MediaTek (2454.TW) Zen Horizon Framework Deep-Dive Research

Hunting ten-year five-baggers among great growth stocks — pressing the upside question: "Can it get much bigger?"

• QHow large is its market ceiling? Is it expanding an existing pie, or creating an entirely new market?6/10

  A

  The ceiling is split between one high and one low: smartphone chips are competing for an existing pie that has already peaked and is still shrinking; the AI ASIC opportunity that truly determines the upside is a share of a large new pie, but MediaTek today is still only the third or fourth player at that table.

  Start with the core business that contributes nearly half of revenue. Smartphone chips are a textbook case of expanding an existing pie, and that pie itself is contracting: global smartphone annual shipments have fallen back from a peak of about 1.4 billion units and have long hovered around 1.1–1.2 billion units. Counterpoint's SoC share data shows MediaTek held about 34–36% in 2025, making it number one by shipments, but the share is expected to edge down to about 34% in 2026. The report body's “40–45%” is too high; on primary tracking data, 34–36% is closer. With share already leading and the market not growing, this business has no real structural “ceiling” to break through. It is a cash cow, not a growth engine. Smart Edge, including Wi-Fi/routers, smart TV SoCs, and the Chromebook platform Kompanio, is similar: it is taking share in multiple mature end markets.

  The real upside narrative rests entirely on the “new pie” of AI ASIC. Two corrections matter here. First, cloud custom ASICs are indeed a new market being created and expanding very quickly. Industry forecasts cited by TrendForce put 2027 TAM at about USD 70–80 billion, with MediaTek targeting 10–15%. Second, the report body describes “USD 2 billion” as a full-year 2026 target, which is inaccurate: both TrendForce and DigiTimes make clear this is a single-quarter run-rate for Q4 2026, with the annualized base larger, and it could scale to “several billion dollars” in 2027.

  So on Baillie Gifford's dividing line of “expanding an existing pie vs creating a new market,” MediaTek's answer is split: old business = expanding, or really defending, a pie that has already peaked, with the ceiling already within reach; new business = taking a share of a newly created, steeply growing pie, with an enticingly high ceiling. Yet MediaTek is neither the creator of this new market (Broadcom, through Google TPU and Meta, has already made ASIC 30%+ of revenue and defines the track), nor has it yet won any exclusive mega-order from a hyperscale cloud customer. It is positioning for the “dual-source” opening in Google TPU (v8t/v9, aiming for v10) and OpenAI's edge project. The direction is right and the imagination space is genuinely large, but whether it can turn a “high ceiling” into its own ceiling remains an unresolved wager today, not an established fact.

  Jun 10, 2026
• QCan its revenue at least double over the next five years? Will growth be driven mainly by volume, price, or new businesses?4/10

  A

  Doubling revenue over five years, from FY2025's NT$596.0 billion to roughly NT$1.2 trillion, requires about a 15% CAGR. Neither smartphone “volume” nor “price” can get it there. The only possible driver is the AI ASIC business, which today still contributes only single-digit revenue and lacks sufficient certainty.

  Set the baseline first. FY2025 revenue was NT$595,966M, up +12.3% YoY, but by Q1 2026 revenue had turned to NT$149.2B, down -2.7% YoY. In other words, as it entered 2026, company growth was effectively shifting down: the old engine was stalling, and the new engine had not yet taken over. A five-year double requires about a 15% annual compound rate, clearly above the report body's “FY2026 +8%” and management's full-year guide of mid-to-high single digits. The gap must be filled by new business.

  Break the three drivers down one by one:

  • Volume (smartphones) can hardly be counted on. Smartphone chips account for about 49% of revenue, but global handset shipments have long stayed around 1.1–1.2 billion units with no growth. MediaTek's share is already number one (about 34–36%) and is expected to edge down in 2026. Shipments offer no incremental room.

  • Price (smartphone ASP) can help, but only within limits. The flagship push of Dimensity 9400/9500 has indeed lifted ASP (the report body says by about 12–15%), but mid- and low-end ASPs are under pressure, and Q1 2026 gross margin had fallen to 46.3%, down -1.8pp YoY, with HBM/DRAM price increases adding erosion. Moving upmarket is a defensive “price offsets volume” move. It is not enough by itself to double the whole company.

