tiering as it does on panel innovation. As Chinese manufacturers continue to close the feature gap and improve their profit margins, Samsung’s new product layout will face a critical test: determining if proprietary AI and brand equity are enough to protect the world's top TV brand from aggressive price competition.

As always, we pay particular attention to display panel prices as they are the front-line of the CE product pricing cycle.  While memory price increases have taken center stage in recent months, the display panel percentage of BOM for laptops ranges from 20% to 25%, for TV sets, between 40% and 50%, and for monitors, over 60%, making them a key determinant in the CE product pricing model.  That said, we are currently at an unusual juncture, with a number of “one-time” factors playing into that equation, with panel prices only one factor in what has become a much more complex pricing architecture.

The Subsidized Squeeze: Chinese New Year and the 85-inch Shift
February saw only modest moves in IT panel prices (Monitors and Notebooks), while TV panel prices settled slightly above our expected range.  We believe this is a result of the utilization cuts that panel producers made during the Chinese New Year holiday, tightening supply and generating enough unease among panel buyers that they were willing to pay higher prices to build TV panel inventory.  While demand remains relatively static globally, better than expected premium TV set sales in China during the holiday, spurred by the renewal of “New for Old” government subsidies, pushed panel buyers to refill depleted panel inventory.  This was primarily felt in panel sizes 85” and above but pushed panel producers away from more conventional sizes (<85”), reducing supply and raising prices.  We expect this to continue into March but to a lesser degree as demand remains flat as the Chinese subsidy momentum subsides.

The Energy Tax: Crude Volatility and Carrier Surcharges
As of As of March 13, the 'short-term effects of the oil situation' have manifested in a new layer of costs. Major carriers have enacted an Emergency Fuel Surcharge (EFS), roughly $150–$160 per container for long-haul routes, to account for Brent Crude breaching the $100/bbl. mark following the closure of the Strait of Hormuz. For a 65” TV, this adds a non-negotiable $0.76 per unit to the landed cost before it reaches a US port.
Chipflation 2.0: AI-Driven Shortages in the Smart TV BOM
From the BOM perspective, since the end of 2025 component prices have continued to climb, particularly memory and advanced packaging products., and while TV sets use significantly less memory than laptops, the push toward  HBM (high Bandwidth Memory) used for AI has left commodity memory like the DDR4 used in TVs in short supply..  Even though contract prices were up 4x last year, they have already risen almost 60% in 1Q of this year.  This has pushed memory, as a percentage of BOM, from 2.5% of manufacturing cost last year to over 7% as of the end of this month.
In addition, the demand (again AI) for CoWoS (Chip on Wafer on Substrate) components has made it more difficult to maintain a consistent supply of SoC’s used in TV video processors, especially at entry and mid-price tiers.  This has caused a ~15% increase in the cost of SoC’s used for video processing due to a lack of advanced packaging materials.  Further, the oil issue has even deeper ramifications at $100/bbl. as a feedstock to the plastic resins that form TV set chassis.  This will add another ‘soft cost’ to the BOM of ~$3.00.
The $20.10 Anomaly: Quantifying the Tariff/Energy Trade-Off
That said, the surprise here is that when we do an analysis of all the puts and takes that affect the CE space over the last 30 days, the buyer of a 65” LCD TV should have seen a theoretical $20.10 reduction in price, even with the price of both oil and memory increasing (and panels) during the period.  The Customs Bureau stopped collecting the IEEPA tariffs on February 20, roughly a 23% (high-volume CE imports) levy, replacing it with a 10% global flat-rate tariff that will last for 150 days (7/24/26).  In terms of our 65” TV, that change had a larger dollar value than all of the component and logistic increases, netting out to a positive $20.10 swing.  Simply, the tariff windfall ($32.50) far outweighed the energy/logistics/BOM drag ($12.10).
The Capture Conflict: Who Gets the Windfall?
Of course, the consumer will likely not see most of that theoretical improvement in price as there are a multitude of outstretched hands that come first in the CE supply chain.

We note that this infographic is good for today only, as any extension of the oil price rise will erode the tariff benefits, but we do note that while the oil situation is certainly a crucial one that has a widespread effect across the entire CE supply chain, the tariff issue has a longer-term effect that gives brands margin flexibility that they did not have coming into this year.  Whether they capture it internally to boost profitability or use it to boost sales remains a big, and as of yet unanswered question, but we expect while all supply chain participants are aware of the impact of the tariff change, they are currently more focused on the short-term effects of the oil situation.
​
Conclusion: The 150-Day Margin Buffer
In a standard year, the simultaneous rise of $100/bbl crude, a 60% memory spike, and higher-than-expected panel prices would have mandated a significant MSRP hike for the US 65” TV market. However, the February 20 Supreme Court ruling has provided the industry with a temporary, yet substantial, "get out of jail free" card.
By replacing a 23% IEEPA levy with a 10% Section 122 duty, the administration has inadvertently created a $20.10 per-unit margin cushion. This windfall is currently acting as a silent stabilizer, absorbing the $12.10 logistics and BOM drag that would otherwise have crashed into retail price points.
The critical question for the next quarter is one of duration. With Section 122 set to expire on July 24, 2026, and the USTR already pivoting toward new Section 301 investigations, this margin buffer is likely a "150-day anomaly." We expect brands to maintain current price rigidity, ignoring the theoretical $20.10 consumer savings, to build a war chest for a potentially more expensive H2 2026. For now, the supply chain isn’t fighting for the consumer’s dollar; it is fighting to capture a piece of the SCOTUS-driven windfall before energy and labor spikes eat it alive.

IPO Overview and Fab Infrastructure
HKC (pvt) received final approval from the Shenzhen Stock Exchange for an IPO in which the company is expected to raise 8.5 billion yuan ($1.238 billion US).  HKC is China’s 3rd largest display producer (13.4% share of China) behind BOE (200725.CH) and Chinastar (pvt), generating between $350 million and $500 million (US) on a monthly basis (2025).  HKC currently runs four fabs, all of which are Gen 8.6, with a total capacity of 350,000 sheets/month. 

• H1     - Chongqing
• H2     - Huzhou
• H4     - Mianyang
• H5     - Changsha

Of that total 230,000 sheets/month use an Amorphous Silicon (a-Si) backplane, as is used in typical LCD TV sets and large panel display products. 120,000 sheets per month use an IGZO (Indium-Gallium Zinc Oxide) backplane, necessary for high resolution, variable refresh rate panels.  The main thrust of the capital raise (38%) is earmarked for expanding HKC’s IGZO capacity at their Changsha fab by between 30,000 and 50,000 sheets per month by converting a-Si capacity to IGZO.  As they develop and refine this capacity it will give them a broader platform on which to produce IGZO-based LCD products, such as RGB Mini-LED, and will give them support for their eventual expansion into OLED IT display production.

The IGZO Pivot: Bridging to IT and OLED
The IPO also funds (30% of proceeds) HKC’s eLEAP Photolithographic OLED production process with an OLED R&D Development Center and an upgrade of the company’s OLED pilot line at the Changsha fab.  While the R&D Center will continue the company’s OLED development, the funding will move the pilot line closer to commercial scale, in order to prove out the eLEAP technology in a mass production setting.  We note that eLEAP OLED technology is a proprietary technology that has been licensed from Japan Display (6740.JP).  At one point HKC was exploring a partnership with Japan Display to develop, commercialize, and build an eLEAP OLED fab, however that relationship fizzled when JDI insisted on high licensing fees.  Instead, HKC acquired some of JDI’s Gen 6 equipment when JDI closed its Mobara fab and receives technical support from JDI, but license the eLEAP technology and is building out the more advanced pilot line independently.

The eLEAP Shortcut: Skipping the Development Cycle
The distinction here is important as HKC, by licensing this technology, skips a large chunk of the development cycle and potentially shortens the time it will take for them to directly compete with SDC and BOE in the Gen 8.6 OLED IT space.  That said, this is still unproven technology, and while it shows promise at the pilot line level, taking it to mass production is another ball game.  SDC and BOE have spent years developing their Gen 8.6 OLED IT production technology and we expect HKC’s path to a competitive product will take some time.  HKC will be using JDI’s equipment and technology for a Gen 6 line as a precursor to an actual Gen 8.6 OLED fab, which indicates that the time to HKC’s Gen 8.6 IT OLED mass production is still years out.
We note also that the cost of HKC’s eLEAP buildout has already been accruing.  HKC paid JDI a fixed technology transfer fee estimated to be between $150 million and $200 million, but JDI has used its largest and controlling shareholder (Ichigo Trust) to hide the actual amount from investors.  The royalty payment goes to JDI’s parent (who actually owns the IP) and will be based on area (fixed fee) and price percentage.  This keeps Japan Display from losing out should HKC lower prices to gain share.

