LITE — Lumentum Holdings Inc.

IDENTITY

Name        Lumentum Holdings Inc.  ("Lumen", the unit of light, + "-tum")
Ticker      LITE
Form        publicly traded C-corp
Filed in    Delaware
HQ          San Jose, CA — 1001 Ridder Park Drive
Employees   13,757
Sector      Technology / Communication Equipment
⚠ FY ENDS   late June.  FY2026 = the year ended 2026-06-27.

Lineage: spun out of JDS Uniphase, 2015. JDSU peaked at $146.32 in Mar 2000 and fell 88% in twelve months; FY2001 loss $50.6B, largest in US history. LITE is the optics half of that wreckage.

⚠ The parallel, and where it BREAKS. Same: picks-and-shovels supplier · a giant "ignore it" non-cash charge · insider selling during the run. Different: JDSU built for demand that never came (dark fiber) and bought growth with inflated stock (86% goodwill). LITE cannot build fast enough for demand already here, and is paying cash for its own fab. ★ The charges are OPPOSITE. JDSU's $50.6B marked failure — assets turned out worthless. LITE's $7.7B marked success — converts only convert when the stock soars. Same defense, different event.

⚠ "Inc." does not imply C-corp — S-corps use "Inc." too. The reliable rule is that publicly traded → must be a C-corp. A C-corp is a separate legal person: it pays its own tax, then I pay again on any dividend I receive. That is double taxation, and it is why Buffett prefers buybacks to dividends — a buyback returns capital without triggering the second layer. My max loss is my basis.

Delaware matters for a specific reason. Its Court of Chancery hears corporate disputes with no juries — career corporate judges only. That is what companies are buying when they incorporate there: predictable, expert adjudication rather than a jury's sympathies.

Capital returns: no dividend. [A] $102M of stock bought back in FY2026 — consistent with the logic above.

---

WHAT THEY SELL

Lasers and optical modules that convert electricity → light → electricity, so GPUs in an AI data center can talk over fiber instead of copper.

★ SOURCE, NOT SENSOR — precisely. LITE ships both directions (a transceiver = transmit + receive, so their modules contain photodetectors too). But the moat is entirely on the transmit side; the receive side of their own product is near-commodity.

Three product lines:

1. EML laser chips — the crown jewel
2. Transceivers — finished 800G / 1.6T plug-in modules, built in Thailand
3. Optical Circuit Switches (OCS) — route light without converting it

to electricity; cut power up to 40%

Reported two ways:

• By segment: Cloud & Networking ≈ 88% of revenue; Industrial Tech is

the leftover from the old JDSU business — incl. the 3D-sensing lasers in Face ID. Real sensing, but NOT the AI business I own.

• By category: Components (individual parts) vs Systems (parts

packaged into modules). That is the component/system hierarchy, not marketing language.

---

THE PRODUCT — how it actually works

EML = Externally Modulated Laser. It keeps the laser burning continuously and blinks the light using a separate shutter built beside it on the same chip; switching the laser itself would make its wavelength wobble ("chirp") and smear the pulse over distance.

Units

• Gbps = gigaBITS per second, not gigabytes. 8 bits = 1 byte, so

200Gbps = 25GB/s. Networking is always bits, storage always bytes.

• A lane is one light path. Modules stack them:

200Gbps × 8 lanes = 1.6T. That is why 200Gbps/lane is the prize — stuck at 100Gbps you'd need 16 lanes for the same module.

Why it's hard: two devices integrated on one sliver of indium phosphide, with the shutter opening and closing 200 billion times a second.

Why copper fails — and it is NOT speed. Copper carries signal at about 2/3 c; fiber at about 0.68 c. Effectively the same. Copper loses on three other things: power and heat, crosstalk between adjacent wires, and attenuation — which worsens as frequency rises, smearing the pulses into each other until the bits are unreadable.

★ Copper's usable distance collapses as speed rises. At 200 Gbps/lane copper reaches roughly 1–3 metres; fiber reaches kilometres. A single rack is taller than three metres. That gap is the entire business — it is not that copper is slow, it is that at these speeds copper can no longer cross the machine.

★ Light does not eliminate the power problem — it moves it, out of the wire and into the transceiver. Conversion is never free: it costs latency, money, and parts that can fail.

---

THE MOAT — and its clock

[F] EML makers, ALL speed grades: ~5
    LITE · COHR · AVGO · Sumitomo · Mitsubishi
[F] Shipping 200Gbps-per-lane IN VOLUME today: essentially just LITE
[F] LITE EML share, all grades: ~50–60%
[A] 200Gbps parts = >25% of LITE's EML revenue (Jun-2026 qtr)

★ The moat is the SPEED GRADE, not the product. Five firms make "EMLs" the way five make "cars" — that tells me nothing. It doesn't erode; it ends the quarter a rival qualifies at 200Gbps.