  • New business (AI ASIC) is the only doubling variable. This is the core point of the entire analysis: TrendForce explicitly states that USD 2 billion is the Q4 2026 single-quarter run-rate, and says it “may scale to several billion dollars in 2027”; DigiTimes also supports the point that cloud customer demand for custom chips is pulling orders forward. If Q4 alone can run at USD 2 billion, then annualizing to USD 8–12 billion in 2027–2028, plus low-single-digit growth from the smartphone base, makes a five-year double mathematically possible.

  Conclusion: a doubling scenario can be written, but it is highly concentrated in a new business that today is only single-digit revenue and has not yet won any exclusive mega-order from a hyperscale cloud customer. This differs from the Baillie Gifford ideal of a doubling story where demand is clear and growth is naturally driven by multiple engines. MediaTek's doubling is a binary bet of “AI ASIC delivers or the company treads water,” not a high-probability linear extrapolation. If growth materializes, it will be driven mainly by “new business,” not volume or price. If AI ASIC disappoints, neither volume nor price can close the gap.

  Jun 10, 2026
• QFive years from now, what will take over as the next growth engine? Does this “second curve” exist today?5/10

  A

  The “second curve” does already exist today, and it is AI ASIC. But it is still in the embryonic stage of “orders visible, revenue not yet ramped”: a real but unvalidated curve, not a growth pole that has already found its footing.

  Baillie Gifford's question asks “who takes over five years from now.” For MediaTek, the answer is clear: smartphone chips are the main curve that must eventually be replaced, because they are mature, share has peaked, and incremental growth is limited. There is only one candidate to take over: cloud custom AI ASIC. Whether it “exists today” has to be viewed in two layers:

  As a business, it already exists and progress is accelerating. The Q1 2026 earnings call made AI ASIC a central narrative. Management doubled the 2026 target from USD 1 billion to USD 2 billion (according to TrendForce, this is the Q4 2026 single-quarter run-rate), said the first AI accelerator ASIC project is “progressing smoothly,” and said a second project is already in design, targeting mass production by the end of 2027. Customer clues are also more specific than the report body suggests: beyond NVIDIA GB10/NVLink Fusion, which the body emphasizes, TrendForce specifically says MediaTek is positioning for Google TPU dual-source share (v8t/v9, aiming for v10 alongside Broadcom), OpenAI's edge AI project, and an unnamed U.S. hyperscaler. Directionally, this is a real second curve with a long runway and leading-customer leads.

  But as “revenue that can take over,” it has not yet arrived. At this stage, AI ASIC contributes only single-digit revenue (the report body estimates about 9–10% in 2026), and the key “exclusive custom mega-order” has still not been announced. Broadcom already has Google/Meta, and Marvell already has AWS/Microsoft; MediaTek is the late challenger at this table. Under Baillie Gifford's standard, a second curve counts as “existing” only when there is not merely “a project,” but a visible and repeatable path for demand to convert into revenue. MediaTek has the former; the latter still needs proof.

  One risk also needs to be stated honestly: this second curve is highly concentrated in a handful of hyperscale buyers, and those buyers themselves use a “self-developed + multi-source” bargaining structure (Google also keeps Broadcom in the mix). Custom ASIC is project-based and replaceable; it does not have the entrenched share base of smartphones. So this is a “real but fragile” second curve. Its existence deserves a high score, but its durability and exclusivity still depend on Q2/Q3 2026 earnings and the first hyperscale order. Outside AI ASIC, the automotive, satellite communications, and Power IC opportunities mentioned in the report are too small to take over from smartphones within five years. They are only third-tier options.

  Jun 10, 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next three to five years?6/10

  A

  The core advantage is an integrated process-cost-scale capability as TSMC's largest non-U.S. customer, plus deep cellular/connectivity patent accumulation. But this moat is “moderately narrow,” and over the next three to five years it probably will not widen: it is squeezed in smartphones by Qualcomm and a shrinking market, must be built from zero in AI ASIC, and its foundation (Arm IP) is in someone else's hands.

  The report body scores the overall moat at 6/10. That judgment holds, and the components are also supported:

  • The real hard foundation is scale and cost (body: 7/10). As one of TSMC's largest non-U.S. customers for N3/N2, MediaTek gets priority and bargaining power for leading-edge process access. Annual shipments of 700–800 million SoCs spread FY2025 R&D of NT$148.3 billion, about 24.9% of revenue, across a very large base. This is the core base that others cannot replicate quickly.

