The Japanese Fiber Majors: One AI-Optical Theme, Three Very Different VehiclesComparative deep-dive — individual names vs an equal-weighted basket

The view, up front

Japan’s three listed “electric-wire” majors — Furukawa, Sumitomo Electric, and Fujikura — have become the cleanest public proxy for a single secular driver: the build-out of optical interconnect inside AI datacenters. All three are century-old cable conglomerates; all three multi-bagged over the past year on the same AI-optical story; and all three sold off near-identically in the May–June 2026 “Japan cable-maker rout.” That co-movement is the central fact. This is one theme expressed through three vehicles, not three independent bets.

The vehicles differ enormously in three dimensions that determine what you actually own:

• Purity. Fujikura earns ~81% of its group operating profit from its AI-exposed telecom segment; Furukawa ~42%; Sumitomo ~18% ¹²³. Buying Sumitomo is mostly buying automotive wiring harnesses with an AI call option stapled on.
• Valuation. Fujikura trades at ~34x EV/EBITDA, Furukawa ~32x, Sumitomo ~16x on the model’s consistent EBITDA basis. You pay roughly twice the multiple for purity.
• Balance-sheet and operating gearing. Fujikura is net cash with a 16% operating margin; Sumitomo carries modest leverage at an 8% margin; Furukawa is the most financially geared (net debt/EBITDA ~2.3x) and the highest-beta (1.17) of the three ⁴.

The recommendation: for an investor who wants exposure to AI-datacenter optical interconnect but cannot confidently pick the winner, an equal-weighted basket of all three is the disciplined default. It keeps full thematic beta while diversifying the idiosyncratic blow-up risk that just played out — Fujikura’s below-consensus guidance, Furukawa’s leverage, Sumitomo’s auto/tariff drag. What the basket does not do is diversify the theme: a multiple reset hits all three within a few points of each other (the model’s sensitivity table puts a 30% de-rating at –30% to –32% for every name). Because the names sit at 3–7x off their 52-week lows and the May rout showed the re-rating is fragile, this is a high-beta thematic satellite to size as a satellite — not a core holding — and entry matters. For investors with a firmer view: concentrate in Fujikura for maximum quality and purity (and accept peak-multiple, peak-cycle risk), or own Sumitomo as the cheap, defensive way to keep a toe in. Furukawa is the one to be most careful with — it trades nearly as expensive as Fujikura on roughly half the AI purity, with the most leverage and the most beta.

1. The setup: who these companies are

Furukawa, Sumitomo Electric, and Fujikura are Japan’s “electric-wire majors” (the densen gosanke, literally the wire industry’s “big three”) — a recognized market cohort whose tickers even run in sequence (5801/5802/5803) ⁵. Each was founded around copper wire and cable more than a century ago, and each has since diversified into a different mix of automotive, energy, electronics, and telecom. What unites them now is optical fiber: all three are top-tier global makers of the glass strand and cable that carry the world’s data, and all three sit in the global top six of fiber-optic cable revenue (Prysmian #1, Corning #2, Sumitomo #3, Furukawa #4, CommScope #5, Fujikura #6) ⁶.

The base optical-fiber business is, by itself, a slow grower: the global market for pure fiber strand was roughly US$12.7B in the base year and is forecast to reach ~US$18.7B by 2031, a ~5.8% CAGR ⁷. That is not the story driving these stocks. The story is the AI-datacenter overlay sitting on top of it.

Why now. The same physical product — high-count optical fiber, connectors, and transceivers — is being pulled into a far faster-growing end market (AI datacenters) at the exact moment fiber is in short supply. That demand mix-shift, not the underlying telecom market, is what re-rated the cohort.

2. The demand driver: AI-datacenter optical, not telecom fiber

The thing being bought is the optical plumbing of the AI datacenter. As GPU clusters scale, the interconnect between chips, servers, and buildings shifts from copper to optics, and the fiber counts per cable, the connector density, and the transceiver speeds all step up together.

