Delta Electronics Zen Horizon Framework Deep Dive

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

  Conclusion first: the ceiling is high enough, and it has both dimensions, expanding an existing pie and opening a new market. But it is not the kind of unnoticed new market Baillie Gifford looks for. It is a large pie that has already been fully priced, with incremental growth released in two stages.

  Expanding the existing pie, the main case: Delta has more than 50% global share in AI server AC/DC power supplies and white-space liquid-cooling CDUs, yet its share of overall data center power infrastructure is only <1%, by Goldman Sachs' measure. This gap, white-space leader but machine-room-level bystander, defines the ceiling: the data center power market is about US$35.1 billion in 2025, rising to US$50.5 billion in 2030, with a 7.5% CAGR, dominated by Schneider Electric, Vertiv, ABB, and Eaton. Delta is using solid-state transformers and 800VDC In-Row systems to move from in-server power into the gray-space, machine-room-level distribution layer they control. That is the real incremental growth from expanding an existing pie.

  Opening a new market, the second layer: liquid cooling jumped from <1% in 2024 to about 9% in 2025, with shipments exceeding US$1.6 billion. Together with the new 800V HVDC architecture co-developed with NVIDIA, grid-to-chip cabinet-level power delivery, it rewrites data center power distribution at the architecture level and comes closer to creating a new market.

  Honest restraint: the largest increment, the 800V main architecture, does not ramp until 2027, and Delta itself says liquid-cooling growth is slowing. The near term is not a cliff, as the four major cloud vendors' 2026 capex is still rising by more than 60%, but a peak in the second derivative is a mid-to-late-cycle risk. This large pie and its extensions have already been priced in early. The ceiling is not the problem. The missing piece is margin of safety at the current price.

  Jun 4, 2026
• QCan its revenue at least double over the next five years? Will growth mainly be driven by volume, price, or new businesses?6/10

  A

  Conclusion first: yes, and consensus thinks it will not need five years. That is exactly where the risk lies. Delta's FY2025 revenue was NT$554.9 billion, +31.8%, so doubling requires roughly NT$1.11 trillion. Sell-side consensus already sees FY2027E revenue at about NT$1.0 trillion, the FactSet median from 23 firms. Another source gives FY2026E of about NT$731.6 billion, +31.9%, and FY2027E of about NT$1.058 trillion, up more than 30% again. In other words, revenue is nearly doubling within two years, well inside five years. A five-year doubling only requires about a 15% annualized rate, so the bar is not high.

  The drivers are almost entirely AI, and volume, price/mix, and new business all come from the same source: ①volume, AI server power supply ramp plus liquid-cooling shipments, with full-year 2025 shipments above US$1.6 billion; ②price/mix, as high-ASP AI power supplies are 5-10 times the price of traditional server power supplies, helping Q1 2026 gross margin reach a record 37%; ③new business, with 800V HVDC only entering its main ramp in 2027, while energy storage and gray-space distribution still need to be proven.

  But measured with Baillie Gifford's honest ruler, the quality of the doubling is diluted. This is not a smooth five-year compound curve. It is a one-off step pulled forward into 2026-2027 by the AI capex supercycle: the same consensus implies only about a 13% revenue CAGR over the next five years, which means growth falls sharply after 2027. That matches the second-derivative peak: capex growth at the four major cloud vendors is about 51% in 2026, 13% in 2027, and 5% in 2028. The roughly 70% of revenue that is non-AI, with automation modest and transportation/EV clearly weak, cannot carry the baton. So the answer is: revenue is likely to double, but it relies on a one-off AI step that the market has already fully priced, not a sustainable long-run compound slope. That supports Watch, not chasing higher.

  Jun 4, 2026
• QAfter five years, what will take over as the next growth engine? Does this second curve exist today?4/10

  A

  Conclusion: the second curve already exists and is ramping, but it is AI data center power plus liquid cooling itself. The issue is not whether there is another curve. It is that the next baton, 800V/gray space, only ramps after 2027, while the current curve's growth rate is already peaking. For a stock priced for high growth, that is the central tension.

  There are two layers to who takes over after five years. The first layer, already formed and ramping: AI server power supplies already accounted for >20% of revenue in FY2025 and grew about +60% YoY; liquid cooling was about 9%, from <1% in 2024, with shipments above US$1.6 billion; AI-related revenue was about 30% in total. This is a real high-ASP, high-margin curve, with traditional server power supplies at 5-10 times lower ASP, and Delta has more than 50% global share in AI server AC/DC power supplies and >50% in white-space liquid-cooling CDUs. But it is today's engine, not the one after five years.