  • Technology and patents are broad (7/10), but the foundation is not owned. It has more than 17,000 patents and accumulated know-how in Wi-Fi 7 / cellular / multi-core heterogeneous design, but the CPU core still depends on Arm licensing, a sharp contrast with Qualcomm's self-developed Oryon CPU. Arm's v9 architecture and price increases (listed as scenario C in the report, with license fees possibly rising 30–50%) are a medium- to long-term risk hanging over gross margin. It means part of the moat wall is rented.

  Why is the moat unlikely to widen over the next three to five years? Look at the two battlefields:

  Smartphones: the wall is being squeezed from both sides. MediaTek's share is already number one (about 34–36%), but Qualcomm is counterattacking in flagships with Snapdragon + Oryon and entering through PC chips; Samsung Exynos may also recover internal share once yields improve. Meanwhile, the overall handset market is not growing. With share near the top, the pie flat, and competitors opening an architectural gap through self-developed CPUs, the smartphone moat is more likely to narrow at the margin than widen. Switching costs (body: 5/10) are already medium: major OEMs such as Xiaomi/OPPO/vivo/Transsion generally dual-source from Qualcomm and MediaTek. No one is locked in.

  AI ASIC: the moat must be dug from zero, and the weakest link is ecosystem. Network effects are only 4/10: MediaTek does not have Qualcomm Snapdragon's developer NPU ecosystem, let alone the platform barrier of NVIDIA CUDA. In custom ASIC, a “project-based, customer-led, multi-sourceable” business, Broadcom has already made ASIC 30%+ of revenue through Google/Meta, with clearly deeper moat depth. Even if MediaTek wins dual-source share for Google TPU, that is “being allowed in to take a slice,” not building an exclusive barrier.

  Conclusion: this is a real but moderate moat, and the direction is unfavorable. Its cost-scale foundation is stable, but it lacks proprietary CPU IP, lacks a platform ecosystem, faces a peaked and pressured smartphone business, and must build behind giants in AI ASIC. Baillie Gifford prefers moats that widen and reinforce themselves, such as CUDA, TSMC's process lead, or network effects. MediaTek is the opposite: its advantages are defensive and erodible. Over the next three to five years, the moat is more likely to move sideways or shrink slightly than widen.

  Jun 10, 2026
• QIf its core business is disrupted, does it have the DNA to reinvent itself? How does it handle mistakes and bad news?6/10

  A

  Yes, and this is MediaTek's most underappreciated strength: historically, it has reinvented itself at least twice after its core business was disrupted or beaten down, proving that the company has genuine transformation DNA. That is precisely the basis for its confidence in betting on AI ASIC today.

  The implicit premise of Baillie Gifford's question is: once the core business is disrupted, can it rebuild a second version of itself from the wreckage? MediaTek's history gives a fairly positive answer, based on the evolution path laid out in the report body:

  • First reinvention (2007–2014 → the 2015–2019 trough → the 2020 counterattack): In its early years it rose through turnkey “shanzhai handset solutions,” then saw its market share suppressed by Qualcomm. From 2015–2019, flagship failures, gross margin falling to 35–40%, and a long stock decline left the core business almost disrupted. It did not lie flat. In the 5G transition window, it counterattacked with the Dimensity 700/800/1000 series and surpassed Qualcomm in quarterly smartphone chip shipments for the first time in Q3 2020 (IDC data). Going from a beaten-down low-end vendor to number one by shipments was a textbook recovery from the bottom.

  • Second reinvention (2023–2025 premiumization): It pushed Dimensity 9000/9200/9300/9400 into vivo X series, OPPO Find X, and Xiaomi flagships, lifted ASP by about 12–15%, and shed the label of being able to serve only the mid- to low-end market.

  • Third reinvention now underway (2026+ AI ASIC): When it foresaw the smartphone market peaking, it proactively redirected R&D firepower toward cloud custom chips. This itself is present-day evidence of “self-reinvention DNA,” not a cornered response after being forced into action.

  How does it handle mistakes and bad news? Two examples show a pragmatic attitude without whitewashing. First, management publicly acknowledged worsening smartphone conditions and YoY profit decline in Q1 2026 (TIFRS net profit -17.4% YoY, operating margin down from 20.0% to 15.8%), without using the AI story to hide pressure on the base business. Second, while sharply raising the AI ASIC target, executives instead reminded investors to “take a long-term view” and avoid short-term chasing. Cooling down their own stock when the price was at record highs and the market was euphoric is a notably restrained and honest signal, consistent with the management quality Baillie Gifford values: reporting bad news as well as good news.