Near-Term Revenue: Mini-LED and Next-Gen Backlighting
The third leg of the IPO is a bit more direct and is meant to generate nearer-term revenue while the other projects are expected to deliver in the future.  In concert with the 2 billion yuan IPO funds for this project, HKC is also funding an 8 billion yuan (10 billion yuan or $1.456 billion US total) project to develop the production of Mini-LED backlight modules.  This includes COB (Chip-on-board) processes to increase the number of dimming zones and will also produce Direct view Micro-LED tiles. 
The company recently released the HKC M10 Ultra monitor which is based on RGB Mini-LED technology.  Typical Mini-LED displays use all blue LEDs with a quantum dot conversion film to create RGB colors.  In this case, the backlight provides the color but on a broader scale, not on a sub-pixel basis.  This means that the color filter that used to be the sub-pixel color source is now an ‘enhancer’ making the sub-pixel colors more precise.  This allows less ‘filtering’ and a brighter display, albeit at a higher cost.

The 21 Billion Yuan Question: Financial and Yield Constraints
As the transition that HKC is attempting to make from a commodity LCD panel producer to a premium OLED product producer is a tricky one to say the least, the risks associated with the IPO are substantial.  HKC is currently carrying 21 billion yuan in short-term debt and has a Debt/Asset ratio of almost 69% which might make some concerned as to the true motivation behind the IPO.  In terms of operational risk, a more standard approach to Gen 8.6 IT OLED, such as the one taken by both Samsung Display and BOE, might reduce that risk but the company has chosen to use an untested (in mass production) technology which could lead to an extended period of low yields.
​
The V-Stripe Factor: IP Litigation and Sub-Pixel Clarity
HKC also has another risk that could come into play at a later date.  While it seems less than important to the average user, the layout of sub-pixels on displays, is a big deal.  The diamond sub-pixel pattern that is typical in many OLED displays is ideal for video reproduction, however it is less so for text and has been known to cause jagged edges or blurring on some displays.  Over the years various brands attempted to change the pattern to reduce or eliminate the text distortion.  Sony (SNE) accomplished the task in their video production monitors but a price above $25,000 for a 17” monitor kept it out of commercial production.  JOLED (defunct) did it as part of their inkjet printing system and more recently Samsung Display (pvt) did it with their V-Stripe layout for their QD/OLED displays.  HKC, as part of the Japan Display technology, also uses a V-stripe layout that differs slightly from Samsung’s, but should Samsung Display decide to litigate the V-Stripe IP, particularly the processes required to place the sub-pixels in that configuration and win, HKC’s entire OLED R&D, development, and licensing program would be in jeopardy.

The India Strategy: The Dixon JV and Regulatory Tailwinds
​HKC does have an additional card to play.  They just received final approval for a JV between the company and Dixon (540699.IN), India’s largest in-country EMS.  Dixon assembles for a variety of primarily Android and Domestic IT firms in India, and in that space has a ~45% market share, They also compete (small share) against multi-nationals like Foxconn (2354.TT), Pegatron (4938.TT), and Wistron/Tata (pvt) in the iPhone ecosystem but have had little to offer that their competitors cannot match.  HKC will have an outlet for their IT products in Dixon, who can offer customers discounted packages if the opt for HKC displays, benefiting both HKC and Dixon.  But while the Android market is ripe for the JV, Dixon is at the early stages of becoming an approved component vendor for Apple (AAPL), which means that potential incremental market is still years away.  That said, the Indian government recently increased the incentive outlay for components, giving the JV additional incentive to build its component supply business.

Conclusion: The eLEAP Gambit
HKC stands at a critical strategic inflection point. By moving away from the "commodity trap" of Gen 8.6 a-Si TV panels and aggressively funding the IGZO and eLEAP OLED transitions, the company is attempting to rewrite its identity as a Tier 1 technology innovator. However, the path to 2027 is paved with significant execution risks. While the licensing of JDI’s eLEAP technology provides a theoretical shortcut to competing with Samsung and BOE, the transition from a pilot line to Gen 8.6 mass production, without the safety net of a major South Korean partner, is a feat rarely achieved in this industry.
Furthermore, the company’s high debt-to-asset ratio (69%) leaves little room for the yield-loss "learning curve" typical of new photolithographic processes. The "wild card" in this narrative is the Dixon JV in India, which finally cleared its major regulatory hurdle on March 10, 2026. This 74:26 partnership provides HKC with a secured, high-utilization outlet for its IT modules in a rapidly expanding market. If HKC can successfully leverage this "Make in India" bridge to generate near-term cash flow while its OLED technology matures, it may just pull off one of the most ambitious technological leapfrogs in recent display history. For investors, the HKC IPO is not just a bet on a display manufacturer; it is a bet on whether licensed Japanese IP can successfully break the Korean stranglehold on the high-margin IT display ecosystem.

Remember Micro-LEDs? They are the stuff of high-resolution displays that cost correspondingly high numbers, reaching into the hundreds of thousands of dollars.  Only a short few years ago they were going to revolutionize the display industry and replace both LCD and OLED displays with brightness levels unable to be generated with any other display technology.  As with most technology, the practical application is typically much more difficult than the hype and while Micro-LED display technology is still being developed and occasionally hits the retail market in some form, it has proven more difficult to manufacture than originally expected.
The problem is that Micro-LEDs are small, very small, and are typically grown on a layer of Gallium Nitride on Sapphire.  Once grown they must be removed from the wafer and transferred to a substrate, typically glass or PCB.  As there are three (R,G,B) sub-pixels for each pixel in a display, a 4K TV requires roughly 8.3 million pixels or almost 25 million micro-LEDs.  To make things even more difficult, red Micro-LEDs are grown from a different material than blue and green Micro-LEDs, so moving the Micro-LEDs from their growth wafers to a substrate means picking LEDs from at least two wafers.  Using a laser to lift off the Micro-LEDs from the growth wafer is now the industry standard but yield is also a major factor that can temper the advertised speeds of current transfer equipment. 
We note that while peak transfer speeds of 100 million to a billion Micro-LEDs/hour are advertised, the reality is that actual throughput is closer to 20 million transfers/hour, or a bit less than one 4K display.  An LCD 4K display comes of a mass production line roughly every 15 seconds, which points to why the cost of Micro-LED displays remains high.  Yield is also a consideration as even Five nines of accuracy for Micro-LEDs yields roughly 2,500 dead pixels per 4K device.
Does this mean that Micro-LEDs are not viable for displays?  Not really, as the display industry continues to make improvements in each of the key Micro-LED processes, but it does push out Micro-LEDs as a display application that can compete with LCD and OLED for a few more years.
The Shift to AI Infrastructure
So why spend time on Micro-LEDs?  Because there is another application that promises to breathe life into Micro-LEDs and (ta-da) it is associated with AI and looks to be much more “immediate” as a practical application.  The application, known as Micro-LED Co-Packaged Optics (CPO), is a substitute for copper cables and traces that act as interconnects between server racks and network switches in data centers.  Copper DACs still dominate the market, at between 45% and 50% of new port installations, especially for short (>1.5m) connections, and Linear Pluggable Optics (LPO) are running 2nd  at 25% to 30% of new port installs, but CPO is the most rapidly growing technology and represents the top of the technology chain.
 Here is how it works:
In typical silicon photonics a laser generates the ‘carrier light”, sort of a dial tone.  These lasers are external devices and remain “on” at all times.  This means the power consumption is predictable as is the wavelength (usually 1311nm).  In order to keep the 700 Watts of heat generated by the GPU from damaging the laser, it is housed in a separate module and pipes the light into the optics that are packaged with the GPU.  The GPU data triggers the CPO and the external laser transmits the modulated light down the fiber.  At the other end, the light is translated back to an electrical signal.
This works well but presents challenges:

• Heat – CPO requires that the laser be near the GPU.  This causes the laser wavelength to drift.  Too much drift and the receiver cannot “see” the data.
• Power – Lasers are always on and maintain a carrier signal even when no data is flowing which means it is power inefficient.
• Sensitivity – Lasers are sensitive to heat and need thermal protection
• Complexity – Lasers require separate ‘modulators’ that add to complexity.

CPO 2.0: The Micro-LED Advantage
So, if CPO has so many drawbacks, why use it?
Here’s where things have changed.  CPO 2.0 is based on Micro-LEDs instead of lasers.  Micro-LEDs are much more thermally stable than lasers.  Stable up to 125°C while lasers need isolation and cooling.  Both lasers and Micro-LEDs are difficult to repair but Micro-LEDs last longer.  As lasers are always on they require power at all times.  Micro-LEDs only require power for the time they are on and that time is a short pulse.  