  The second layer, the next baton, exists today but has not yet ramped, and that is the core of this question: ①the 800V HVDC cabinet-level new architecture, where Delta and NVIDIA jointly developed an 800VDC solution supporting 1.1MW cabinet-level power, aligned with NVIDIA's 2027 Kyber 1MW cabinet; ±400V enters mass production in 2026, 800V starts small-volume shipments in 2H 2026, and only becomes a major revenue contributor in 2027; ②machine-room-level gray-space power distribution expansion, where Delta's overall data center infrastructure share is only <1%, by Goldman Sachs' measure, leaving enormous space but with Delta as the attacker, having to take business from Schneider Electric, Eaton, and ABB; ③GPU on-board vertical power delivery, VPD, in partnership with Infineon, plus energy storage and microgrids.

  Honest restraint: this next baton lands exactly in the gap. The current AI curve's growth is peaking, with hyperscaler capex growth of about 51% in 2026, 13% in 2027, and 5% in 2028, while Delta itself says liquid cooling will grow more slowly in 2026 and overall AI growth will no longer accelerate. The main 800V/gray-space contribution has to wait until after 2027, and gray space is still in the early <1% stage. Non-AI EV and automation are clearly weak and unlikely to take over. So the second curve is real, quantifiable, and Delta is extremely well positioned, but it is highly tied to the current curve, both linked to AI capex. It is not a new engine independent of the AI cycle. That is the root of the tension between a high-quality business and an overdrawn price.

  Jun 4, 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next three to five years?6/10

  A

  Conclusion: the moat is real, and on the narrow AI front it is widening. But its width depends heavily on a single architecture dividend, while the edges are being eroded at the same time. This is a moat with a deepening core and leaking outer edges, not the kind of broad, error-tolerant moat Baillie Gifford prefers.

  The advantage is anchored in three quantifiable areas. First, leadership in power-conversion efficiency: the world's first 80 PLUS Titanium at >96%, the 2024 OCP flagship ORV3 at 97.5%, and an AI 18.5kW PSU at 98%. Data center tenders put heavy weight on PUE, and every percentage point adds design-in stickiness. Second, grid-to-chip full-chain vertical integration: from medium-voltage solid-state transformers and 800VDC distribution to GTC 2026 in-row power shelves with AC-DC efficiency of 98% and 140kW liquid-cooling CDUs designed for GB300 NVL72, Delta is one of the few suppliers able to provide end-to-end systems. Third, scale, R&D, and patents: R&D has long been >8% of revenue and was about 10% in 2024; the company has more than 18,000 granted patents and has been named a Clarivate Top 100 Global Innovator for five consecutive years. The moat is indeed widening in the Blackwell generation. Delta has more than 50% share in AI server power supplies and about 70% on the Blackwell platform, and its joint 800V development with NVIDIA is a real engineering lock-in.

  But the narrowing cracks also need to be acknowledged honestly: the share concentration is in white-space in-server power, while overall data center infrastructure share is only <1%, making the move into the gray space dominated by Schneider Electric and Eaton an attack, not a defense. Lite-On leads in BBUs and is expanding R&D to win orders. Delta does not control the liquid-cooling cold-plate layer, where AVC has more than 40% GB cold-plate share, and Eaton's acquisition of Boyd and Ecolab's acquisition of CoolIT are bringing giant capital into the field. Management says further gross-margin upside is limited, already a sign that pricing power is peaking. Baillie Gifford-style judgment: the core power moat remains wide and deepening over the next three to five years, but it is tied tightly to the NVIDIA ecosystem and the 800V dividend. If the architecture changes or major customers diversify suppliers, the moat will be more fragile than it looks on paper. That is the cost of the two-layer structure: diversified contracts, concentrated end demand.

  Jun 4, 2026
• QIf its core business is disrupted, does it have the DNA for self-reinvention? How does it handle mistakes and bad news?6/10

  A

  Conclusion: Delta has a rare industry DNA for self-reinvention. Over 50 years it has repeatedly jumped ship before the next technology wave displaced its main business, and management's candor about bad news is above the industry average. This is real support for its long-term rating, but it does not change the current judgment that the share price has already overdrawn the future.