  One qualification should be kept: both historical reinventions occurred inside the smartphone/connectivity circle of competence it knew best (5G and flagship SoCs were upgrades of the same craft). This AI ASIC move is a step into cloud, with a completely different customer structure and competitors (Broadcom, Marvell). The reinvention is harder, and past victories cannot be linearly extrapolated. Still, “willingness to face bad news + two documented successful pivots” puts it clearly ahead of most single-cycle players on this question. The DNA for self-reinvention is real; this time, the exam is simply harder.

  Jun 10, 2026
• QDoes management, especially the founder, have a long-term view and deep alignment with the company? Is it willing to sacrifice current profit for five to ten years out?5/10

  A

  MediaTek answers the long-term view and “sacrificing current profit for the future” parts very well. But “deep alignment with the company” is a clear weakness: founder-family ownership is very low and equity is highly dispersed. That is the core reason it cannot get a full score on this Baillie Gifford question.

  Start with the strengths. The evidence is solid:

  • Long-term view + willingness to sacrifice current profit for five to ten years out: this is a true MediaTek strength. The most direct evidence is that it continued investing FY2025 R&D of as much as NT$148.3 billion, about 24.9% of revenue, increasing the absolute amount from about NT$132.0 billion a year earlier, mainly into AI ASIC and next-generation Dimensity. The direct consequence was profit suppression: FY2025 revenue rose +12.3%, but net profit fell -1.0%, and EPS edged down from NT$66.92 to NT$66.16. Revenue growth and profit were completely decoupled. In other words, management knew it would sacrifice current EPS, yet still chose to fund the 2027–2028 AI compute opportunity. More importantly, when the stock hit record highs and the market was euphoric about the AI story, executives publicly urged investors to “take a long-term view” and avoid short-term trading. This restraint in cooling down their own stock at the peak is exactly the long-termist temperament Baillie Gifford prizes.

  • Founder continuity and strategic stewardship are strong. Tsai Ming-kai, age 77, led the 1997 spin-off from UMC to found MediaTek and is the kind of founder often described as a “father of chip design.” He still retains the chairman and strategic-steward role. CEO Rick Tsai, after years as COO, succeeded in sequence and has run day-to-day operations since 2024. The team is stable, succession has been orderly, and there has been no governance turbulence.

  Now the fatal weakness: alignment is too weak:

  • According to the report body, Tsai Ming-kai and his spouse (Tsai Ming-kai + Li Cui-xin) together hold only about 5.5%. The single largest shareholder is Singapore's GIC, with only 4.46%, and no single shareholder exceeds 5%. Equity is extremely dispersed.

  • Tsai Ming-kai's son Tsai Chi-hsuan has long occupied a major-shareholder seat but has not directly entered the operating management team.

  This means that by Baillie Gifford's standard of whether the founder has put his net worth and fate on the company, MediaTek is “professional-manager long-termism,” not a “founder-heavy fate-sharing vehicle.” Compared with high-scoring examples in this framework, such as a Schrodinger founder stepping back to an advisory role but retaining constraints, or Lunar-style companies with founders owning 30%+, MediaTek's founder economic alignment of 5.5% is visibly thin. It lacks the extreme skin in the game of “if the company sinks, I sink with it.”

  Overall judgment: vision (strong) + willingness to sacrifice current profit (strong, supported by real R&D spending and peak-time cooling comments) + governance continuity (strong), but ownership-level alignment (weak). This is a management team that is professional, restrained, and long-term oriented, but whose founder has not made a heavy ownership bet. Two of the three items are true strengths; one is a structural weakness. This question scores upper-middle, with the deduction coming from alignment rather than vision.

  Jun 10, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable and not dependent on harming society or regulation?6/10

  A

  Both answers are positive, but neither is extreme: customers “would miss it but would not be paralyzed,” because MediaTek is nearly indispensable in mid- to low-end Android and connectivity, yet Qualcomm is an alternative almost everywhere. Its growth model is also quite clean, not dependent on social harm or regulatory red lines, a rare “non-toxic” strength in its profile.