The size of the prize. The AI-attributable slice of fiber and optics is growing several times faster than the base market. One bottom-up digest puts the AI-datacenter-attributable fiber-plus-optics category at ~US$25.8B in 2026 rising to ~US$63.5B by 2029 — roughly a 35% CAGR ⁸. Independent cuts corroborate the order of magnitude: Cignal AI counted ~US$25B of optical-component revenue in 2025 ⁹, and the narrower data-center-interconnect (DCI) sub-segment is projected to grow from ~US$15.4B (2025) to ~US$25.9B (2030) ¹⁰. Methodologically these are different lenses — a bottom-up component count versus a defined sub-segment — and they converge on a mid-$20Bs 2026 demand pool, which is what gives the demand number credibility rather than a single vendor’s headline.

The technology that matters. Three product vectors carry the AI-optical demand, and the three Japanese majors are positioned differently on each:

• Ultra-high-density fiber cable — packing thousands of fibers into a thin cable to wire dense GPU campuses. Fujikura’s spider-web-ribbon (WTC) cable reaches 6,912 fibers in a 29mm diameter ¹¹; Furukawa mass-produces a 13,824-count cable, among the world’s highest density ¹². This is the most mature, highest-volume AI-optical product, and Fujikura leads it ¹³.
• Co-packaged optics (CPO) — moving the optical engine onto the same package as the switch/GPU silicon to cut power. Here Sumitomo is differentiated: it is named a technology partner in NVIDIA’s CPO ecosystem on the fiber/connector and indium-phosphide-laser tier ¹⁴⁸, and Furukawa is pursuing CPO joint development with Shinko Electric ¹⁵.
• Next-generation fiber and faster transceivers — multi-core fiber (MCF) and hollow-core fiber (HCF) for the longest reaches, plus the step-up in transceiver speeds from 200/400G toward 800G and 1.6T ¹³¹³. LightCounting expects AI-cluster Ethernet optics to scale into a multi-tens-of-billions market this decade ¹⁶.

A required reality check. Bernstein’s framing is the discipline against over-extrapolation: inside the rack (“scale-up,” GPU-to-GPU), copper remains dominant and will for some time; CPO begins at limited scale in scale-out networks around H2 2026, with scale-up adoption not until ~2028+ ¹⁷¹⁷. And CPO’s bigger effect is to reallocate the optical profit pool toward engines, fiber-array units, and lasers rather than to expand the total dramatically ¹⁷. The bull case is real but it is a multi-year ramp with a copper-shaped ceiling, not a step-change.

The near-term tailwind that is not speculative: fiber is in active shortage, and the supply tightness is already producing 2026 price increases ¹³. That is margin-positive for all three today, ahead of any CPO transition.

3. The purity spectrum: what you are actually buying

The single most important distinction across the three is how concentrated each is in the AI-optical story. The cleanest disclosed gauge is the share of group operating profit earned by each company’s AI/datacenter-exposed segment.

• Fujikura — ~81% AI-optical. Its Telecommunication Systems (Information & Communication) segment generated ¥152.7B of operating profit, ~81% of group OP, and grew ~1.7x year-on-year ¹¹¹⁸. Fujikura is, in profit terms, an AI-optical pure-play wearing a cable-conglomerate’s name.
• Furukawa — ~42% AI-optical. From FY2026 Furukawa reorganized into six market-based segments explicitly to “accelerate growth in the data center domain” ². Its two datacenter-centric segments — Optical Solutions (¥196.9B sales) and Digital Infrastructure Components (¥198.0B) — together are ~30% of sales and ~¥26.6B, or ~42%, of group OP, guided to ~70% of OP in FY2026 ²²²². Real and rising, but still less than half the profit base.
• Sumitomo — ~18% AI-optical, and even that overstates it. Its Infocommunications segment (all optical) is just 6.4% of sales at ¥326.6B and ¥77.4B of OP, ~18% of group profit ³³³. Within that, the genuinely AI-datacenter piece is estimated at only ~US$0.8B — roughly 2–3% of group sales ¹⁹¹⁹²⁰. Sumitomo’s profit is dominated by its Automotive segment (wiring harnesses, 57.5% of sales) ³³.

A useful grounding cross-check sizes each company’s 2026 AI-datacenter fiber revenue at similar absolute scale — Fujikura’s US subsidiary AFL ~US$0.8B, Sumitomo ~US$0.6B, Furukawa ~US$0.4B ⁸⁸⁸ — which tells you the purity difference is driven as much by how small the rest of Fujikura is as by how large its AI business is. That is the right way to read the spectrum: you are choosing how much non-AI ballast to accept alongside the same theme.