  The best evidence for reinvention is its track record: in 1971, Bruce Cheng started with about NT$300,000 and 15 employees making TV coils; in 1983, it bet on switching power supplies and tied itself to IBM/Apple, becoming a global switching-power leader; in 1995, it entered industrial automation; in 2014, it acquired Norway's Eltek for about US$520 million to strengthen telecom power; and in 2021, it acquired Universal Instruments in the United States. From TV components to switching power, automation, and AI power plus liquid cooling, the company has consistently earned technology money from power-conversion efficiency rather than assembly outsourcing. That is why, in the most practical disruption, AI power delivery, it is a leader rather than a follower: at GTC 2026, it co-developed 800VDC with NVIDIA, launched 90kW power shelves, and jointly developed a 1.1MW cabinet-level solution.

  Its attitude toward mistakes and bad news is equally pragmatic: at the 2026-05 shareholder meeting, Ping Cheng acknowledged that EVs face high vehicle prices and charging infrastructure buildout below expectations, while saying the long-term direction has not changed. Management also admitted that gross-margin upside is limited, and oil prices and supply-chain issues may push up costs. It does not avoid weak spots or sell fantasies.

  Honest Baillie Gifford addendum: the reinvention DNA is a positive, but the current NT$2,425 price and PE-TTM of about 89× have already priced in that execution capability. Question 5 asks whether it can withstand disruption. The answer is yes. It does not ask whether one should buy now.

  Jun 4, 2026
• QDoes management, especially the founder, have a long-term view and deeply aligned interests with the company? Is it willing to sacrifice current profits for outcomes five to ten years out?6/10

  A

  Conclusion: on long-term orientation, Delta is close to a positive Baillie Gifford template. But on interest alignment, the score needs an honest discount. This is a second-generation founder-family succession with operating control, but family economic ownership appears relatively moderate, and the current share price is already overdrawn. Do not treat good governance as a buy reason.

  Long-term orientation, strong. This is a rare company that has written sacrificing the present for five to ten years out into its DNA. Founder Bruce Cheng made environmental protection, energy conservation, and love for the Earth its mission, and explicitly said companies should not only focus on near-term profits but should look further ahead. In numbers, the report notes that R&D has long been >8% of revenue, about 10% in 2024; granted patents exceed 18,000; and Delta has been named a Clarivate Top 100 Global Innovator for four consecutive years. It is Taiwan's first company to join RE100, committing to 100% renewable electricity in global operations by 2030, a tangible design-in advantage in energy-efficiency-heavy data center tenders. The latest evidence is the 2026-05 opening of a 7,500 m² R&D center in Soest, Germany, with capacity for 250 people and explicit coverage of EV 4-in-1 powertrain, showing long-cycle investment even in a recognized weak area.

  Interest alignment, requiring an honest discount. Founder Bruce Cheng stepped down in 2012 and is now honorary chairman; on 2024-05-30 the handover was completed, with eldest son Ping Cheng, CEO since 2012, becoming chairman and CEO, and second son Anson Cheng serving as director and vice president. This is a second-generation succession where the founder family still firmly controls operations. But economic alignment is relatively moderate: according to the report's cited source, foreign ownership is about 64% and director/supervisor ownership about 7%, from a single platform, with no two-source confirmation of precise family ownership. That is below many founder-controlled leaders. The family holds control, not a massive equity stake.

  Pragmatic restraint, a positive. At the 2026-05 shareholder meeting, Ping Cheng acknowledged that EVs face high short-term costs and charging infrastructure below expectations, and that gross-margin upside is limited. No overpromising, no avoidance of weak spots. Overall, Delta passes the management dimension and is even excellent, but it does not create a margin of safety for a current PE of about 89×.

  Jun 4, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable, without depending on harm to society or regulatory arbitrage?6/10

  A

  Conclusion: Delta's indispensability is close to full marks, but growth sustainability is the real issue in this question. It is running into the second-derivative turning point in AI capex.

  ① If it disappeared tomorrow, customers would be immediately blocked. Delta has more than 50% global share in AI server AC/DC power supplies, by broker estimates, is number one in telecom power, and has >50% share in white-space liquid-cooling CDUs. It is also deeply tied to NVIDIA: 90kW power shelves for the 800VDC architecture, 140kW liquid-cooling CDUs designed for GB300 NVL72, GPU on-board vertical power delivery with Infineon, and 18.5kW PSUs with efficiency up to 98%. Its grid-to-chip full-chain capability makes it difficult to replace at equal scale in the short term. A supply cut would directly choke AI compute power delivery and cooling. One caveat: the financial statements show no single customer above 10% of consolidated revenue, but end demand is highly concentrated in the NVIDIA ecosystem, meaning diversified contracts and concentrated end demand.