  Start with indispensability: the degree of “missing it” is medium-high. If MediaTek disappeared tomorrow, the biggest impact would fall on the global mid- to low-end and emerging-market Android phone ecosystem, including India, Africa, and Southeast Asia. It is the number one smartphone SoC supplier by shipments, with about 34–36% share in 2025, and volume models from Transsion, Xiaomi, OPPO, and vivo depend heavily on its high-value-for-money solutions. Wi-Fi/routers and smart TV SoCs also embed MediaTek chips at scale. Without it in the short term, hundreds of millions of affordable smart devices globally would face price increases and shortages.

  But the degree of “missing it” has a ceiling, because substitutes sit right next door:

  • In smartphones, Qualcomm is the second supplier for almost all major OEMs, and the industry generally dual-sources from MediaTek/Qualcomm. Customers are not in a “cannot live without it” position; they are in a “without it, products become more expensive and need redesign/tuning” position. Switching costs are medium, not locked-in.

  • In AI ASIC, indispensability is currently the lowest: custom ASIC is project-based, customer-led, and multi-sourceable, and Google already keeps Broadcom as a parallel supplier. On this line, MediaTek is “one option allowed to participate,” far from irreplaceable.

  Compared with Baillie Gifford's true indispensability benchmarks, such as TSMC for the whole industry or CUDA for AI developers, MediaTek is an “important, low-cost default option,” not an “irreplaceable choke point.”

  Now look at social and regulatory sustainability: this is a clean strength. MediaTek sells smartphone/connectivity/compute chips, a positive-externality business that makes inclusive computing and connectivity cheaper. It pushes advanced compute into affordable devices and expands, rather than harms, social welfare. It does not have the underlying nature of gambling, addiction, or data extraction, where monetization depends on harming others. Its growth model can withstand ethical and regulatory scrutiny. Regulatory risk is “exogenous geopolitics,” not “endogenous wrongdoing”: the risks listed in the report body are that U.S.-China tech conflict may affect future cooperation with U.S. customers (such as GB10 involving NVIDIA export exposure), and Taiwan Strait geopolitics. These are external uncertainties it passively bears, not red lines it crossed itself. It is also worth noting that Taiwan's export controls are relatively loose and the company's compliance base is stable.

  Overall judgment: indispensability is medium-high (nearly a default option in affordable Android and connectivity, but Qualcomm substitutes exist almost everywhere and AI ASIC is weakest) + growth model is highly sustainable, does not harm society, and regulatory risk is exogenous. MediaTek answers this question steadily. It is a company that customers would genuinely miss and that earns clean money, but the degree of “missing it” does not reach the level of “cannot survive without it.”

  Jun 10, 2026
• QHow are the unit economics of this business: gross margin and incremental returns? Do they improve or deteriorate as scale increases? Where does the money it earns go?6/10

  A

  Unit economics are “healthy but weakening at the margin”: gross margin near 47.5% is above the industry, and the net-cash profile is extremely strong, making this a good business that earns money. But incremental returns are falling: gross margin is declining, operating margin has fallen from 19% to 16% in two years, net profit has decoupled from revenue, and AI ASIC is likely a “low-margin scale trade.” As scale grows, unit economics are more likely to deteriorate than improve.

  The existing business has excellent unit economics. FY2025 gross margin was 47.5%, far above the semiconductor industry median of about 35% cited in the report body, driven by Dimensity premiumization and process-scale advantages. The balance sheet is also among the strongest in Taiwan design stocks: zero long-term debt and a long-held net cash position equivalent to about USD 10 billion+ (the report estimates about NT$350.0 billion). Annual shipments of 700–800 million SoCs spread heavy R&D across a large base, and inventory turnover is healthy (reduced to about 72 days in 2025, better than the industry 90–100 days). Looking only at whether “this business makes money and whether cash is solid,” the answer is yes.

  But incremental returns, which are what Baillie Gifford is really asking about, are worsening. This is the core warning sign:

  • Gross margin is sliding: 49.6% in FY2024 → 47.5% in FY2025 → 46.3% in Q1 2026, eroded by mid- to low-end ASP, second-generation N3 costs, and HBM/DRAM price increases.

  • Operating margin has declined for two straight years: 19.3% (FY24) → 17.4% (FY25) → only 15.8% in Q1 2026, versus 20.0% in the same period last year.

  • The most glaring decoupling: FY2025 revenue rose +12.3%, while net profit fell -1.0%. Revenue rose by more than 10%, yet profit declined. This shows that the latest increment did not translate into proportional profit, which is direct evidence that “larger scale has not improved unit economics and may have worsened them.”