4. The numbers: comparative financials and valuation

All three printed record results for the year ended March 2026. The table below is the model’s comps output; every figure ties to a cell in japanese-fiber-majors-comps_v1_2026-06-08.xlsx.

USD conversions at USD/JPY 160.13 ³¹. EV/EBITDA is computed on a single consistent basis — EV ÷ (company-reported operating profit + D&A) — for all three names. Third-party screens show 28.0x / 16.1x / 34.4x ²³²⁴²⁵; the Furukawa screen figure of 28.0x is not internally consistent with the same screen’s own reported EBITDA (which implies ~32x), so the model uses the consistent computed basis.

Two reconciliation points are worth stating plainly, because each is also a thesis point:

• The EV/EBITDA basis. On a consistent OP-plus-D&A definition, Furukawa computes at ~32x — only modestly below Fujikura’s ~34x and double Sumitomo’s ~16x. The widely-quoted third-party screen shows ~28x for Furukawa, but that figure is internally inconsistent with the same screen’s own reported EBITDA (~¥107B, which implies ~32x), so the model uses the consistent computed basis rather than the screen’s published multiple ²³. The takeaway survives either number: Furukawa is priced close to Fujikura on far less AI purity.
• Forward P/E and the guidance gap. The model’s forward P/E uses company guidance, not Street consensus — which matters most for Fujikura. On a like-for-like operating-profit basis, Fujikura’s FY2027 OP guidance of ¥211B (+11.8%) came in roughly 20% below the ¥263.6B analyst (QUICK) consensus the Street had penciled in ²²¹; management guided conservatively and the stock fell as much as 17% on the three-year plan ³². A guidance-based forward multiple therefore looks higher than the consensus-based screen — the conservatism is the signal, not noise.

The shared re-rating. Over the trailing year all three names compounded several-fold — Furukawa +545%, Sumitomo +308%, Fujikura +283% ²⁸²⁹³⁰ — and Fujikura’s multi-year record is extraordinary (+~400% in 2024 as the best Nikkei 225 performer, +~155% in 2025) ³³³⁴. The point of the chart below is not the absolute moves but their togetherness: this cohort re-rates and de-rates as one.

5. The names, one by one

Fujikura (5803.T) — the purest, the priciest, at peak cycle

Fujikura is the highest-quality and highest-purity vehicle. FY2025 was a record across the board: sales ¥1,182.4B (+20.7%), operating profit ¥188.7B (+39.2%) at a 16.0% margin, net income ¥157.2B (+72.5%) ²²²²²². The balance sheet is pristine — a 57.8% equity ratio, ¥181.2B of cash, and an essentially net-cash position ³⁵³⁵ — which on period-end equity puts ROE at ~28% (and higher on an average-equity basis). The AI-optical engine is the whole story: telecom OP up 1.7x to ~81% of group profit ¹¹, driven by differentiated ultra-high-density cable, a Verizon-certified air-blown product, and supply relationships across all the hyperscalers ¹¹³⁶³⁶. Management’s new mid-term plan targets ¥315B of OP at a 28.5% ROE, with up to ¥260B of US investment to roughly quadruple optical-fiber-cable capacity versus FY2022 ³⁷³⁸³⁸.

The bull case is that AI-optical demand compounds, fiber stays tight, and Fujikura’s purity plus best-in-class margins and balance sheet make it the franchise to own; Goldman rates it Buy at a target 19.0x EV/EBITDA — the highest multiple in its Japan coverage ¹³. (The GS targets cited throughout are from its January 2026 update, which predates the May-2026 FY2025 results; treat them as a relative-conviction signal, not a refreshed price view.)

The bear case is valuation and cycle. At ~34x EV/EBITDA and ~13x book ²⁵, the stock prices in continued hypergrowth — Fujikura’s ~¥7.3T market cap now exceeds three-quarters of Sumitomo’s on roughly a quarter of Sumitomo’s revenue ³⁹. It guided FY2027 below consensus ²², and its three-year plan landed so far short of expectations (¥315B FY2029 OP vs a ¥460.5B consensus) that the stock plunged 17% on the day ³²; a US tariff dispute cost ¥12.8B this year ¹⁸; and >50% US revenue concentration adds policy risk ³⁴. You are paying the peak multiple for peak purity at what may be peak cycle — the asymmetry that makes the single name riskier than its quality suggests.