Furthermore, Micro-LEDs are simple and do not require a separate modulator, and while lasers must be aligned to a 9um single mode fiber, Micro-LEDs are able to use 50um multi-mode fiber, making the alignment process considerably easier.  Lasers run “Narrow & Fast” meaning they run at 100Gbps per lane which generates large amounts of heat.  Micro-LEDs run “Wide & Slow”, ~2-4Gbps across thousands of lanes., essentially parallel processing the data.  The bottom line is that most drawbacks of lasers in this environment are opportunities for Micro-LEDs.
Right now, CPO 2.0 (Micro-LED) is more expensive than CPO 1.0 (laser) on a per link basis, but that premium is the result of a more mature laser supply chain and will diminish over time, with estimates for late 2027 as the point at which it will be cheaper to produce CPO 2.0 than CPO 1.0 when including yield improvement.
But it doesn’t end there.  A standard 1.6T laser based module consumes ~30 Watts of power.  That same equivalent link using CPO 2.0 (Micro-LED) consumes ~1.6 Watts.  Using the example of a 100,000 GPU cluster, moving from CPO 1.0 to 2.0 saves approximately 15 million kWh per year, which comes to a savings of ~$1.5 million per cluster.  Additionally the cooling cost of CPO 1.0 is significantly higher than 2.0.  Lastly the cost of multi-mode fiber is roughly 40% cheaper than single-mode, adding to the cost savings for 2.0 over time, so over a 5 year period, while the 2.0 install cost $1,000 more than 1.0 the energy savings over that period would give the 2.0 install a $5,000 per port advantage.

​Conclusion: The Diversification of Micro-LED and the End of the ‘Carrier’ Era
The evolution of Micro-LED from a high-end display prospect to a critical AI infrastructure component represents a fundamental shift in how we think about the energy-per-bit required to fuel the AI revolution. While the display industry continues to refine the massive transfer processes required for consumer screens, the data center has provided an immediate, high-value application that bypasses those scale barriers. This transition from CPO 1.0 to 2.0 is more than just a component swap; it is a permanent shift in the economic calculus of high-speed networking.
As of March 2026, the landscape has changed. With the signing of the Ratepayer Protection Pledge at the White House this month, the era where Big Tech could socialize the cost of inefficient energy use has ended. Hyperscale’s are now directly responsible for every megawatt of new generation they require. In this "pay-to-play" energy landscape, the 10-month break-even period of Micro-LED optics isn't just a competitive advantage—it’s a survival requirement.
By moving from the "Narrow & Fast" carrier-based laser models to the "Wide & Slow" parallel architecture of Micro-LEDs, data center operators are finally scaling past the thermal and financial limits of silicon photonics. Micro-LED may still be the future of the living room, but today, it has found its true calling as the essential, invisible nervous system of the global AI economy—proving that the most disruptive innovations often find their greatest success in the places we least expect them.
 

​The Multimodal Promise
Smart glasses give the user a number of opportunities and Meta’s (FB) Ray-Bans (with display) are the quintessential product.  Depending on the model, you can look at an object and ask the AI, “What is that?”.  The Ai can see what you see and figures out what the object is and provides details.  Of course you can take it even further and ask where it can be purchased, whereupon the glasses will give you visual ‘cards’ with store names and addresses, and in some cases you can purchase the item directly through Instagram Shopping or Facebook Marketplace, although the actual checkout would be on your phone.
 
Capture Mechanics and the Transparency Light
Users can also take images or video while wearing the glasses, and to make it simple for the user, there are three ways in which to do this.

• Press the “Capture” button on the right are of the glasses and hold for a second or two
• You can say “Hey Meta, record a video”
• If you are using the Meta neural band you can double tap your index finger to your thumb

When you record with the glasses Meta has included a white LED on the front face that illuminates when the camera is in use.  This allows image or video subjects to know when and for how long they are being recorded and makes it difficult to ‘secretly’ record someone without their knowledge.  That said, the Meta AI is always ‘listening’ for the words needed to put it into action, but the glasses do not capture images or video without the express instruction from the user.  The camera is not “always on”, meaning if you want the AI to identify something, you have to ask it and the Ai will then turn on the camera, snap an image, and send it to the cloud for processing.  If the camera was on all the time the battery would expire in a short time and the amount of computational data processing would be staggering.
Super Sensing: From Active Inquiry to Passive Observation
However, Meta has recently introduced “Super Sensing” (aka “Live AI”), which enables a more continuous flow of data if you have it enabled.  In fact, the glasses take a picture every few seconds so the user can have an ongoing conversation with the AI about what is being seen, without having to re-activate the AI “Hey Meta”.  The recording light stays on or pulses during this mode, but it seems that sometimes users forget or don’t notice that the glasses are still taking periodic snapshots, even after they put the glasses down.
The Human-in-the-Loop: Data Annotation in Kenya
As the snapshots in this mode are being sent to the cloud, Meta has been using some of these images and videos to train its AIs, and that means that some of the images have to be reviewed by human contractors in order that they can be labeled for the AI (Yellow dog on green grass, green boat in blue ocean, etc.).  A recent controversy has developed over the fact that these 3rd party contractors are therefore viewing what is personal data, which has put privacy advocates up in arms.  A major investigation by Swedish newspapers (Svenska Dagbladet and Göteborgs-Posten) just last month revealed that human contractors in Kenya have been reviewing thousands of "accidental" recordings. These include:

• Intimate Moments: Footage of people in bathrooms, getting undressed, or in bedrooms.
• Sensitive Data: Clear images of bank cards, ATM PIN entries, and computer screens with private emails.

Persistence and Triggers: The "Accidental" Recording Loophole
This can happen in two ways:

• Session Persistence – Sometimes the “Super” session does not end correctly and continues to record snapshots periodically.
• External Triggers – As the glasses are always listening, background noise can accidentally trigger a session.

Ownership vs. Licensing: The Terms of Service "Tax"
While there is a difference between legal ownership and usage rights, the reality is that to the user, they are the same.  You, as the creator of the image or video, own a US copyright and Meta agrees that you retain ownership of your “User Content”.  However when you put on the glasses and use the cloud-based AI features, you give Meta a “License” to everything you create, along with its partner Luxottica (pvt).  "perpetual, non-exclusive, royalty-free, transferable, and sub-licensable worldwide license", which means they can store it on their servers, process it for their AI models, share it with 3rd party contractors, and use it to improve their AI models. 
Meta updated its policy in January to explicitly state that interactions with Meta AI (including the visual data from your glasses) can now be used to personalize ads across Facebook and Instagram.  This “intent” data can be used to personalize ads in the future and also allows the data to be reviewed by humans and even if you delete your activity logs, the insights generated in your personal advertising profile will remain.  Meta does let you opt out of some things.  You can turn of the storage of AI interactions, and you could always disable cloud processing, but that would remove much of what the glasses are able to provide, which leads most to scroll through the privacy notice and say yes to it all.
The Legal Challenge: Challenging "Designed for Privacy"
That said, not everyone feels that way, and last week Meta received a proposed class action lawsuit in the US District Court of the Northern District of California – San Francisco.  The lawsuit seeks to hold Meta responsible for what it alleges is “affirmatively false advertising and failure to disclose the true nature of surveillance and its connection to the company’s AI data collection pipeline”. It further alleges that Meta’s conduct violates state consumer protection laws and offends basic notions of privacy.
The underlying issue in the suit is Meta’s advertising motto for the glasses, “Designed for privacy, controlled by you”.  The suit points to product privacy pages in ads and product description that state, “You’re in control of your data and content” and “Clear, easy device and app settings” to “help you manage your information, giving you control over what content you choose to share with others, and when.”  Further, pointing to the Meta AI Glasses website, which offers tips for consumers on “How to wear your Ray-Ban Meta glasses responsibly,” in order to make others feel “safe and comfortable while you’re wearing your glasses,”
The suit points to how Meta feeds data, captured by consumers through their Meta AI Glasses, to a subcontractor.   At the subcontractor’s headquarters in Nairobi, Kenya, thousands of people working as “data annotators” review this sensitive user data, annotating it with labels like “cars,” “lamps,” and “people” in order to train artificial intelligence used in Meta products.  These annotators report seeing deeply private video clips, including videos of bathroom visits, sex, and other personal moments.  
One person recounted seeing a video wherein a man placed his Meta AI Glasses on a bedside table and left the room. Afterwards, his wife entered the room and, unknowingly, changed clothes in front of the Meta AI Glasses.  The video was sent to people halfway around the world to view and annotate in order to train AI, all without the woman’s knowledge that the video of her had been captured in the first place.  Other data annotators report seeing videos with visible bank cards, private text exchanges, and other private information. This sensitive data lives in Meta’s database, viewed by human data annotators and is available indefinitely for use to Meta’s AI products.
The plaintiffs are asking for an injunction, a disgorgement of profits, and punitive damages, as one would expect, but this is less an AI oriented issue and more an issue of how strong Meta’s disclaimers and privacy policies are.  We expect the real challenge will come from the ‘implied’ limitations in the literature that accompanies the glasses.  If that verbiage is too broad the plaintiffs would have at least a case for misrepresentation, but more likely Meta will add additional wording to the product disclaimers,  and perhaps add more “Please acknowledge” boxes that covers the 3rd party viewing of ‘accidental’ personal information.
Conclusion: The "Privacy-by-Design" Paradox
The emerging legal battle in the Northern District of California highlights a fundamental friction between AI utility and user expectation. At its core, the lawsuit suggests that Meta’s "Privacy-by-Design" marketing may have created a false sense of digital isolation. While the hardware includes visible indicators like the white LED, the software architecture relies on a "human-in-the-loop" pipeline that many users find incompatible with the intimate settings in which the glasses are worn.
The outcome of this litigation likely hinges on three critical factors:

