Skip to content

Instantly share code, notes, and snippets.

@komsit37
Created April 22, 2026 05:44
Show Gist options
  • Select an option

  • Save komsit37/356fe14753825febca4becd1bc1f4b75 to your computer and use it in GitHub Desktop.

Select an option

Save komsit37/356fe14753825febca4becd1bc1f4b75 to your computer and use it in GitHub Desktop.
800 VDC for AI data centers — beneficiary brief

800 VDC for AI data centers — beneficiary brief

Whitepaper thesis: AI rack densities push past 1 MW; the only economic answer is 800 VDC bus distribution + integrated storage, with MV → DC conversion as the long-term endgame. This collapses the AC distribution layer and creates demand pulls in five places:

  1. High-frequency, high-voltage power devices — SiC MOSFETs / GaN HEMTs for the 800V → 12V (64:1) LLC converters. The whole architecture rides on SiC switching efficiency.
  2. MW-scale AC/DC rectifiers (5×3.5 MW per power block) and eventually MV rectifiers / solid-state transformers at 7.5 MW @ ≥98.5%.
  3. DC switchgear and solid-state breakers + 1500 A busducts.
  4. Layered energy storage — capacitors (≤100 ms), supercaps (100 ms-10 s), batteries (>10 s) — at rack and facility level.
  5. Grid-forming / VRT-capable interconnect + control plane that lets DCs become grid-supporting assets.

This brief ranks the 43-name shortlist by 800 VDC leverage, then for each Tier A/B name pulls segment-level revenue / growth / margin from the latest 9M FY2025 tanshin (tdnet fs CODE seg), so we can see what % of consolidated revenue actually rides the theme.


Tier ranking

Tier Code Name Why they win Why they don't
A 6963 ROHM Only JP pure-play SiC IDM (substrate→device); LLC converter sweet spot Discrete segment still loss-making; auto cycle drag
A 6504 Fuji Electric IGBT/SiC modules + AC/DC power supplies + EV chargers — three of the five demand-pulls in one company Semi segment in cyclical reset; auto-IGBT inventory drag
A 6503 Mitsubishi Electric Power semi + GIS / HVDC switchgear + SST roadmap — most complete stack Conglomerate dilution; HVAC dominates earnings
A 6890 Ferrotec SiC substrate via FLH (China); thermal stages + quartz for SiC fabs China geopolitical risk; substrate pricing volatile
A 6920 Lasertec SiC inspection — every new SiC fab buys these tools EUV mask cycle currently dominates revenue, masks the SiC opportunity
A 6882 Sansha Electric Pure-play industrial DC power supplies + EV chargers + semi process power — direct rectifier-block fit Tiny ¥18B revenue; semi seg still loss-making
A 6844 Shindengen Power devices + DC/DC converters; explicit "power-device new co" capex disclosures Auto-electrification cycle dependent; small fwd P/E hides cyclical earnings
A 6501 Hitachi (Hitachi Energy) HVDC, transformers, grid automation, MV rectifier roadmap — direct fit for facility-level + MV stages HE only ~20% of consolidated; segment data not in JP tanshin
B 6508 Meidensha Substation transformers + eco-switchgear + drives — bullseye for the MV side Mid-cycle margins; switchgear seg still loss-making
B 6622 Daihen RF generators (semi etch power) + transformers + welding power Small DC-data-center exposure; semi-cap cycle exposed
B 5333 NGK Insulators NAS battery (grid storage) + ceramics (insulators, SiC substrates) NAS in slow secular decline vs Li-ion; energy seg sub-scale
B 5310 Toyo Tanso Graphite for SiC epi growth — direct upstream consumable Single product line; SiC seg not separately broken out
B 6864 NF Holdings Lab/grid power supplies, frequency converters Tiny ¥10B mcap; segments not disclosed
B 6707 Sanken Power IC, motor-drive ICs (post-Allegro spin) Single semi segment; restructuring noise
B 485A Power X BESS (Mega Power) + ultra-fast EV chargers (Hypercharger) — 800 V architecture native Pre-profit scaling; consolidated loss -¥5B
C 6617 Takaoka Toko Transformers — grid capex tailwind Less DC-specific; smart-meter mix
C 6643 Togami Switchgear Smaller, less DC-specific
C 6507 Sinfonia Drives, motors Industrial — not directly 800VDC
C 6768 Tamura Coil transformers + power supplies Small-coil, not MW-scale
C 4966 Uyemura Plating chemistry Pass-through volume play
C 4062 Ibiden IC substrates Compute-side, not power
C 3445 RS Tech Reclaimed wafer + SiC adjacency Indirect
C 6928 Enomoto Lead frames Volume play
C 6140 Asahi Diamond SiC dicing wheels Consumable
D 6266 / 6291 / 6337 / 6590 / 6728 Semi cap eq Generic semi capex beneficiary Not 800VDC-specific
D 6742 Kyosan Rail signaling Wrong vertical
D 9501 / 9502 TEPCO / Chubu Demand-side beneficiary Slow regulated growth
D 9517 eRex Power retail No HW exposure
D 1942 Kandenko DC build-out contractor Indirect; capex deflator
D 1436 / 4169 / 3083 / 3121 / 3260 / 3913 Pivots Narrative play Pre-revenue speculation
D 6981 Murata Capacitors (relevant!) Capacitor exposure not driving consolidated; mainly mobile/MLCC