Furukawa (5801.T) — the geared middle, and the one to handle with care

Furukawa is the most operationally and financially geared of the three, and the most awkward on valuation. FY2025 sales were ¥1,307.6B (+8.8%) with OP up 35.8% to ¥63.9B — but at just a 4.9% margin, a third of Fujikura’s ²¹²¹. Reported net income of ¥72.5B (+117%) and a 19.1% reported ROE (17.4% on period-end equity in the table above) flatter the picture: ~¥29.1B was one-off extraordinary income, so the operating read is the cleaner +36% ²²¹. Furukawa is building its AI-optical position aggressively — a reorganization to “accelerate growth in the data center domain,” the Lightera optical brand, the ¥4.4B acquisition of (loss-making) Fujitsu Optical Components, and a 67% stake in connector-maker Hakusan ^240414243. The datacenter segments are guided from ~42% to ~70% of group OP ².

The bull case is the operating leverage: a low-margin base means datacenter mix-shift drops disproportionately to profit, and FY2026 guidance of ¥95B OP (+48.8%) reflects exactly that ²¹. Furukawa beat every target in its prior mid-term plan ².

The bear case is that the market already pays for it. Furukawa trades at ~32x EV/EBITDA and ~7.6x book — nearly Fujikura’s multiple on roughly half the purity and a third of the margin — while carrying the most leverage (net debt/EBITDA ~2.3x) and the highest beta (1.17) of the three ²³²⁴. Goldman is Neutral at an 8.5x target multiple ¹³. The +545% trailing-year move is the largest of the cohort ²⁸, which cuts both ways: most upside if the theme runs, most downside on a reset. Of the three, Furukawa offers the least valuation cushion for its risk.

Sumitomo Electric (5802.T) — the cheap, diversified, defensive call option

Sumitomo is the large, cheap, diversified blue-chip — and the least AI-levered. FY2025 sales crossed ¥5T for the first time (¥5,110.2B, +9.2%) with OP of ¥418.2B at an 8.2% margin ³³³. But the company is dominated by automotive wiring harnesses (57.5% of sales) and power/energy infrastructure (23.1%); optical is just 6.4% of sales, and the genuinely AI-datacenter slice is ~2–3% ³³³²⁰. The AI-optical piece is nonetheless high-quality and accelerating — Infocommunications OP margin reached 19%, and the segment is guided to ¥500B sales (+53%) and ¥130B OP in FY2026 with capex tripling ⁴⁴³³³ — and Sumitomo holds genuine strategic optionality as a named NVIDIA CPO partner ¹⁴.

The bull case is valuation and defensiveness: at ~16x EV/EBITDA, ~3.5x book, a 25.6x trailing P/E and a ~1.2% yield ^26262624, Sumitomo is half the multiple of the other two, with the lowest beta (0.89) and the diversification of a true industrial conglomerate. Its 2028 mid-term plan targets ¥6T sales and ¥600B OP at a >13% ROE ⁴⁵⁴⁵⁴⁵. If AI-optical compounds, Infocommunications could grow toward ~20% of the company ⁴⁴ — a real call option you are buying cheaply.

The bear case is the mirror image: you get the least AI exposure, so in an up-tape Sumitomo lags (it returned “only” +308% versus Furukawa’s +545%) ²⁹²⁸, and the bulk of the company is exposed to autos and tariffs rather than datacenters. Tellingly, Goldman’s Buy target of ¥7,350 sits below the current price — the market has run past even the bullish sell-side case ⁴⁶. Sumitomo is the way to own the theme with a seatbelt — but the seatbelt is most of the car.

6. The basket decision

The case for treating these as a basket rests on one empirical fact and one structural one.

Empirically, they move together. They are a named cohort that multi-bagged in unison and sold off in unison — all three fell roughly 7% on the 2026-06-08 session alone, and the broader May-2026 cable-maker rout hit all three after Fujikura’s weak guidance ⁵³². High co-movement means a basket does little to diversify theme risk.

Structurally, the single-name risks are idiosyncratic and offsetting. Fujikura’s risk is a growth-guidance miss; Furukawa’s is leverage and an expensive multiple; Sumitomo’s is auto/tariff drag and AI under-exposure. An equal-weighted basket diversifies precisely these company-specific risks while retaining full thematic beta.

The model makes the trade-off explicit:

• Equal-weight basket: ~27x EV/EBITDA, ~39x P/E, ~8.0x P/B, ~20% ROE, ~0.95 beta. It captures the secular driver and dampens single-name blow-up risk.
• Cap-weight basket: ~25x EV/EBITDA — lower, because Sumitomo’s ¥9.5T market cap pulls ~48% of the weight, diluting the AI exposure toward the cheapest, least-pure name. Cap-weighting is the more defensive, less pure basket; equal-weighting is the cleaner thematic expression.
• Downside is shared. The sensitivity table shows a 30% sector multiple compression takes all three down 30–32% (leverage makes Furukawa worst, net cash makes Fujikura best) — the basket falls ~31%. A basket spreads idiosyncratic risk; it does not protect against a theme-wide de-rating.

How to choose:

7. Risks and what would change the view

• AI-capex digestion. The dominant risk. Any hyperscaler capex pause de-rates the entire cohort together; the May-2026 rout was a preview. Watch hyperscaler capex guidance and the 800G/1.6T transceiver-speed transition ¹³.
• Fiber over-supply. Today’s shortage and price hikes are the near-term tailwind ¹³; capacity additions (including Fujikura’s planned 4x US expansion ³⁸) could swing the cycle to over-supply within a few years.
• The CPO timeline disappoints. If copper holds in the rack longer and CPO scale-up slips past 2028, the highest-multiple names (Fujikura, Furukawa) lose their forward justification first ¹⁷¹⁷.
• Fujikura execution / guidance. Already a live risk — a below-consensus three-year plan and a ¥12.8B tariff hit ¹¹⁸. Further conservatism or US trade friction (>50% US revenue) would pressure the priciest name ³⁴.
• Furukawa leverage. At ~2.3x net debt/EBITDA and a 4.9% margin, Furukawa has the least cushion if the cycle turns ²²¹.
• FX. Reported in JPY; USD/JPY at 160.13 is near the top of its range, so yen strength is an earnings headwind for the US-heavy names ³¹.
• Split mechanics (technical, not fundamental). All three are splitting — Fujikura 6:1 (done April 2026), Furukawa 10:1 (1 July 2026), Sumitomo 4:1 (record date 30 June 2026) ⁴⁷²¹⁴⁸. Splits change per-share optics, not value; the comps here are stated on market-cap and aggregate bases to stay split-invariant.

What would change the view to “concentrate, don’t basket”: clear evidence that AI-optical demand is durable through a capex cycle (multi-quarter hyperscaler order visibility) and that Fujikura’s premium is supported by widening margin/share — that combination would favor concentrating in the purest name over diluting into the cohort.

8. Bottom line

This is a rare case where the cleanest way to play a powerful theme is also a diversifiable one. The three Japanese fiber majors give you the same AI-datacenter-optical driver at three different concentrations, valuations, and risk profiles. After a year in which all three tripled-to-quintupled and then wobbled together, the prudent expression is an equal-weighted basket sized as a thematic satellite — full theme beta, diversified single-name risk, entry-sensitive. Investors with conviction on durability and quality should concentrate in Fujikura; investors who want the theme with downside protection should own Sumitomo; and Furukawa, the expensive, geared middle, earns the most caution. The one thing the basket cannot do — and no construction can — is protect against the theme itself cooling. That, not stock selection, is the risk to underwrite.

Sources

Full provenance (verbatim quotes, source URLs, local snapshots, retrieval dates) is in the findings sidecar 2026-06-08-japanese-fiber-majors-deep-dive.md.findings.jsonl (216 findings, S001–S216). Headline anchors: company FY2025 results (tanshin / Fact Book / results presentations, filed May 2026) for all financials and segments; stockanalysis.com and FMP for same-day market data; Goldman Sachs “Era of Optics” / Japan Technology (via Noldor) for ratings and target multiples; Valuates/QYResearch, Cignal AI, MarketsandMarkets, LightCounting, and a Noldor fiber-optics demand digest for market sizing; Bernstein (via Noldor) for the CPO/copper reality check. The comparative model is japanese-fiber-majors-comps_v1_2026-06-08.xlsx; charts are rendered from it and re-verified against its cells at audit.

Prepared for John Hendren. Source: named primary providers, Hendren analysis. Not investment advice; figures are indicative and as of 2026-06-08.