  Will AI ASIC improve or worsen unit economics? Most likely worsen them, at least in the short term. Cloud custom ASIC is a “customer-led, thin-margin, high-volume” business. The ASIC gross-margin structures of Broadcom/Marvell are generally below the gross margin of MediaTek's existing smartphone chips. To win Google TPU dual-source share behind Broadcom, MediaTek has weak bargaining power, and early new business is about “trading profit for scale and an entry ticket”. The most likely scenario is that AI ASIC enlarges revenue but pulls the overall gross margin down further; growth comes at the cost of diluting unit economics.

  Where does the money go? Three destinations, with healthy proportions. ① The largest portion goes into R&D (FY2025 NT$148.3 billion, about 24.9% of revenue), betting on next-generation Dimensity + AI ASIC, which is reinvestment for growth. ② A high dividend payout returns cash to shareholders (2025 cash dividend of NT$53.50/share, payout ratio about 81%), more generous than TSMC's 50–60%, but this also shows that it sends a substantial portion of cash “back out” instead of putting all of it into high-return new opportunities, somewhat conservative for a pure growth stock. ③ It maintains a net cash buffer. Capital allocation discipline is clear and there is no reckless spending, but the combination of “high payout + declining incremental returns” looks more like the financial profile of a mature cash cow than the compounding machine Baillie Gifford favors, where unit economics improve as scale spreads costs.

  Jun 10, 2026
• QWhat conditions must all hold for it to rise fivefold in ten years? Are those conditions realistic? What expectations are embedded in today's stock price?3/10

  A

  A fivefold return in ten years requires three difficult things to happen at the same time: AI ASIC delivers, smartphones hold up, and valuation does not contract. The realistic probability is low. At today's price of about NT$4,475, the market has already priced in most of this optimistic script: the stock sits inside the report's optimistic intrinsic-value range, far above the top of the fair-value range, with almost no margin of safety. Future returns are more about “delivering on expectations” than “upward re-rating.”

  First quantify what “fivefold in ten years” requires. Current price is about NT$4,475, with market cap around USD 226–230B. Fivefold means reaching about USD 1.1 trillion+ within ten years, or about 17.5% annualized. To support that market cap, three conditions must all hold:

  1. AI ASIC must move from “project” to “large business”: the USD 2 billion Q4 2026 single-quarter run-rate must arrive on time, scale to several billion dollars annualized in 2027–2028, steadily win 10–15% share of a USD 70–80 billion TAM, and sign at least one repeatable mega-order from a hyperscale cloud customer (Google TPU dual-source entering v10, or OpenAI edge project conversion).

  2. The smartphone base must not collapse: against Qualcomm's Oryon counterattack, Samsung Exynos insourcing, and no growth in the overall market, it must hold about 34% share and avoid further major gross-margin erosion (already down from 49.6% to 46.3%).

  3. Valuation must not contract: today's TTM PE of about 65x and Forward PE of about 50x are already historical highs (the report cites a ten-year average of about 22x and a five-year average of about 38x). A fivefold return in ten years requires either earnings to rise fivefold while valuation stays high, or valuation to expand further. The former requires EPS to rise from about NT$66 to NT$300+, while the latter is even harder from an already expensive base.

  If any one of these three fails, the fivefold case breaks. They are also highly correlated: if smartphones collapse, AI cannot support the entire valuation. The report's Pre-mortem downside scenarios, with AI ASIC failure probability at 35%, smartphone share loss probability at 25%, and Arm price increase probability at 20%, are exactly the “failure paths” for these three conditions. A ten-year fivefold outcome is not impossible, but it requires a chain of low-probability events to hit sequentially. Its realism is low.

  What expectations are embedded in today's price? The market has already paid most of the down payment for the optimistic script. The report's three valuation ranges are: conservative NT$1,800–2,500, fair NT$2,500–3,400, and optimistic NT$3,500–4,800 (implying AI ASIC shares the market with Broadcom and grows 20%+ annually). The current price of about NT$4,475 is directly inside the optimistic range and about 30% above the NT$3,400 top of the fair range. Forward PE of about 50x is nearly 70% above the industry median cited in the report. In other words, the market has already assumed the best outcome of an “AI ASIC breakout.” Notably, sell-side consensus average target price of about NT$3,327–3,874 is actually below the current price. The highest Goldman NT$5,000 target was issued when the stock was only NT$2,610 in late April, implying about 92% upside at that time; since then, the stock itself has already captured most of that move and has approached the 52-week high of NT$4,970.

  Conclusion: conditions are demanding, probability is low, and price is stretched. At this level, the asymmetric odds that Baillie-style investors want, where upside greatly exceeds downside, have largely disappeared. The good script is priced in; the bad script, where the report's Pre-mortem gives -50% to -60% downside, is not. This is the fundamental reason the report rates it “Watch,” sets a reasonable buy range of NT$1,800–2,500, and requires a 38–56% pullback.

  Jun 10, 2026
• QWhy has the market not realized all this yet? Is it because the market does not understand, looks down on it, or cannot look far enough? What would become the “narrative inflection point”?3/10

  A

  The key reversal is this: for MediaTek, the market has not “failed to realize it.” Quite the opposite, the market has already fully, even excessively, recognized the AI ASIC story and priced it into a 65x PE. So Baillie Gifford's question, designed for “undervalued great companies” that the market does not understand, underestimates, or cannot see far enough, must be inverted here: is the market looking too far, believing too much, and treating an unfulfilled expectation as an accomplished fact?

  Lay out the facts first. The market has clearly “understood and looked far” already:

  • The stock ran within one year from a 52-week low of NT$1,130 to a June 2 high of NT$4,970, and the gain was once so overheated that the Taiwan Stock Exchange temporarily froze trading, while market cap exceeded USD 165.0 billion;

  • Goldman raised its target price from NT$2,454 all the way to NT$5,000, and aggressively revised 2028 EPS from NT$177 to NT$406, based precisely on AI ASIC; 22 of 24 sell-side firms rate it Buy.

  This shows the AI ASIC story is not undiscovered gold, but a consensus theme that has been repeatedly told, chased, and priced by the market. The report body's statements that “valuation has already overdrawn the possible 2027–2028 breakout of the AI ASIC business” and “at the lower end of the optimistic range, with no margin of safety” are consistent with this fact.

  What the market may truly have failed to see clearly is instead two risk-side points. This is the honest way to use Baillie Gifford's question in reverse:

  • The meaning of “USD 2 billion” has been optimistically misread: many momentum buyers treat it as a full-year 2026 target, but it is actually only the Q4 2026 single-quarter run-rate, and there has still been no announced exclusive mega-order from any hyperscale cloud customer (Broadcom already has Google/Meta, and Marvell already has AWS/Microsoft; MediaTek is still in the “positioning” stage for Google TPU dual-source and OpenAI edge projects). The market treating “target” as “order” is the biggest current cognitive risk.

  • The base business is deteriorating, but the story is covering it up: Q1 2026 revenue was -2.7% YoY, profit -17.4% YoY, and operating margin fell from 20.0% to 15.8%. Pressure on the smartphone base has been overshadowed by the AI narrative. It is telling that management itself publicly urged investors to “take a long-term view” and avoid short-term trading when the stock hit a record, effectively cooling the market's excessive optimism.

  What would be the “narrative inflection point”? There are clear triggers in both directions:

  • Upside confirmation, turning story into reality: in Q2/Q3 2026 earnings, AI ASIC reaches USD 500–800 million cumulative revenue for the half year, and the company announces an exclusive custom mega-order at the AWS/Google/Meta/Microsoft/ByteDance level. That would upgrade “expectation” into “delivered” and could open the next leg. Dimensity 9500, expected in Q3 2026, holding the high-end smartphone line and early TSMC N2 adoption would also help.

  • Downside falsification, collapsing valuation: the USD 2 billion Q4 run-rate misses (only reaching 1.0–1.5 billion), 2027 still has no hyperscale order, or Google TPU dual-source is swallowed by Broadcom alone. The market would then mark valuation back from 65x to a phone-chip vendor's 20–25x, matching the report's Pre-mortem -60% case.

  Conclusion: MediaTek's issue is not that “the market does not understand or cannot look far enough.” It is that “the market is looking too far and has prepaid for the best outcome.” The narrative inflection point is not on the “value discovery” side; it is on the “whether expectations can be validated by results” side. If they can be validated, high valuation can persist. If not, a Davis double-kill follows. For Baillie-style investors, buying now no longer earns money from an insight gap; it means taking the other side of an already-consensus optimistic expectation at a price with no margin of safety.

  Jun 10, 2026