Tier A — segment-level analysis

All tables are 9M FY2025 (Apr-Dec 2025) vs 9M FY2024, ¥M unless noted, from each company's latest 3Q tanshin (segments/<code>.seg).

6963 ROHM — SiC IDM, the LLC-converter pure play

Segment Rev 9M-FY24 Rev 9M-FY25 YoY Profit 9M-FY24 Profit 9M-FY25 Margin 9M-FY25
LSI 155,888 169,360 +8.6% 3,986 19,937 11.8%
半導体素子 (Discrete — SiC/IGBT/MOSFET) 144,128 155,837 +8.1% −20,493 −16,488 −10.6%
モジュール 25,664 24,915 -2.9% 2,710 3,011 12.1%
その他 (resistors) 18,960 19,401 +2.3% 1,890 3,332 17.2%
Consol 344,642 369,515 +7.2% −11,080 +9,703 2.6%
  • Target segment: 半導体素子 (Discrete) — houses the SiC MOSFET line. 42% of consolidated revenue.
  • Read: Discrete is still loss-making (-10.6% margin) but the loss narrowed from −¥20.5B to −¥16.5B on +8% revenue. The SiC capex amortization is the drag; the question is when SiC volume ramps faster than D&A. The whole equity story is whether Discrete returns to positive margin within FY26-27. 800 VDC architecture pull is a fresh demand vector that wasn't in the auto-only thesis.
  • Caveat: ROHM also changed depreciation method (定率→定額) which boosted Discrete profit ¥5.4B — half of the apparent improvement is accounting.

6504 Fuji Electric — three demand-pulls under one roof

Segment Rev 9M-FY24 Rev 9M-FY25 YoY Profit 9M-FY24 Profit 9M-FY25 Margin 9M-FY25
エネルギー (T&D, EV chargers, BESS) 229,553 255,690 +11.4% 20,269 33,085 12.9%
インダストリー (Drives, automation) 286,756 319,128 +11.3% 17,126 19,749 6.2%
半導体 (IGBT/SiC modules) 164,028 170,552 +4.0% 21,543 14,987 8.8%
食品流通 (Vending) 83,991 77,936 -7.2% 12,274 9,319 12.0%
その他 26,733 27,746 +3.8% 2,427 2,539 9.2%
Consol 791,064 851,055 +7.6% 68,431 74,027 8.7%
  • Target segments: エネルギー + 半導体 = ¥426B = 50% of consolidated revenue.
  • Read: エネルギー is the standout — +11% revenue, +63% segment profit YoY, margin from 8.8% to 12.9%. This is where DC PSU / EV charger / grid storage demand lands. 半導体 is in a cyclical reset (margin 13.1% → 8.8%) on auto IGBT inventory; the SiC module ramp is the next-leg story.
  • Caveat: Energy seg includes traditional T&D and food-vending HVAC — not a pure DC play. Need MD&A to break out DC PSU.