• The "Informed Consent" Standard: Meta will argue that its terms of service technically disclose the possibility of human review. However, the plaintiffs are betting that "buried" disclosures cannot override a high-visibility marketing campaign that explicitly promised users were "in control."
• The Failure of Anonymization: Evidence that facial blurring and data filtering frequently fail, especially in the low-light or high-motion environments where "accidental" recordings occur.  This can undermine Meta's primary defense that they take reasonable steps to protect identity.
• The "Super Sensing" Liability: The continuous nature of "Live AI" creates a unique risk where the device transitions from a purposeful tool to a passive observer. If the court finds that "session stickiness" or "external triggers" led to the capture of intimate data without a clear affirmative act from the user, Meta may face significant punitive damages.

Ultimately, this case may force a regulatory shift in how wearable AI is marketed and managed. We expect that regardless of the immediate legal verdict, the industry will move toward "Local-First" AI processing, where identifying and labeling occur on the device itself rather than in the cloud, effectively cutting the cord to offshore human reviewers, although that has its own drawbacks. Until then, the Meta Ray-Ban glasses remain a powerful example of the "AI tax": a trade-off where the price of a more helpful assistant is the potential presence of an invisible, human audience.
 

The Timing of the "Price Reveal"
New TV models from major brands are typically announced at the Consumer Electronics Show in early January, but pricing is not usually revealed until March or April, right around when they become available for pre-order.  There are a number of reasons why TV set brands hesitate to reveal prices early in the new year, some of which are fiscal and some psychological.
The “Price Reveal” pulls the curtain away from the hype that surrounds the new models early in the year and adds a sense of reality that allows the consumer to make a judgement as to whether the propaganda is worth the money.  In fact, almost all brands have revealed 2026 TV set prices, with only Samsung Electronics (005930.KS) the holdout at this point in the year.  That said, LG Electronics (066570.KS), who revealed the pricing for much of their flagship and mid-tier TV sets recently, is holding back the pricing on their budget tier sets, likely until they get an understanding of Samsung’s pricing.
LG’s Premium Strategy: Holding the Line on OLED
The most important point relating to LG’s TV set lineup is the pricing itself, which many expected to increase as the cost of memory and other components continues to rise into the new year.  In fact, LG has kept initial TV set prices almost entirely in-line with last year’s initial set pricing, with only one model size different than last year, and that price is actually lower than last year.

The LCD Defensive: Aggressive Cuts in Non-Premium Tiers
However, LG’s TV set lines are more than just OLED.  The company also offers Mini-LED TV sets, Quantum dot enhanced LCD TVs, and generic LCD TVs, and the pricing on these “Non-Premium” lines is even more different from expectations.  With only one exception (Flat), all of LG’s Mini-LED TV sets, both this year’s high and low price tiers), are priced below the initial price these sets were offered at last year, antithetical to conventional wisdom.
 

Taking this unusual circumstance further, when looking at LG’s more generic LCD TV line, both with quantum dots and without, the discounts to last year’s pricing are even greater, averaging almost 11%.  We do expect the pricing for the budget tier of LG’s OLED line to remain flat, at the least, but depending on Samsung’s competitive positioning, it could fall below last year’s initial prices.

​Industry-Wide Trends: BOM Pressure vs. Brand Identity
As noted, as the largest TV set brand globally, the TV set pricing data from Samsung is quite important for the industry.  Samsung is quite aware of this fact and holds out its reveal as long as possible to keep competitors guessing, but even without Samsung we can poll other large brands and see if they follow LG’s lead or if the trend is in other directions.
We note that the data for each brand is tempered by BOM pressure as noted, but also by brand strategy, which can have a significant effect on new year TV set pricing, along with trying to manage 2025 remaining inventory.  Higher initial prices will drive consumers to last year’s models, while lower prices will bring them to the new line of sets  2026 is already shaping up to be a year of ‘influences’ with even more factors pushing brands in one direction or the other.  Tariff issues and transportation headaches have already begun to complecate an already complicated CE space.  That said, we collected data similar to LG’s data for four other brands which we summarize below.
Competitor Breakdown: Five Different Paths to 2026

• LG - Defensive Stability & OLED Protection
  • Total Weighted change - ↓2.13%
• LG is focused on protecting its premium OLED market share. They kept Flagship (G-Series) and Mid-Tier (C-Series) pricing nearly identical to last year (0% to -2%). However, they are aggressively cutting prices in their LCD and QNED (QD) lines (-9% to -13%).  LG is signaling that OLED is a premium and stable investment while using their lower-end LCD sets to fight off the aggressive price-cutting from TCL and HiSense in the mass market.
• Sony (SNE) - Premium Aspirations & Margin Preservation
  • Total Weighted Change: +2.97%
• Sony remains relatively stable overall. While their Flagship models saw nearly a 20% jump, their "Elite" and "Performance" tiers stayed flat at 0%, which balanced out the total brand impact.  Sony is betting on superior processing and brand equity to justify higher entry prices for their newest tech. By keeping other tiers flat rather than cutting them, they are prioritizing profit margins over raw volume, refusing to join the "race to the bottom" that can typify TV set pricing.
• HiSense (600060.CH) - Aggressive Up-Tiering
  • Total Weighted Change: +8.55%
• Despite some price cuts in their "Value" tier, HiSense has the highest overall increase. This is driven by significant double-digit price hikes in their high-volume "Performance" and "Mainstream" categories.  HiSense is attempting to shift its brand image from "budget" to "premium performance”.  It seems they are using their low-end sets to attract customers but expect to upsell to models that carry higher margins.
• TCL (000100.CH) - "Big Screen" Pivot
  • Total Weighted Change: -1.87%
• TCL is showing a slight downward trend for the brand. Their huge price increases in the "Ultra-High" (98"+) segment were almost entirely offset by aggressive price drops across their Budget, Entry-Mini, and Performance lines.  TCL is banking on the "Bigger is Better" trend. By making 65" and 75" sets much cheaper, they drive volume; by raising prices on the 98" and 115" sets, they capitalize on their unique manufacturing advantage in giant panels where there is less competition.
• Vizio (WMT): Market Share Recovery
  • Total Weighted Change: -16.40%
• Vizio is the only brand in the set with a double-digit decrease across the entire brand. They are cutting prices across every tier that had year-over-year data, signaling a very aggressive value-focused strategy for 2026.  Vizio is in full "disruption mode" to claw back market share lost to TCL and HiSense.  They are aiming to be the default choice for the price-conscious consumer at big-box retail, likely prioritizing unit volume and platform growth (SmartCast) over hardware margins.  While likely a necessary early strategy to identify the brand with its new owner, the longer-term prospects for that strategy are less robust..