  ② The growth model does not harm society or regulation, but sustainability has real constraints. On the positive side, improving power-conversion efficiency directly enables energy savings and carbon reduction. Delta used 91.1% renewable electricity in 2025, with Americas operations reaching 100% ahead of schedule, consistent with its 2030 global RE100 target. There is no obvious social-harm attribute. But the biggest variable in growth is the second derivative of demand. Independent estimates show hyperscaler capex growth falling from about 51% in 2026 to about 13% in 2027 and about 5% in 2028, and Delta management also says 2026 AI growth will no longer accelerate, while liquid cooling will keep growing but at a slower pace. Add roughly 40-45% China revenue/capacity exposure facing decoupling and tariffs, plus Lite-On competing on price and lead time, and indispensability does not equal perpetual high-speed growth. Honest Baillie Gifford view: the moat is not in dispute, but this is a slowing windfall, not a ten-year fivefold compounding slope.

  Jun 4, 2026
• QWhat are the unit economics of this business, in gross margin and incremental returns? Does it get better or worse as scale increases? Where does the money it earns go?5/10

  A

  Conclusion: unit economics are genuinely improving, but this is driven by a high-ASP AI mix plus operating leverage, and it is tied to the peak of the AI capex cycle. It is not a cycle-independent monotonic rise. Delta earns technology money from power-conversion efficiency, not the hard-earned pennies of assembly outsourcing, and the moat is showing real evidence in gross margin.

  Gross margin and incremental returns: the five-year curve shows FY2022 revenue of NT$384.4 billion, gross margin 28.8%, operating margin 10.8% → FY2024 NT$421.1 billion, 32.4%, 11.3% → FY2025 NT$554.9 billion, +32%, 34.3%, 15.2%, net income NT$60.1 billion, +70.6%, EPS NT$23.14, with operating margin jumping nearly 4 percentage points in one year. Q1 2026 gross margin of 37% and operating margin of 17.8% both reached record highs, with net income +101%. The margin jump comes almost entirely from an upward product mix shift: AI server power ASP is 5-10 times that of traditional server power, and AI-related revenue is already about 30% of total revenue. The clearest evidence of incremental returns is segment operating leverage: FY2025 operating profit in the infrastructure and power segments rose +287% and +93% YoY, respectively, far outpacing revenue growth. So scale is indeed making the business better for now, but the essence is operating leverage driven by a rising share of high-margin AI mix.

  Where the money goes: R&D has long been >8% of revenue, about 10% in 2024, with granted patents above 18,000. Capex was NT$46.1 billion in FY2025 and will rise by >10% again in FY2026, used for factories in Thailand, India, and the United States to hedge geopolitics, while also increasing depreciation. Shareholder returns are thin, with a dividend yield of about 0.48% and PB of about 17.6×. This is reinvestment-first.

  Honest restraint: current margins are in the top zone of the AI capex cycle. Management itself says gross margin has limited further upside, while oil prices and supply-chain bottlenecks may push up costs. If AI capex growth moving from about 51% in 2026 to about 13% in 2027 materializes while depreciation has already stepped up, both returns and margins may come under pressure. Lite-On order competition and liquid-cooling price wars further weigh on pricing power. The improvement in unit economics is real, but it is tied to a cycle peak and has not yet been proven across the cycle.

  Jun 4, 2026
• QWhat conditions must hold for it to rise fivefold in ten years? Are those conditions realistic? What expectations are embedded in today's share price?2/10

  A

  Conclusion first: a fivefold rise over ten years, from about US$200B market cap to about US$1 trillion and from about NT$2,425 per share to about NT$12,125, requires four demanding conditions to hold at the same time. Today's share price already prices the most optimistic version of those conditions as fact. That is the core of the Watch rating. If the valuation multiple does not change, a fivefold market cap requires net income/EPS to rise about fivefold as well, from FY2025 EPS of NT$23.14 to about NT$116, or roughly 17.5% compound growth over ten years. But this is on top of a power-electronics shovel seller whose shares have already surged about 517% in 12 months, with market cap above US$200B and the stock ranking second in Taiwan. The difficulty is extremely high.

  The four required conditions, and their realism: ① AI server power plus liquid cooling must keep beating expectations, and Delta must substantially build out machine-room-level gray space, where its current share is only <1%, by Goldman Sachs' measure; ② AI capex must not drop sharply in 2027, but the second-derivative peak is already a multi-source consensus, with the four major cloud vendors' capex growth at about 51% in 2026, 13% in 2027, and 5% in 2028, and Delta itself saying 2026 AI growth will no longer accelerate and liquid cooling will grow more slowly; ③ the valuation multiple must not compress for ten years, yet the current PE of about 89× is nearly 4 times the historical midpoint of about 23×, and just a reversion to the midpoint would erase about three quarters of market value, so demanding no multiple movement for ten years is close to harsh; ④ China 40-45% exposure/geopolitics must not deteriorate, and Lite-On taking orders must not erode share and margins. If any one of the four loosens, the fivefold path breaks.

  What expectations does today's share price embed? It almost embeds full delivery of the bull case: the current NT$2,425 is already within the report's bull range of NT$2,200-3,030 and near the sell-side consensus target of NT$2,545, while independent DCF estimates, Morningstar at NT$588 and GuruFocus at about NT$584, are only 1/3-1/4 of the current price. In other words, the market has already priced most of the good outcomes for the next decade into the starting point. Baillie Gifford seeks a fivefold upside that the market has not yet recognized. Here it is the opposite: the share price has risen about 92% annualized, while EPS has risen only about 26% annualized, so most of the rerating came from multiple expansion rather than earnings. Honest judgment: Delta is a top-tier shovel seller and does not lack execution, but another fivefold rise from an already overdrawn starting point requires fivefold fundamentals plus zero valuation compression at the same time. That is not very realistic. The conditions for fivefold upside have mostly already been paid for by today's price.

  Jun 4, 2026
• QWhy has the market not recognized all this yet? Is it because investors do not understand it, look down on it, or cannot see far enough? What would become the narrative inflection point?3/10

  A

  Conclusion first: Delta is not a typical case of the market not recognizing something. It is precisely a case where the AI shovel-seller narrative has been over-recognized. Baillie Gifford LTGG looks for great growth businesses that can rise fivefold over ten years and that the market has not yet understood, then asks why nobody has seen it yet. Delta's shares have risen about 517% in one year, its market cap has exceeded US$200B, making it Taiwan's second-largest stock, its PE-TTM is about 89×, nearly 4 times the historical midpoint of about 23×, and the sell-side consensus target is about NT$2,545, with 18 buys and 1 sell among 19 firms, already near the current price. This is not failure to understand. It is over-clear understanding and over-full pricing.

  Break the narrative inflection point into three possibilities:

  • Not understood? Basically false.Sell-side analysts are uniformly bullish and have chased target prices up to the current price. The AI shovel-seller status, with global server power share above 50%, is already consensus common knowledge. There is no cognitive blind spot.
  • Looked down on? It is the reverse.The real bear case comes from independent DCF: Morningstar fair value is NT$588, the current price is about 156% above it, and GuruFocus is about NT$584, only 1/3-1/4 of the current price. The sell side is chasing, DCF is warning, and the directions are completely opposite. That split is itself the strongest risk signal.
  • Cannot see far enough? This is the only genuine upside option, not a market-failure-to-understand case.If 800V HVDC and cabinet-level liquid cooling become standard, Delta lifts its machine-room-level gray-space share sharply from <1%, and revenue and EPS growth digest a nearly 90× valuation, then today's price will not look expensive in hindsight. Delta has repeatedly caught the next wave, from PC power to communications, automation, and AI. Underestimating its execution is a real risk.

  So the real narrative inflection point is two-sided: upside equals AI power/liquid cooling continuing to beat guidance, with server power expected to grow about 70% in 2026, gross margin holding at 37%+, and 800V ramping in 2027 to digest the valuation through growth. Downside, which deserves more attention, equals a quarterly miss versus high expectations, capex growth at the four major cloud vendors falling sharply to about 13% in 2027, Lite-On order competition or liquid-cooling price wars pressuring gross margin, or geopolitics worsening. If any one materializes, nearly 90× TTM PE offers little cushion, and de-rating is the most likely path to permanent capital loss here. Honest conclusion: Delta's problem was never that the market failed to recognize greatness. It is who recognizes first that the narrative should contract after being overdrawn.

  Jun 4, 2026