6503 Mitsubishi Electric — most complete vertical stack

Segment Rev 9M-FY24 Rev 9M-FY25 YoY OP 9M-FY24 OP 9M-FY25 Margin 9M-FY25
インフラ (T&D, GIS, HVDC, defense) 789,048 926,454 +17.4% 47,662 79,281 8.6%
インダストリー・モビリティ (FA, EV) 1,205,804 1,202,091 -0.3% 64,740 94,157 7.8%
ライフ (HVAC) 1,625,061 1,670,294 +2.8% 139,649 136,034 8.1%
デジタル 57,182 57,702 +0.9% 6,915 7,293 12.6%
セミコンダクター・デバイス (SiC, IGBT, opto) 194,680 188,735 -3.1% 31,884 35,120 18.6%
その他 128,576 110,734 -13.9% 43,100 41,640 37.6%
Consol 4,000,351 4,156,010 +3.9% 303,555 294,757 7.1%
  • Target segments: インフラ + セミコンダクター = ¥1,115B = 27% of consolidated revenue.
  • Read: インフラ is the breakout — +17% revenue, +66% profit YoY. Hitachi Energy's success is bringing demand to MELCO too on transformers and GIS for DC connections. Semi is a small but high-margin slice (18.6%) where SiC mix is driving the resilience despite revenue dip. Combined target seg margin ~10%.
  • Caveat: HVAC (ライフ) is still ~40% of revenue and ~46% of profit; for the 800 VDC theme to move the consolidated needle, インフラ needs to keep compounding at this rate.

6890 Ferrotec — SiC substrate via FLH

Segment Rev 9M-FY24 Rev 9M-FY25 YoY Profit 9M-FY24 Profit 9M-FY25 Margin 9M-FY25
半導体等装置関連 (Semi eq + SiC substrate FLH) 122,784 134,314 +9.4% 10,838 12,487 9.3%
電子デバイス (Power-semi substrates, thermal) 36,303 41,963 +15.6% 5,923 8,837 21.1%
車載関連 23,127 22,477 -2.8% 2,837 2,008 8.9%
その他 (saw blade, solar Si) 20,150 12,890 -36.0% 624 -169 neg
Consol 202,365 211,645 +4.6% 19,696 21,840 10.3%
  • Target segments: Semi-related + Electronic Devices = ¥176B = 83% of consolidated revenue.
  • Read: Electronic Devices (where the FLH SiC substrate spin-off sits) grew +16% with margin expansion 16.3% → 21.1%. Semi-related grew +9% with margin steady at 9.3%. This is the cleanest pure-play exposure on the JP shortlist by % of revenue.
  • Caveat: China-domiciled subsidiary risk; SiC wafer pricing has been weak globally — the volume is the story, not ASP.

6920 Lasertec — SiC inspection (single-segment)

  • Single-segment (検査・測定装置). 9M-FY26 (Jul-Dec 2025) revenue ¥51.8B (-19.7% YoY), op profit ¥18.0B (-22%) per out/6920.md. The decline reflects EUV mask inspection cycle.
  • Read: SiC inspection is a small but explicitly-disclosed product line (typically <10% of revenue from the IR deck, not separately segmented). The 800 VDC pull is a multi-year option, but Lasertec's near-term P&L is dominated by the EUV cycle.

6882 Sansha Electric — pure-play DC power supplies

Segment Rev 9M-FY24 Rev 9M-FY25 YoY Profit 9M-FY24 Profit 9M-FY25 Margin 9M-FY25
半導体事業 (Semi process power) 4,367 4,582 +4.9% -419 -511 -11.2%
電源機器事業 (Industrial power supplies, EV chargers) 13,568 13,597 +0.2% 1,037 858 6.3%
Consol 17,936 18,180 +1.4% 618 347 1.9%
  • Target segment: 電源機器 = 75% of revenue.
  • Read: Pure-play but tiny (¥18B) — at this scale, a single 800VDC AI-DC contract win is transformative. Margin compressed YoY (7.6% → 6.3%); semi seg still loss-making. Highest revenue-leverage on a beneficiary basis but smallest absolute exposure.

6844 Shindengen — power devices + EV DC/DC

Segment Rev 9M-FY24 Rev 9M-FY25 YoY Profit 9M-FY24 Profit 9M-FY25 Margin 9M-FY25
パワーデバイス事業 23,271 24,525 +5.4% -348 1,969 8.0%
パワーユニット事業 (DC/DC, motorcycle/EV) 49,911 53,878 +7.9% 3,854 2,981 5.5%
パワーシステム事業 4,310 5,304 +23.1% 808 969 18.3%
Consol 77,611 83,807 +8.0% 758 3,117 3.7%
  • Target segments: All three = 100% of revenue — Shindengen is fully in scope.
  • Read: Power Device segment swung from loss to +¥2B profit (margin ~8%). Recent filings explicitly disclose new "power-device new co" stock acquisition (Jan 2026) — tells you management is doubling down. Power System is small but margin 18% — could be the AI-DC PSU sub-line.
  • Caveat: EV motorcycle exposure (Honda) is the dominant Power Unit driver — not DC bus.