Pricing power — passes hard today. LITE sets the price; the customer negotiates the volume.

[A] demand > supply by ~30%, AFTER a 20% capacity add
[F] customers hunting a second source
[C] double-digit 200Gbps price increases expected in 2026

⚠ Counterweight [A] two customers = 26% + 12% of revenue (Jan–Mar 26). Buyers that big always negotiate. ⚠ The clock [C] industry capacity ~doubling to 50.7M units/mo in 2026. The cure for high prices is high prices.

---

WHAT KILLS IT

★ The threat is not a better EML — it is NOT NEEDING ONE. Kodak lost to no film, not better film. Nobody is going to out-manufacture LITE at EMLs.

1 · CPO / co-packaged optics — the structural risk

An EML does two jobs: generate light, and blink it. The blinking is the moat. CPO moves the optics onto the switch chip's own package and hands the blinking to a silicon modulator; the laser demotes to a dumb always-on CW-DFB.

[F] CW-DFB capacity is led by AVGO + Sumitomo (then COHR, LandMark/
    LuxNet) ≈ 74% of market. LITE is NOT the leader there.
[C] broader CPO ramp   H2 2026
[G] NPO deployments    late 2027
[G] inflection         2028

⚠ AVGO is both the loudest CPO advocate and a CW-DFB leader. Incentive. ★ Silicon cannot lase (indirect bandgap) — somebody always supplies the laser. The risk is demotion, not extinction: from the hard part to the cheap part. Demotion shows up in GROSS MARGIN, not revenue. ⚠ Honest counterweight: CPO has been "two years away" since ~2015. Pluggables keep winning because a failed module can be swapped; a failed co-packaged optic means replacing the switch. ✓ LITE is hedged — [A] a multi-hundred-$M NVDA order for CPO external laser sources, plus OCS as a second leg. → 2028 risk. Did not touch the 3–5 month trade. It DOES undercut thesis layer (1), "capex 2–5 years and LITE is the winner" — 2–5 years lands 2028–2031, exactly when the architecture may stop needing the EML.

2 · Customer concentration [A] two customers = 26% + 12% of revenue (Jan–Mar 2026). Buyers that size always negotiate.

3 · Nvidia's $2B — read it as a BUYER, not an investor

[A] $2B cash → 2,876,415 NEW LITE preferred @ $695.31, 2026-03-02
    ≈ 3% of the company. Converts 1:1; the "preferred" label is
    almost certainly for antitrust timing.
[A] the IDENTICAL $2B went to COHR the same day. Both NON-EXCLUSIVE.

★ NVDA manufactured itself a second source. A monopoly supplier is its problem, not its asset. Non-exclusive is the tell — exclusivity was available and they declined it. ⚠ CORRECTION on the record: $695.31 was ~0.8% BELOW the 02-27 close of $700.91 — i.e. the market price, not an informed discount. The gap to my $833.91 is five months of tape, NOT superior information. 📌 What the tape did instead: 700.91 → 783.25 (announcement, +11.7%) → 558.44 four sessions later. The market priced the dilution and the duopoly immediately. Bull read: nobody puts $2B into a supplier they plan to design out. Bear read: they funded a duopoly so they never pay monopoly prices. Both true — same fact, opposite sides of the invoice.

4 · Dilution, disaggregated. ~70M → ~90M shares FY2025→now (+29%). NVDA was only ~2.9M of that; the convertible bonds were ~17M — the same event as the $7.7B GAAP charge. (Caveat: converts were already in the diluted count, so true economic dilution is smaller than 70→90 looks.) Calibration: LITE gave up ~3% once for $2B; AAOI gives up ~14% per quarter.

---

THE NUMBERS

WATCH EVERY QUARTER — the turn shows up here first:
  1 non-GAAP operating MARGIN (Q4 FY26: 36.6%) — falling = power going
  2 SEQUENTIAL vs prior quarter, NOT year-over-year
  3 gross margin — it breaks before operating margin does

[A] FY2026 revenue  $3.014B, +83.2% YoY (from $1.645B)
    year ended 6/27/26, reported 8/11/26
    verified: 3.014 / 1.645 = 1.832 → +83.2%
[G] FQ1'27 revenue  $1.225–1.275B  (Jul–Sep 26, reports ~11/05)
[C] FY2027 EPS      ≈ $33          (20 analysts)

Which tag is the price built on? The fact is $3B of revenue. The $830 price is paid against the [C] line — the weakest tag on the page. That gap is the whole argument. → Chapter III.

⚠ GAAP FY2026 net income is a large LOSS and it is not real — a one-time non-cash charge sits between the operating line and the bottom line. Chapter II — GAAP vs non-GAAP.

---

VALUATION (Chapter III · priced 2026-08-26 @ $939.03)

[A] price $939.03 · mkt cap $84.2B · 89.7M sh
[A] P/S trailing 27.9x · P/B 17.9x (book $52.41/sh)
[A] FCF yield 0.36% (FY26 FCF $300M) · OCF yield 0.89%
[M] ROIC ~42%  (Q4 non-GAAP op annualized / invested capital ex-cash)

★ THE SAME STOCK HAS THREE HONEST P/Es. Nothing about the company differs between them — only which year I choose to price.

[A] 114x  on FY2026 non-GAAP actual        $8.23
[G]  56x  on LITE's own Q1'27 guide, annualized  $16.80
[C]  28x  on consensus forward             $33.01

The spread from 114x to 28x is entirely how much future I treat as already banked. At $939 the market prices the [C] line — the weakest tag on the page — which needs EPS to quadruple in about a year.

The split verdict, and both halves are true:

• The business is outstanding. 42% ROIC. Buffett's own test.
• The price is not. 27.9x sales, 0.36% free-cash yield — a T-bill

pays roughly ten times that cash yield with no risk. Buffett passes, and not because the company is bad.

⚠ Multiple compression is faster and larger than the upside from a beat. If EPS lands at $20 instead of $33 and the multiple settles at a still-rich 30x, that is $600 — down 36% from here. Nothing needs to go wrong; consensus merely needs to be optimistic.

What paying $939 requires me to believe: the Q1'27 guide is a floor and not a peak quarter · ~40% operating margins hold while industry capacity doubles · no CPO drag arrives early.

📌 Re-entry frame. My three entries averaged $833.91. $939.03 is +12.6% above that, and the written discipline is enter on washouts — this is the opposite of one. Noted as a fact, not a plan.

---

CUSTOMERS & DEMAND

[A] MSFT+GOOGL+AMZN+META capex = $165.1B, Apr–Jun 2026, +87% YoY
    YoY growth, last 4 qtrs: 65 → 64 → 81 → 87 (ACCELERATING)
    ⚠ OMITS ORACLE — understates the true buildout
[F] 1 Nvidia GB200 NVL72 rack ≈ 5,000 optical connections

Hyperscaler capex is a spender, LITE is a supplier. Their expense line is Lumentum's revenue line.

---

MY POSITION — FLAT

Closed 2026-08-25. No LITE position.

TRIP 3   bought 34 sh @ $833.91 avg  =  $28,352.93   (08-19/20/24)
         sold   34 sh @ $881.55 avg  =  $29,972.62   (08-25)
         realized  +$1,619.69  (+5.71%)

LITE lifetime, acct 660 — 3 round trips, 3 green:
  07-31 → 08-04   +$1,382.35
  08-11 → 08-12   +$1,472.95
  08-19 → 08-25   +$1,619.69
  ────────────────────────────
                  +$4,474.99

⚠ SCORE THE EXIT — sold a day EARLY against my own written plan. The plan said flat at the Wed 08-26 close. I sold Tue 08-25 at $881.55. LITE closed 08-26 at $939.03.

  34 sh × ($939.03 − $881.55)  =  $1,954.32 left on the table

The trade was still green and the purpose was served — flat before the NVDA print. But the exit beat its own deadline by a day and cost more than the trip made. This is the documented pattern: exiting early under pressure. Recorded, not excused.

THESIS & EXIT

Not about the business — about my trade.

Thesis (opened 2026-08-03, CLOSED 2026-08-26 — archived, not judged). AI capex runs 2–5 years · LITE is the picks-and-shovels winner · the name is a volatile cyclical, so I enter on its washouts. Target was 1,100–1,200 within 3–5 months.

⚠ Closed as a historical artifact when the whole thesis layer was retired at my direction — neither proven nor refuted. The $1,150 target and the late-Jan-2027 hyperscaler guide-down clock go unscored. The argument below is preserved because the reasoning is still what I'd re-enter on; the record is closed. Full outcome note: theses table, id 1.

Position exit — EXECUTED 2026-08-25, one day ahead of the written plan. Flat before the NVDA print, which was the purpose. See MY POSITION for the score. Thesis remains OPEN and unexpressed.

Thesis exit — retired UNFIRED when the thesis closed. Neither trigger went off; nothing about the moat changed. Keep both — they are the re-entry test as much as the exit test.

1. Language. LITE drops "supply constrained" / "demand exceeds supply"

from its own earnings commentary; or pricing goes from "increases" to "stable"; or lead times shorten.

1. Second source. A named rival (COHR · AVGO · Sumitomo · Mitsubishi) is

qualified into a customer's design at 200Gbps/lane. Qualified — not a demo, not sampling. Qualification is when the customer legally has somewhere else to go.

(1) leads (2): LITE sees its own order book before the market sees the rival.

---