Conclusion - Bifurcated Markets: Margins vs. Market Share
The 2026 "Price Reveal" has effectively unmasked a deeply bifurcated industry strategy. While the narrative early in the year focused on BOM (Bill of Materials) pressure and rising component costs, the actual pricing data suggests that brand strategy is currently outweighing fiscal necessity. We are seeing a market split into two distinct camps: those prioritizing brand elevation and margin preservation, and those engaged in an aggressive pursuit of market share.
Sony and HiSense represent the "margin first" contingent. Sony is leaning heavily into its heritage as a premium Tier-1 brand, betting that consumers will pay a nearly 20% premium for flagship processing. HiSense is perhaps making an even bolder move; by aggressively up-tiering their mainstream pricing by over 8%, they are signaling an end to their "budget brand" identity, attempting to move into the vacuum left by brands that have historically held the mid-to-high ground.
Conversely, Vizio, TCL, and LG are playing a defensive volume game, though with different tools. Vizio’s double-digit collapse in year-over-year pricing is a clear "shock and awe" campaign intended to re-establish the brand's footprint under its new ownership. TCL continues to leverage its vertical integration to commoditize the 65” to 75” market while keeping a premium "tax" on the ultra-large sets they alone can manufacture at scale. LG, meanwhile, is effectively using a "shield and sword" strategy—using stable OLED pricing as the shield to protect their premium image, while using their LCD line as the sword to slash into the mass-market share of their Chinese competitors.
The Samsung Pivot: Waiting for the Final Piece
The final piece of the 2026 puzzle remains Samsung. As the lone holdout, Samsung’s eventual pricing will act as the ultimate pivot point for the year. If they follow Sony’s lead toward margin expansion, it will provide a much-needed "pricing umbrella" for the rest of the industry. However, if they choose to follow LG and Vizio into a price war to protect their #1 global position, 2026 will quickly transform from a year of "strategic influences" into a race to the bottom that will test the fiscal endurance of every brand on this list.  We will know in about a month.

Over the last year AI hype has transformed from generative AI, the Chatbots we were all so impressed with months ago, to Agentic AI, those snippets of code that operate behind the curtains and allow us more ‘refined’ cognitive pastimes while they slog through the daily grind of e-mails, memos, and meeting notes. 
Generative AI is a collaborative tool, with the interaction between the model and the user the key to success.  We think of generative Ai as being a somewhat over-eager assistant who is willing to do almost anything you ask but can be a bit careless.  Said assistant, who has no experience, sits with you day after day, eager to help you, sometimes so much so that they can become a bit effusive.  When assigning a task to your generative assistant, you must be very precise and specific about details as the more specific you are the greater the chance that the AI will give an answer that is specific to your query and correct.

The Shift from Chatbots to Agents
Agentic AI can be thought of as the  assistant you have had for years. One for whom might look messy and disorganized but one you can leave a simple note and know the work will  be done while you are out of the office and done just the way you like it.  In fact, sometimes you don’t even have to leave a note, and the work gets done without you input.
While there is relatively little data compiled on agentic tasks at this early stage, the most common task for agentic AI has been summarization.  The common “TL;DR” (Too long; Didn’t read”) shows a generational frustration with long e-mails, You Tube videos over 30 seconds, legal documents, or sadly, books.  Agentic AI is used to avoid even seeing those items in their long form and just reading the summaries as they appear on your desktop.  Agentic AI is also used for classification and triage.  Automatic e-mail and memo sifting/tagging is a common agentic function, allowing all the ‘unimportant  e-mails to fall into the ‘junk’ bucket for a quick human review before they are deleted, while the ‘important ones are elevated to the top of the stack, all before you even see them.
​
The "TL;DR" Culture and the Rise of Triage
Agentic systems come in neither a blanks slate, available for task assignment, nor a pre-programmed package.  As a reminder, AI systems are tools and tools either need to be initially developed with a purpose in mind (a hammer has two-pound nails in and pull them out) or taught a purpose, but agentic AI splits the difference. Here’s how:

• The Model - Most Agents start  with a foundation AI model.  The model has been taught lots of things, everything from physics and literature to writing Python code, but as it stands it can do nothing on its own.
  • The Framework – The model is wrapped in a framework that provides pre-programmed ‘instincts.  The framework adds the following:
    • Planning Loop – This gives the model a path to sequentially follow tasks (“Do X, then Y, then Z”)
    • Handshake – This code allows the model to connect with your computer, the internet, and  other company or private resources that you access on a daily basis.
    • Memory – A “Yellow Pad” so the agent can remember what it learned when it finished task 1 and will not forget it when it finishes each successive task.
  • Job Description – As an agentic AI user, this is your stage.  You define what you want the agentic AI to do for you.  Here are some examples:

Scheduling & Logistics
Your Request - "I need to meet with the Marketing team and our external vendor, Sarah, sometime next week for 45 minutes. Find a time that works for everyone, send the invites, and if Sarah hasn't replied by Friday, send her a polite nudge."
Agent Action - It checks your calendar, emails the team to find their availability, cross-references Sarah’s time zone, drafts the calendar invite, and sets a "watch" on its own internal clock to follow up if it doesn't see a "Meeting Accepted" notification.

Deep Synthesis
Your Request - "Our competitor just released their quarterly earnings report. Compare their growth in the EU sector to ours, put the data into a table in a Google Doc, and Slack me the three biggest risks you see for our Q3 strategy."
Agent Action - It downloads the competitor’s PDF, scrapes your company’s internal database for EU sales, performs the math, creates the document, and sends you a summarized Slack message with the "Top 3 Risks."

Data/Lead Triage
Your Request - "Go through the 'Contact Us' form submissions from the last 24 hours. If any mention a budget over $50k, add them to Salesforce and assign them to Jim. If they are just looking for support, draft a reply with a link to our help docs and archive the ticket."
Agent Action - It reads the incoming text, performs a "Sentiment and Intent" analysis, logs into Salesforce to create the lead, and interacts with your email API to send the support responses.

System Maintenance
Your Request - "I’m getting reports that the client portal is slow. Run a diagnostic on the server logs from the last hour, identify any IP addresses with weird traffic spikes, and if you find a potential DDOS attack, temporarily block those IPs and alert the security team."
Agent Action - It accesses the server via a secure terminal, parses thousands of lines of log data, identifies patterns, executes a "Block" command on specific IPs, and sends a high-priority alert to the human team.

Real-World Delegation: Scheduling, Synthesis, and Triage
With agentic models you can think of yourself as the CEO (unless you already are).  The model is the office manager and those tasks shown above, which can be captured in  “Scheduled  Scripts” are your employees.  You can have them repeat some tasks regularly.  You can call up a more specific task when necessary, or you can create a new one for a specific instance.
When you use generative AI, unless you have an account with that Ai’s provider, the AI itself only knows the content of your query.  The hosting platform collects things like your ip address, the type of device you are using, the browser type, your screen resolution, and places a cookie that temporarily puts an id on your computer, all similar to what is collected when you visit sites on-line.  The generative AI retains the conversation as it progresses so it is able to answer follow-up questions that are part of the chat, but when the chat ends the Ai itself will delete the conversation.  Most platforms will retain the chat for anywhere from 72 hours to 30 days.  This is the limit of what the AI knows about you unless your account with the AI provider, allows you to retain your chats for your own review.
Agentic Ai is very different.  It must have a persistent memory, and it must be integrated into whatever system is using it.  The persistent memory comes in the form of a vector database that builds up over time, eventually becoming a 360° profile of you, the user.  It remembers that time you specified a certain brand, a certain file type, and which tools you prefer to use, but it goes much deeper than that.  It learns what tasks you trust it to do on its own, which newsletters you like to read and which ones you keep.  It tracks your response cadence, the speed at which you respond to various communications.  It learns your “tone” or writing voice and will match it when writing replies, and it can make internal decisions about the priority of those preferences based on simple things like whether you saved all e-mails about a certain topic.  It can learn about your relationships and can understand that a particular contact can be significant to other contacts in your file.  It can see relationships in your daily life that you do not always see, like connecting a PDF in your “download” folder to an e-mail thread from weeks ago because they shared a particular project name or code.

The Privacy Paradox: The 360° User Profile
The bottom-line is, for agentic AI to work well it has to know all about you, and for some this might seem a particularly onerous invasion of privacy.  While a digital intrusion its is not much different than the years of observing your habits that a long-time assistant might make.  While that slice of digital privacy might be problematic for some it is not what we worry about.  Our issue is that once you give the keys to the castle to someone else they can be stolen, and while agentic AI is still in its infancy, such pilferage has already begun.

The Security Crisis: Poisoning and the "Keys to the Castle"
Agentic AI can be attacked in two ways, either on a large scale or a small one.
Large – A patient hacker can ‘poison’ the AI’s training data by adding just a small amount of data.  This is a small number of documents in a massive dataset, but it teaches the AI to associate a certain phrase with a command to bypass security filters.  The AI operates normally until the hacker sends an e-mail or document to the model that contains the trigger phrase, and the Ai follows by dropping its security and doing whatever the hacker specifies.  Recent studies have indicated that it takes only poisoning 0.01% of a database to plant a permanent backdoor in the model.
Small – In this situation the hacker sends a seeming innocuous PDF or e-mail directly to you.  Hidden in the metadata or even printed in white so as to be invisible, the hacker includes instructions that poison your model in a similar fashion  As the agent ‘reads’ your e-mail it absorbs the backdoor instructions, such as including a hidden “CC” to the hackers location  for every e-mail you send.  Taken further, the hacker could instruct the agentic Ai to send him your vector file, and while it might seem like just a list of coordinates, there are tools that allow that data to be converted into plain text, essentially opening up every aspect of your life, business, decision you have made since the agentic AI was installed.  This isn’t just name-address-password, it is everything the agentic AI has collected about what you care about, what you do on a daily basis, and what you don’t.  What could be better to create a false you?  To add insult to injury, what do hackers use to unravel the vector databases they steal? Another AI model.

Conclusion: The Double-Edged Sword of Autonomy
The transition from Generative to Agentic AI represents the most significant leap in personal productivity since the advent of the personal computer. We are moving away from a world where we struggle to keep up with our digital lives and into one where our software finally works as hard as we do. By sifting our inboxes, managing our logistics, and synthesizing our data, these agents offer us the most precious of modern commodities: time. However, as we move from "Instruction" to "Delegation," the stakes change. To be truly useful, an agent must possess the "keys to the castle", access to our files, our schedules, and our very way of thinking. This intimacy creates a paradox of progress: the more an agent knows about us, the more effective it becomes, but the more devastating a potential compromise becomes.
The future of Agentic AI will not just be won by the most capable models, but by the most secure frameworks. As we embrace these digital proxies to handle our "daily grind," we do so with eyes wide open, recognizing that while an agent can be our most powerful ally, its persistent memory and deep integration require a new standard of digital vigilance. In 2026, the goal is no longer just to build an AI that can think, but to build one that we can trust to act in our name without opening a back door to our lives.

The End of the Initial Purchase Era
In 2024 the government of China was faced with a dilemma.  The country’s population had been rising in affluence since the early 2000’s and the economy had seen massive growth.  However by 2017/18 consumer electronics’ products, bought by this new Chinese middle class, were up for replacement, and CE growth transitioned from initial purchase growth to a more normal replacement cycle.  The value of purchases continued to increase as middle class affluence continued to grow, but in 2017 unit volumes in flagship products began to decline for the first time.

The Urban-Rural Connectivity Gap
By that time ~98% of the Chinese population had at least a mobile phone and the transition from a ‘dumb’ mobile phone to a smartphone had already reached between 55% and 60% of the population.  This was driven by the urban population’s desire for better connectivity and more “user-experience” oriented applications.  However the fall in unit volume, for the first time since China had been collecting such data, was a head’s-up for the Chinese government.  There were other factors that precipitated this change, such as the over 98% of the Chinese population who accessed the internet through their phone, and that smartphone conversion was now moving from the urban population to more remote regions, but the unit volume outcome was a blow to the growth that the Chinese government had been touting for years.. 
​
The "Urban Bubble" and the Delayed Response
Chinese government officials lived in the same urban environments as affluent consumers and were sheltered somewhat from the slowing smartphone growth.  They were aware that the growth of more mature CE products, particularly appliances, had been slowing and put in place some relatively narrow subsidies for these items in the early 2020’s, but it was not until 2024 when smartphones were beginning to be considered as part of the government’s  “New for Old” program.  In fact, smartphones were not included until January 2025, after almost 8 years of declining sales.
From Green Energy to National Solvency
That program, which began in a slightly different form in 2024, was originally intended, at least so the Chinese government says, to focus on energy efficiency.  Initially 8 home appliances were targeted:

• Refrigerators
• Washing Machines
• Televisions
• Air Conditioners
• Computers
• Water Heaters
• Household Stoves
• Range Hoods

Along with the automotive sector, with a specific focus on all-electric or plug-in hybrids, the point of these subsidies was that ‘new’ models would be more energy efficient and lower the broad Chinese energy use profile.  In fact, the subsidy was based on the energy efficiency of the ‘new’ device, with a 20% subsidy for those devices with a Grade 1 designation (Most Energy Efficient) and 15% for those with a Grade 2 (Less Energy Efficient).
​
The Fiscal Pivot: Centralizing the Burden
In the early days of the subsidy (2024) much of the funding was done through local governments who fronted the capital.  This local emphasis was the reason behind the program’s slow start and considerable friction between provincial governments and the Central government, but in July of 2024 the Central government issued 150 billion yuan ($20.7 billion US) of ultra-long special treasury bonds (20, 30, and 50 year).
The Central government changed the plan at that time from a roughly 50/50 burden on the provinces, to a general 1:9 ratio, with the central government now bearing most of the program cost.  In actuality, the percentage changes based on the prosperity and location of the provinces involved, with less prosperous provinces in the 5% range and the more affluent in the 15% responsibility range.
We note, in the table below, that the consumer allocation percentage of the float is less than half of the float (and spend) with the larger allocation going to industrial equipment.

Future Tech and the Aging Demographic
This year’s program will expand further, with the NDRC indicating that while 2024 and 2025 was oriented toward unit volume (moving inventory), 2026 will be more about “Future Tech” and the “Silver Economy”.  Future tech is easy to understand.  It includes new items like:

• Smart glasses (AI/AR) (15% rebate capped at 500 yuan ($72.65 US)),
• Fitness bands (not previously included) and some
• Health related products

But the “Silver Economy”[1] bears a bit more scrutiny.  The official definition not only specifies senior citizens but also leans toward those in their 40’s and 50’s, who have to “…prepare for the challenges of an aging population”.  The government takes this new philosophy even further vowing to build 10 dedicated “Silver Economy” industrial parks across the country to ‘cluster’ companies specializing in elder-tech.  It is also a central theme in the 2026 – 2030 Five-Year-Plan that is currently under development and includes age-friendly home renovations, hardware like heart or sleep monitors, and, of course, companion or exoskeleton robotics.  We expect these items will find there way into the subsidy program at some point.

Mandating Quality: The War on "Involution"
At the same time the government also narrowed the number of home appliance categories from 12 to 6, eliminated most Grade 2 subsidies, and lowered the caps on high-end appliances from 2,0900 yuan ($290.61 US) to 1,500 yuan ($217.96 US) to open the funding to a large population swath.  That said, in a more uncharacteristic move, the central government also implemented in the 2026 program what it calls “Anti-Involution”.  Involution is a Chinese word for a “race to the bottom”, the point at which companies are so willing to compete on price that profitability disappears, innovation is stalled, and workers suffer.
New “Price Conduct Rules” that are part of the 2026 subsidy plan make sure that manufacturers do not sell product below cost to capture share and use these rules to rein in ‘disorderly’ low-price e-commerce competition by blacklisting violators for 6 to 12 months.  It seems that the central government, after years of idly standing by while Chinese companies undercut foreign businesses by selling at or below cost to gain share, they seem to notice that it also has the effect of causing deflation, and when that is internal to China it becomes a very sensitive subject  A number of Chinese smartphone brands are very familiar  with the issues surrounding price “conduct” and are well-known for those tactics.

The AI Filter: Subsidizing the Next Generation
This year (2026)  the requirements go even further.  Any smartphone looking to generate a 15% subsidy must meet the following three new requirements relating  to AI:

• Dedicated NPU – The phone must have a stand-alone NPU (Neural Processing Unit) and cannot rely on Ai produced on the phone’s general CPU or GPU.
• On-Device Processing – With variations, this requirement states that the phone’s hardware must be capable of running a 7-billion parameter model without the cloud.
• Memory – All phones must have at least 12GB of RAM, forcing budget phones, which are typically 8GB out of subsidy territory.
• Domesticity – In order to receive a subsidy the phone must be on the central government’s approved list.  To maintain a place on that list it must contain a domestically-produced processor or a compliant one.  Typically Huawei’s (pvt) Kirin processors, Xiaomi’s (1810.HK) Surge, and certain Qualcomm (QCOM) Density processors, along with AI framework processors like those from Baidu (BIDU) (Ernie) and Alibaba (BABA) (Qwen)

The Compliance Trap: Apple and the Regulatory Pincer.
In the 2nd half of this year the Chinese government is expected to release a set of AI Handset Standards (2.0) that will mesh with the subsidy requirements.  The focus here will be AI privacy, essentially keeping AI from leaking sensitive user and corporate data.  This will be significant for Apple (AAPL), as the company will find itself caught between the strict subsidy price caps and AI compliance issues.  Apple is able to pass the NPU requirement as both the A19 and A19 Pro chips in the iPhone 17 series have a neural engine that qualifies.  The iPhone 17 Pro, the Pro Max, and the iPhone Air all meet the RAM requirement (12GB) but the basic iPhone 17 does not with 8GB, meaning it can be cut out of the subsidy pie if the memory is not changed, a margin question for Apple, especially given the rapid rise in memory pricing.  Apple also faced the price cap rule (6,000 yuan) for the iPhone 17 in 2025, but was able to ‘adjust’ the initial price to 5,999 yuan last year, making the 500 yuan subsidy available to this basic model in China. 
But there is another more fuzzy issue, that referring to Apple Intelligence.  Apple has been forced to partner with Alibaba and Baidu to ‘backend’ Apple intelligence in China and  Apple’s AI must be able to refuse to answer 95% of the 2,000 questions asked by the government concerning sensitive political issues.  Apple is still in the testing phase, but could find itself in an “all”, “none”, or “partial” mode where it might not be able to access subsidies in every province.  Apple can assuage these potential problems by offering higher trade-in values for other Apple products to offset the loss of the subsidy, but it is still up in the air about the final level of compliance.  On the opposite pole is Transsion (688036.CH).   Transsion  is probably the most well-known for the price cutting tactic mentioned above, as the company dominates the low-end market in China and Asia, but has the ability to rein in its aggressive pricing nature to capture this more draconian subsidy program.
​
Conclusion: Engineering the Upgrade
The shift from 2024’s volume-based subsidies to 2026’s "Future Tech" and "Silver Economy" mandates represents more than just a retail stimulus; it is a calculated attempt to force-march the Chinese consumer into a domestic, AI-integrated ecosystem. By tying financial incentives to high-threshold hardware specs, like the 12GB RAM floor and dedicated NPUs, Beijing is effectively de-platforming the "low-quality" growth that defined the previous decade.
For domestic champions like Huawei and Xiaomi, these rules provide a subsidized path to premiumization. For foreign players like Apple, the path is narrower, requiring a delicate dance between maintaining global brand margins and meeting increasingly localized, "fuzzy" AI compliance standards.
Ultimately, the "New for Old" program's success will be measured by whether it can truly break the cycle of "Involution." If the government can successfully pivot the middle class from "buying cheap" to "buying smart," they may finally solve the deflationary puzzle that has haunted the post-2017 market. However, with the central government now bearing 90% of the fiscal weight through ultra-long treasury bonds, the stakes are high.  Beijing is not just subsidizing a replacement cycle; it is betting the national balance sheet on the hope that AI and the Silver Economy can reignite a cooling economic engine.
As we move into the second half of 2026, the rollout of "AI Handset Standards 2.0" will be the ultimate litmus test. It will determine if China can create a self-sustaining, high-value tech market, or if it has simply replaced a "race to the bottom" with a high-priced, state-funded "race to the top."
 


[1] The official Chinese government definition of “Silver Economy” - "The sum of economic activities that provide products and services to senior citizens and prepare for the challenges of an aging population."

Yesterday Samsung Electronics (005930.KS) officially announced their latest flagship line of Galaxy smartphones, the Galaxy S26, the Galaxy S26+, and the Galaxy S26 Ultra.  Smartphone release announcements are typically a parade of incremental hardware updates and changes, culminating in the ultimate question, did the price change, and if so, is it worth it to upgrade?  While the price question always remains (Only the Ultra increased in price by $50 – The S26 and the S26+ remained the same), this year’s release, usually a highly focused bally-ho over changes that are almost negligible to most smartphone buyers, was oriented differently, which we see as a significant change.
Under the Hood: Moving Matrix Math to the Center
There was still plenty of focus on hardware changes, with much attention given to the upgrade to the chipset in the Galaxy Ultra.  While the Galaxy S25 Ultra sported the Qualcomm (QCOM) Snapdragon 8 Elite chipset (Technically the 8 Elite Gen 4), this year’s Galaxy Ultra moved up to the Qualcomm Snapdragon 8 Elite Gen 5 chipset,  This solved a heating issue that plagued the S25 ultra, but more significant was the fact that Qualcomm moved the hardware that performs matrix math (The parallel processing used by large language models) from a combination of the NPU (Neural Processing Unit) and the GPU (Graphics Processing Unit) to the CPU (Central Processing Unit) itself.  This resulted in much faster responses to AI requests, both internal and from the user.  In itself this was a plus, but most important was the focus on AI rather than more visible hardware improvements.
From Reactive to Proactive: The Rise of "Nudge"
This focus on AI continued, particularly with the S26 Ultra.  In most AI phones, when you need the AI to help with a particular function or application, such as when when someone asks for a copy of an image, you ask the AI to send it to that particular person with a query.  In the S26 Ultra Samsung has added “Nudge”, best described as an AI agent.  Now when someone asks for a copy of an image a ‘”nudge” icon automatically appears on the photos and a simple click on the icon allows the agent to take over the process of mapping out the steps needed to send the image to the person with whom you are messaging, and executes the process for you.  This would be called “Proactive AI” rather than “Reactive AI”.
The ”Proactivity” travels deeper as it is part of the Samsung One UI 8.5 operating system (Based on Android 16).  This means that instead of being a typical “Wait-and-See” system that remains idle until you click on an icon, it is constantly observing your local context in order to predict your next move.  The OS framework also contains “Hooks” that allow Gemini and Bixby to “click” buttons and navigate through applications, similar to agentic engines like Comet. 
Deep Integration: The Ecosystem "Hooks"
This “Nudge” system is able to hook into eight core internal applications:

• Gallery – The system makes proactive suggestions concerning specific photos when the context asks for them.
• Calendar – Offers to create events or check for conflicts when it sees a date or time on your screen.
• Contacts – Pops up contact cards or phone numbers during conversations.
• Reminders – Suggests setting an alert if you are discussing a deadline, project, or task.
• Smart Things – Offers to control home devices when things like “Coming Home” or “Sleep” are mentioned.
• Quick Share – Recognizes people in your photos and suggests sending them the file directly.
• Notes & Calculations – Brings up relevant notes or formulas that relate to what you are currently typing or viewing.
• 3rd Party Applications – Unlike competing “nudge-like” systems (Primarily from Google (GOOG)), as long as you are using the Samsung keyboard the Nudge system can hook into many 3rd party messaging application, with support already confirmed for WhatsApp, Instagram DMs, Telegram, Samsung Messages, and Google Messages..
• Other “Hooks” – Not only can Nudge make suggestions concerning 3rd party applications but is also able to perform tasks (Multi-step workflows), such as booking Uber (UBER) directly from a Nudge pop-up.  Throughout this year Samsung will be announcing further integration with Instacart (CART), DoorDash (DASH), and Amazon (AMZN), giving Nudge the ability to automatically fill a cart with items in a recipe you are looking at or source and purchase items on Amazon based on an image you are looking at on your phone.

The Friction Points: Privacy, Power, and "Clippy" Fatigue
While “Nudge” can be a time saver, there is some downside.

• Privacy – While the data Nudge uses is local to the Snapdragon chipset, some will be uncomfortable that the system is “reading” everything they write, including private messages, e-mails, and notes.  Discussions concerning legal or medical issues can still push Nudge to make what might be embarrassing suggestions if you are sharing the phone with someone else.
• Overload – Remember Microsoft’s (MSFT) “Clippy”, the virtual assistant that popped up on your screen?  The fact that Clippy is gone indicates that an overly aggressive AI could easily become an annoyance.
• Reliability – As Nudge must resolve conversational context, it can lead to errors for dates or incorrect photo suggestions, and more disturbing hallucinations are also a possibility.
• Power – The Qualcomm Snap 8 Gen 5 chipset is 35% more efficient, but even so, with Nudge on, the NPU is active at all times.  Some users have indicated that turning off Nudge gives an additional 45 to 60 minutes of screen time before recharge, so its use will drain the phone’s battery faster than normal.
• Lock-in – As the Nudge system is based on the Samsung keyboard, Nudge users are limited to Samsung devices, eliminating GBoard (Google) and Swiftkey (Microsoft).

As with any new system, consumers will have to decide how much it helps versus the drawbacks, but what is most significant to us is the fact that Samsung, known for its hardware expertise, seems to have shifted its focus to software, something we have been championing for some time.  The ability for smartphone brands to make significant hardware changes that push consumers to upgrade has diminished over the last few years as competition has increased.  This has continued to push  consumers to hold their phones for extended periods, leading to little growth in the smartphone market.  

​The Economics of Intelligence: High Margins over High Capital
Software does not have the high capital cost of hardware manufacturing, and consumer adoption to AI has given smartphone brands an opportunity that they have not had in a number of years, and Samsung seems to have taken the opportunity in a big way.  Google (Pixel smartphone line) has also been a proponent, as have Honor (pvt), Oppo (pvt), Vivo (pvt) and Apple (AAPL), each in its own way, although some are more hands on, meaning you have to invoke the Nudge-like application before it will perform.  The idea that software can be as much of an attraction to consumers as hardware is certainly not a new one, given estimates between $550 billion and $670 billion for the mobile software market, but brands have always relied on hardware as the selling point for new smartphone models.  The shift to software will test the mettle of brands that have relied on hardware for years, and if the brand enthusiasm carries through for another year or so, the change in perspective will represent a substantial shift in focus for smartphone brands.
Conclusion: A Catalyst for a New Mobile Era
In conclusion, the launch of the Galaxy S26 series represents more than just a new product cycle.  It signals a fundamental restructuring of the smartphone industry’s DNA. By prioritizing Proactive AI and agentic software over incremental hardware specs, Samsung is acknowledging that the "Hardware Era", defined by megapixel counts, multiple cameras, and screen resolution might have reached its peak.
This shift to an "Intelligent OS" model addresses the industry's greatest existential threat,  the stagnating global replacement cycle. As consumers hold onto devices for nearly four years, the value proposition must move from the physical device to the utility it provides. By integrating "Nudge" directly into the kernel of the OS and leveraging the Snapdragon 8 Elite Gen 5’s unique CPU architecture, Samsung is attempting to transform the smartphone from a passive portal into an active, autonomous partner.
If successful, this pivot will force a market-wide recalibration. Competitors will no longer be judged solely on their ability to manufacture hardware, but on their ability to develop deep software "hooks" and agentic ecosystems. In a software market worth upwards of $670 billion, the S26 Ultra isn't just a phone, it is a manifesto for the next decade of mobile computing, one where the "killer app" is the operating system itself. The industry is moving from selling metal to selling ease of use, and Samsung’s proactive turn may well be the catalyst that defines a new era.
 

The Great Digital Divide
We live in a divided world where polarization has become the norm and an open mind the exception.  While politics is the usual playing field for such a divide, the technology world has supplied us with a new one over the last few years, AI.  Some, particularly those developing the technology, believe AI is unlike any other technology advancement and will change the way humanity will evolve for centuries to come.  Others believe it is a scourge that will push humanity to a secondary position and eventually morph the human race into a herd of narcotized sheep.

The Third Perspective: AI as a Utility
But there is another contingent.  A group that looks at AI as nothing more than a tool that can assist mankind in its endeavors to both for the positive, to forge a better world or to the negative, to gain power over his/her fellow man, things that we have been doing since the dawn of man.  Having used AI on a daily basis since its early days of commercial existence, we fall into this middle camp and look at Ai as a “Tool with benefits” rather than the  dualistic earth-shaker that supporters of both camps seem to see.  Perhaps this is because, in our daily work, we see the points where AI can be successful and the places where it is not, but at the same time we have spent time conversing with AIs about this very question, and the AIs we have spoken with point out their good and bad points quite easily.

Defining the Nature of a Tool
Tools are devices that help mankind perform tasks.  They can be quite simple, as were stone cutting tools that helped man cut wood to build shelters or more complex tools like the plow that allowed scalable agriculture to develop.  But with the development of each tool the lives of the population were fundamentally changed.  Tools perform a function, usually a simple one that is fixed at the point when the tool is produced.  Some tools can do more than one thing, as a hammer can pound in nails and can pull them out, but it cannot adapt and perform other functions without being redesigned and rebuilt.

The Proponent’s Argument vs. Reality
Proponents say that AI is different.  It can adapt and change.  It can reason and learn, and above all, AI is probabilistic. making it more than a tool.  However, when we look at what AI systems are and how they work, it becomes more obvious that they are tools, sophisticated tools, but tools none the less.  Ai designers point to the fact that AI systems are robust, that they are able to make sense of complex data even when it is ‘noisy’ and includes irrelevant information and are able to build on their knowledge, but at least now, some of those claims are less than true.

Why AI Still Fits the Toolbox
Our goal is not to denigrate the development of AI or cast aspersions on those involved in their commercialization, but merely to point out that while AI is a valuable tool that we see as a major step forward, it is still a tool and while more adaptable than many tools, it still falls into the tool category and not the god-like creation some say it is.  Here is why:

• Usefulness – Most tools are able to function as soon as they are constructed.  AI tools only become useful when they are trained and are only as useful as the data on which they are trained.
• Perception - You cannot “trick” most tools as they have no perception of their use.  While some say that AI is perceptive, you can “trick” an AI system because its understanding is based on pixels and math, not on a real-world understanding of context.  Adversarial Machine Learning has proven that adding “noise” to an image, which is invisible to humans, can cause the AI to see a toaster as a school bus, as, in reality, the Ai is not “seeing a toaster or school bus, it is ‘calculating” a toaster or school bus.
• Fragility – There is a concept called Distribution Shift with AI systems.  The illustration would be if you teach an AI to sort red apples in a bright room and then show it red apples in a dark room, it might decide that the object is black coal.  Unlike other tools, the AI does not stop functioning, it gives a confident wrong answer.  This occurs because its knowledge is tied to a specific statistical distribution that it was taught, not the underlying reality of the object.
• Learning & Forgetfulness – Traditional tools are additive.  If you add a handle to a tool it does not forget how to work.  If you try to teach an existing AI something new, such as trying to teach a translation AI to do complex math, the AI can forget how to translate .  This is called Catastrophic Interference and has been a big stumbling block in moving AI from being a specialized tool to the general intelligence category.  This stems from the fact that AI systems do not store data in folders, like we do on computing devices.  They represent the connections between AI neurons as weights that represent how strong each of those millions of connection are.  When an AI system learns to recognize a cat, all of those weights relate to defining “cat-ness”.  If you try to teach that AI to also recognize dogs, the process begins to change some of those shared weights to “dog-ness” and the AI is less able to understand “cat-ness” and can often forget the original “cat-ness” altogether.

The "Update" Illusion
wait, we are often told that the AIs we are using are learning all the time.  In fact, Ai designers get around the forgetfulness issue by using RAG (Retrieval Augmented Generation), opening up the AI to a search engine.  The AI is then able paste the search data into its response, although the AI has not incorporated the search data in its weightings.  The ‘learned’ the data is only us to augment that particular answer and is deleted along with the chat when it is over.  There are other tricks that designers can use, but AIs need to be completely retrained in order to learn a new function.

Architectural Instability: The Shared Space Problem
The forgetfulness issue is a big one because it points to a major difference between AI systems and humans and what further points to the view that AIs are tools, not a god-like creation that will eventually overtake humankind.
AI system share space.  When an AI system learns about something the entire network shifts.  Teach it something else and the AI uses that same network for the new data.  There is no ‘reserved space” for the earlier data, so with each addition the AI  sees less of the original data and more of a conglomeration.  Sort of how using a dark marker in a notebook bleeds through to the next page.  Eventually is becomes a mess.

The Biological Advantage: Human Consolidation
Humans consolidate data.  We move new information to a temporary point in the brain and during sleep, that information is moved to a long-term storage location.  During that move the brain locks down information that it deems important so it cannot be overwritten.  This mechanism allows use to maintain old memories while making room for new ones,
Batch Processing vs. Sequential Wisdom
Further, AI systems learn in batch mode, essentially all at once.  In order to add to thet knowledge everything previously learned must be learned again along with the new data.  Humans are sequential learners. This is because the brain weights information by both importance and emotion, something AI tools cannot do as they have no emotional context.
​
Conclusion: The Hand, Not the Replacement
When we pull back the curtain on the "black box" of AI, we don’t find a digital god or a sentient rival; we find an incredibly sophisticated reflection of our own collective information. The very things that make AI feel "broken" to some—its fragile memory, its need for "tricks" like RAG to stay current, and its tendency to confidently misinterpret a dark room—are actually the boundary lines that define it as a tool.
AI doesn't have to be a competitor for the human throne to be revolutionary. Just as the telescope didn't replace the eye but allowed it to see further, and the engine didn't replace the leg but allowed it to move faster, AI is an extension of our ability to process a world that has become far too data-heavy for the naked mind. It is a "Cognitive Swiss Army Knife" that can pop open the right blade for almost any task, but it still lacks the calloused hands and common sense of the person holding it.
By stepping away from the "all or nothing" hype, we can appreciate AI for what it really is: a remarkable, albeit slightly temperamental, achievement of human engineering. It’s a better mousetrap, sure—but one that still needs a human to decide where to place it and what's worth catching. In the end, the most impressive thing about AI isn’t how the machine "thinks," but how we choose to use it to work, create, and build a better version of the world we already live in.