6501 Hitachi — Hitachi Energy is the prize, but it's ~20% of consolidated

  • Tanshin doesn't disclose segments at this granularity (Hitachi reports under their own IFRS structure). From the latest IR deck:
    • Hitachi Energy: revenue ~$13B FY24, expected ~$15-17B FY25, EBITA margin expanding 11% → 14%, 5-year orders backlog, capex tripling to expand transformer / HVDC capacity.
    • Hitachi consolidated: ¥9.7T revenue → HE ≈ 20% of group, but ~30% of operating profit growth in FY24-25.
  • Read: The single largest absolute-yen beneficiary on the list. 800 VDC architecture demands MV transformers + grid-forming inverters + grid automation — exactly Hitachi Energy's catalog. Backlog is the constraint, not demand. But you only get 20-cent dollars of HE through the parent.

Tier B — segment-level analysis

6508 Meidensha — substation eq + drives

Segment Rev 9M-FY24 Rev 9M-FY25 YoY Profit 9M-FY25 Margin
電力インフラ事業 58,336 69,068 +18.4% 7,694 11.1%
社会システム事業 52,591 55,566 +5.7% -1,707 -3.1%
産業電子モビリティ 50,454 46,492 -7.9% -1,037 -2.2%
フィールドエンジニアリング 24,837 27,673 +11.4% 3,966 14.3%
不動産 2,397 2,396 flat 1,059 44.2%
Consol 191,083 203,855 +6.7% 7,082 3.5%
  • Target segment: 電力インフラ = 34% of revenue.
  • Read: Power infrastructure had the standout — +18% revenue, ~50% profit growth, margin 11.1%. Direct beneficiary of US/JP transformer shortage. The other segments (社会システム, 産業電子モビリティ) are loss-making and dragging consolidated margin to 3.5%. A clean pure-play in this one segment if you back it out.

6622 Daihen — Energy Mgmt is the relevant slice

Segment Rev 9M-FY24 Rev 9M-FY25 YoY Profit 9M-FY25 Margin
エネルギーマネジメント (Transformers + power systems) 83,810 86,659 +3.4% 9,634 11.1%
ファクトリーオートメーション (welding) 21,441 22,534 +5.1% 738 3.3%
マテリアルプロセシング (RF generators for semi etch) 50,429 54,080 +7.2% 5,461 10.1%
Consol 155,825 163,423 +4.9% 12,478 7.6%
  • Target segments: Energy Mgmt + Material Processing = 86% of revenue.
  • Read: Energy Mgmt is the transformer/HV side (+3% rev, 11% margin). Material Processing is RF generators for semi etch — semi-cap cycle beneficiary, +7% rev, 10% margin. Both are in scope but for different reasons (HV grid + SiC fab capex). Solid mid-cycle margins; less explosive growth than Meidensha or Hitachi Energy.

5333 NGK Insulators — energy is the smallest of three

Segment Rev 9M-FY24 Rev 9M-FY25 YoY Profit 9M-FY25 Margin
エンバイロメント (auto-cat substrate) 288,567 293,110 +1.6% 54,108 18.5%
デジタルソサエティ (semi/electronics ceramics) 123,937 149,132 +20.3% 20,095 13.5%
エネルギー&インダストリー (NAS battery + insulators) 42,946 45,666 +6.3% -850 neg
Consol 455,451 487,908 +7.1% 73,048 15.0%
  • Target segment: E&I = 9% of revenue, loss-making. Semi/digital is the actual growth engine (+20%, 13.5% margin) but that's AI-DC compute side, not 800VDC power.
  • Read: NGK is mis-tagged for this theme. NAS battery is sub-scale and loss-making; the real story is the auto-catalyst substrate franchise + semi ceramics. Don't buy NGK on the 800 VDC thesis.

5310 Toyo Tanso — single product, no SiC breakout

  • Reports by geography (Japan / US / Europe / Asia), not by product. SiC graphite is embedded in Japan + Asia revenue but not separately disclosed.
  • 9M-FY25 (Jan-Sep 2025) consolidated revenue ¥46.2B (vs ¥53.1B prior 9M, -13.0%), op profit ¥6.8B. Margin holding ~14%.
  • Read: Generic graphite cycle dominates near-term P&L. SiC is the option value. Without segment disclosure, hard to size — depends on management commentary in the IR deck.

6864 NF Holdings — segments not disclosed in tanshin

  • Single-segment reporter; no useful breakdown. ¥10B mcap, P/B 0.8, profit small. Frequency converters and lab power supplies — too small to be a serious 800 VDC vendor at scale, but a watchlist name on takeover/concentration optionality.

6707 Sanken Electric — single semi segment

  • Single-segment (半導体デバイス事業). Post-Allegro spin, the consolidated business is power IC + motor-drive ICs + LEDs. 9M-FY25 revenue declined; management noted auto inventory adjustment. Margin recovery is the swing factor, not the 800 VDC theme directly.

485A Power X — pre-revenue scaling

Segment (FY24 = full year) Revenue Seg Profit Margin
BESS (Mega Power, Cube) 4,143 855 20.6%
EVCS (Hypercharger) 1,628 -498 neg
電力 (energy retail) 389 -55 neg
Consol (after corp ¥5.2B) 6,161 -4,942 neg
  • Read: BESS unit economics already work (+21% margin). EV charging is the burn item. Consolidated still loss-making by ¥5B/yr on corporate overhead. The cleanest "800 VDC pure-play architecture" name on the screen — Hypercharger uses 800V DC bus directly. Risk: small float, IPO-vintage scaling, no AI-DC reference customer disclosed yet (IIJ协业 announced 2026-02).

Summary table — exposure × growth × margin × contribution

For Tier A/B, the "target segment" rolled up:

Code Name Target seg(s) % of rev Seg YoY Seg margin Conviction
6963 ROHM Discrete (SiC/IGBT) 42% +8% −11% High option value, low quality today
6504 Fuji Electric Energy + Semi 50% +9% 11% High — Energy seg is actually inflecting
6503 Mitsubishi Electric Infra + Semi-Device 27% +13% 10% High — Infra +17% / +66% profit
6890 Ferrotec Semi-eq + EDevice 83% +11% 12% High — most concentrated exposure
6920 Lasertec (single seg) 100% -20% 35% Mid — EUV cycle masks SiC pull
6882 Sansha Power Equipment 75% flat 6% Mid — small, watchlist for contract wins
6844 Shindengen All three 100% +8% 4% Mid — auto-electrification dependent
6501 Hitachi Hitachi Energy ~20% ~+20% 14% High — backlog is the constraint
6508 Meidensha 電力インフラ 34% +18% 11% High — purest mid-cap HV play
6622 Daihen Energy + Material 86% +5% 10% Mid — solid, less explosive
5333 NGK E&I 9% +6% neg Low — mis-tagged for this theme
5310 Toyo Tanso (no breakout) n/d -13% (consol) 14% Low-mid — option only
6864 NF Holdings (single) 100% small mid Low — too small
6707 Sanken (single semi) 100% declining mid Low — restructuring noise
485A Power X BESS+EVCS 100% growing seg+, consol- Speculative — pure architecture native

Where the highest-quality leverage sits

  1. Hitachi Energy (via 6501) — best-in-class margin trajectory in the right business, but you only get 20-cent dollars of it through the parent. Best name if you want quality.
  2. Meidensha (6508) — only mid-cap with a single segment delivering +18% rev / 11% margin in the exact right business. Best pure-play.
  3. Mitsubishi Electric Infra (6503) — +17% rev, +66% profit, 8.6% margin and accelerating. Best risk-adjusted because of conglomerate underwriting.
  4. Fuji Electric Energy (6504) — +11% rev, +63% profit YoY. Cheapest of the big 3 (fwd P/E 19.9 in index.md).
  5. Ferrotec (6890) — purest by % of revenue exposed, but China FLH execution is the swing factor.

Where to be careful

  • NGK (5333) — tagged by the screen but the relevant segment (E&I, NAS battery) is loss-making and only 9% of revenue. Don't buy this on the theme.
  • ROHM (6963) — best long-dated SiC option but Discrete segment is still −11% margin and depreciation accounting boosted reported numbers ¥5.4B. Need to see real margin expansion in next 2-3 quarters.
  • Lasertec (6920) — SiC tools are real but lost in EUV-cycle noise; near-term P&L will mislead the thesis.
  • Pivots (Tier D) — speculative; not playable on fundamentals.

Cross-references

  • Source data: segments/<code>.seg
  • Per-ticker profile: out/<code>.md
  • Methodology: notes.md
  • Supply-chain map: supply-chain-brief.md
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment