MRVL, LITE, AAOI: The Only Multiple That Survives a Shortage Ending — and the Market Has It Exactly Backwards

Daily Analysts · Monday 24 August 2026, pre-market. All prices are Friday 21 August 2026 closing prices. Marvell $237.04 (−5.57%) · Lumentum $866.71 (−1.43%) · Applied Optoelectronics $124.82 (−3.32%) · Coherent $289.52 (−0.18%) · Broadcom $368.45 (+1.21%) · Nvidia $214.72 (−0.98%) · SPY $765.72 (+0.41%) · QQQ $713.44 (+0.35%).

1. Where I agree and where I disagree with the prevailing take

I agree with the consensus that the optical/interconnect layer is the genuine bottleneck in AI build-outs, that demand is capacity-bound rather than order-bound, and that the shortage in EML and CW lasers persists into 2027. That is well-covered ground and the supply-chain work backs it: an optical component shortage running roughly 50% below demand, with Nvidia locking up ~$4B of laser supply and pushing rivals past 2027, and silicon-photonics foundry capacity constrained at every layer — SOI wafers, InP substrates and foundry slots — with capacity being reserved through customer prepayments.

I disagree on three points, and each one is a trade:

1. The Marvell/Google warrant is not 7% dilution. It is a ~0.9% economic cost at today's price with $12.18B of cash coming the other way. The market priced the share count and ignored the consideration. Detail in §3.
2. Lumentum is not a loss-making company. Screens show TTM EPS of −$82.21 and FY2026 GAAP EPS of −$92.96. That is an artifact of a ~$7B non-cash charge from equitising convertible notes. The operating business printed a 50.4% gross margin and $3.23 of non-GAAP EPS last quarter. Detail in §4.
3. AAOI's problem is not the $600M ATM. Everyone is writing about the dilution. The dilution is 5.7% — irritating, not fatal. The actual problem is that the ATM exists because the company burned roughly $900M of free cash flow on $596M of trailing revenue. Detail in §5.

My genuinely new contribution is the gross-profit-multiple ranking in §2, which inverts the popular ordering of these three names, and the observation that it holds on both trailing and forward numbers — so it is not a base-effect illusion.

2. The ranking table — one number that reorders the group

In a shortage, revenue growth tells you nothing about who wins, because everyone is sold out. What tells you who wins is what share of each revenue dollar the company keeps. Here is the group, priced on gross profit.

*Forward gross profit: MRVL on the company's raised FY2028 revenue target of $16.5B at 52% GM; LITE on annualised FQ1 FY2027 guidance midpoint ($1.255B × 4 = $5.02B) at 50.4% GM; AAOI on management's FY2026 revenue guide of >$1.1B at 30% GM. These are my calculations from the disclosed figures below, not consensus estimates.

3. Marvell — the warrant everyone read as dilution is a $12.2B pre-paid order book

What was actually filed

Marvell filed an 8-K on 19 August 2026 disclosing a warrant issued to Google for up to 58,970,907 shares at a strike of $206.58, vesting principally against revenue tranches and discretionary purchases, tied to an expanded custom-silicon agreement supporting Google's TPU ecosystem. Full exercise equals $12.18B of cash paid to Marvell. The stock rallied ~7–10% on the news on 18–19 August, then gave it all back and more, closing Friday at $237.04, −5.57%, pacing decliners on profit-taking while the S&P rose 0.43%. Broadcom, the incumbent TPU partner, closed +1.21%.

The arithmetic the sell-down ignored

58.97M shares on ~821M outstanding is 7.2% — that is the number in every headline. But Marvell is not giving the shares away; it is selling them at $206.58. The cost to existing holders is only the discount to market:

Nine-tenths of one percent. The market marked the stock down 5.57% — roughly six times the economic cost — for a contract that simultaneously commits the world's second-largest custom-silicon buyer to a multi-year volume ramp. My view: that is the wrong trade and it is fadeable.

The operating business, from the filing

From the Q1 FY2027 earnings exhibit filed with the SEC: net revenue $2,418M, up 28% y/y, a record; data-center revenue $1,832.7M, or 75.8% of total; GAAP gross margin 52.1%; operating margin 14%; GAAP EPS $0.04. Guidance for Q2 FY2027: $2.7B ±5%, roughly +35% y/y, with non-GAAP EPS of $0.93. Management's language in the release is unusually direct: “We expect revenue growth to continue accelerating each quarter throughout fiscal 2027 … We are seeing exceptional AI-related bookings, and as a result, we are significantly raising Marvell's revenue outlook for both fiscal 2027 and fiscal 2028,” naming 800G and 1.6T scale-out optics, 51.2T Ethernet switches and custom XPU and XPU-attach solutions. The FY2028 revenue target was raised to $16.5B. For context, FY2026 revenue was $8.195B and FY2025 was $5.767B.

At $194.6B, Marvell is 18.0x the annualised Q2 guide and 11.8x the FY2028 target. Consensus into the print is ~$2.7B and roughly $0.93 non-GAAP EPS; the report lands after the close on Thursday 27 August.

4. Lumentum — the −$92.96 EPS on your screen is not real, and the balance sheet is now the asset

Lumentum's FY2026 (ended June 2026) figures as they appear on data feeds: revenue $3,014M, gross margin 44.24%, operating income $536.2M, net income −$6,935M, EPS −$92.96. Our own fundamentals feed shows TTM EPS of −$82.21 alongside an ROE of 30.68% — a contradiction that should be the tell.

The reconciliation: the GAAP loss is a one-time non-cash charge of roughly $7.8B related to the equitisation of convertible notes. As the stock ran from a 52-week low of $115.21 (21 Aug 2025) to a high of $1,085.68 (12 May 2026), the converts were settled in equity, generating an inducement/settlement charge of a size that swamps every operating line in the P&L.

The operating quarter

Fiscal Q4 2026 (reported 11 August 2026): revenue $1,006.3M, gross margin 50.4%, EPS $3.23, cloud-and-networking revenue $649.4M, and FQ1 FY2027 guidance of $1.23B–$1.28B. The stock rose 14.89% to $942.75 on the print.

Put those numbers in sequence, because the sequence is the thesis. FY2025 revenue was ~$1,645M. FY2026 revenue was $3,014M, +83.2%. Q4 alone was $1,006.3M — one third of the entire fiscal year in the final quarter. And the FQ1 FY2027 guide midpoint of $1,255M means a single quarter is now 76% of what the company sold in all of FY2025, +25% sequentially and roughly +150% year over year. Gross margin went with it: 44.2% for the full year, 50.4% in Q4.

Valuation: $71.0B on an annualised FQ1 guide of $5.02B is 14.1x forward revenue, or 28.1x forward gross profit. On earnings, Q4's $3.23 annualises to $12.92 (67x); my own estimate — and I flag it as an estimate — is that FQ1 lands near $4.30–4.60 given +25% sequential revenue at 50%+ margin, which annualises to roughly $17–18 and puts LITE near 48–50x forward earnings growing revenue at triple digits. Expensive. Not absurd.

5. Applied Optoelectronics — the growth is real, the economics are not there yet

Let me be straightforward: the demand commentary from AAOI is the most exciting in the group, and I still would not own it here.

From the Q2 2026 earnings call (6 August 2026): revenue $191.9M against $102.95M a year ago (+86%), gross margin 29.8%, EPS $0.06. 800G revenue was $12.8M, more than double Q1. Q3 guidance is ~$255M at the midpoint — a 33% sequential step — with full-year 2026 revenue guided above $1.1B against FY2025 revenue of roughly $456M. Capacity is being pushed toward 200,000 units per month for 800G and 1.6T. Management's framing on the call was as clear as it gets: “Demand to support next generation AI infrastructure remains so robust that our near-term revenue is bounded almost entirely by production capacity and key component availability” and “if we could produce more, we could ship more right now.” A large customer is expected to return as a 10%-plus customer in Q3.

Now the other side of the ledger

On 21 August the company entered a $600M at-the-market equity distribution agreement with Raymond James and Needham, its third trip to the equity market in 2026, taking total 2026 issuance past $1.1B. At $124.82, $600M is ~4.81M shares, or 5.7% of the share count.

And the valuation, which is the part I have not seen anyone else state plainly: at $11.0B on $172M of trailing gross profit, AAOI trades at 64x trailing gross profit — about 20% more expensive than Lumentum's 53x — while earning a 28.9% gross margin versus Lumentum's 50.4%, and burning cash instead of generating it. You are paying a premium for the worse business. On forward numbers the gap narrows but does not close: 33.3x versus 28.1x.

6. The macro overlay — this week is a minefield for all three

None of these positions exist in a vacuum. The 10-year closed Friday at 4.734% and the 30-year at 5.273%, within a few basis points of a 19-year high. Our standing regime level is that a 10-year sustainably above 4.75% re-engages AI multiple compression — we are essentially there. Three specific downstream consequences for this group:

• Rates → multiple. Marvell at 11.8x FY2028 revenue and Lumentum at 14.1x forward revenue are pure long-duration equity. A 10-year that clears 4.85% takes the highest-multiple names first, and LITE (14.1x) is more exposed than MRVL (11.8x) despite being the better-margin business. Below 4.45% and the whole complex re-rates higher.
• Nvidia, Wednesday 26 August. The entire optical complex is a derivative of Nvidia's shipment cadence. A soft guide compresses LITE and AAOI harder than MRVL, because Marvell has a diversified custom-silicon and switching business while the module names are almost pure GPU-attach.
• Warsh at Jackson Hole, Friday 28 August. The new Fed chair must reconcile a 9–3 July hold — with all three dissents for a hike — against a Treasury actively suppressing long yields. A hawkish read is a direct hit to 2.3-beta MRVL and 3.85-beta AAOI. Marvell reports on Thursday 27th, into that.

Ordering matters: NVDA Wednesday, MRVL Thursday, Warsh Friday. Anyone sizing full positions before Wednesday is taking three sequential binary events, not one.

7. Suggestions

HIGH CONVICTION [MRVL-2] LONG Marvell Technology

• Reference: $237.04 (21 Aug close)
• Entry zone: $225–$242. Half size before the 27 August print; add the balance after guidance.
• Target: $300 (+26.6%) — 14.6x the FY2028 $16.5B revenue target, below the 18.0x currently paid on the Q2 annualised run-rate.
• Invalidation (price): a daily close below $198 — decisively through the $206.58 warrant strike, the level at which the Google option carries no intrinsic value and the “validation” leg of the thesis has no economic anchor.
• Invalidation (fundamental): Q3 FY2027 gross margin guided below 51.0% on the 27 August call. That is the number that proves the warrant was a price concession rather than a partnership.
• Timeframe: 1–3 months
• Signals (4, independent): (1) Q1 revenue +28% with data centre at 75.8% of mix and a Q2 guide of +35% — accelerating, not decelerating; (2) FY2028 revenue target raised to $16.5B in the company's own release; (3) the warrant is a contractual multi-year volume commitment from Google carrying $12.18B of cash consideration; (4) cheapest in the group on both trailing (43x) and forward (22.7x) gross profit.
• Risk, quantified: this is an earnings-event long. Implied moves on MRVL prints have run ±8–10%; a bear-case guide takes the stock to the $195–210 zone, roughly −13% to −18% from the entry midpoint. Size accordingly — this is a half-position-before, half-after trade, not a full-size hold into the number.

HIGH CONVICTION [LITE-2] LONG Lumentum — buy weakness, do not chase

• Reference: $866.71 (21 Aug close), 20.2% below the $1,085.68 high
• Entry zone: $790–$870. Scale: 60% in the zone now, 40% reserved for a flush toward $760 on any Nvidia-driven sector drawdown.
• Target: $1,050 (+21.1%) — 17.1x the annualised FQ1 guide, a premium to today's 14.1x justified only if the +25% sequential cadence repeats once more.
• Invalidation: a weekly close below $700. That breaks the entire post-11-August breakout structure and would mean the market has stopped believing a guide the company has already put in writing.
• Timeframe: 1–3 months
• Signals (3, independent): (1) FQ4 revenue $1,006.3M at a 50.4% gross margin with $3.23 EPS, and FQ1 guidance of $1.23–1.28B, +25% sequential; (2) the convertible overhang has been equitised away — the balance sheet is clean entering a prepayment-driven capacity allocation cycle; (3) laser and foundry scarcity documented as persisting into 2027 across independent supply-chain work.
• Risk, quantified: beta 1.53 and a 14.1x forward revenue multiple. A 10-year print above 4.85% or a soft Nvidia guide is a −15% to −20% event on this name before any company-specific news. The bear scenario at $700–750 is −14% to −19% from the entry midpoint.

SPECULATIVE [OPTICAL-PAIR-1] LONG Lumentum / SHORT Applied Optoelectronics

• The trade: the gross-margin spread. LITE 50.4% versus AAOI 29.8% — 20.6 points — while AAOI trades at a higher multiple of gross profit. When the shortage normalises, module assemblers with sub-30% margins have no cushion; component owners do.
• Entry: initiate at the current ratio of 6.94x (866.71 ÷ 124.82) or better.
• Target: ratio 9.0x (+30% on the spread)
• Invalidation: AAOI posts gross margin above 35% for two consecutive quarters with positive operating cash flow — that would prove the $565M capex programme is converting into structural margin and the short leg is wrong. Secondary stop: ratio below 5.80x.
• Timeframe: 3–6 months
• Why only SPECULATIVE: the analysis is high-conviction; the mechanics are not. AAOI has a 3.85 beta, an $18.50-to-$233.67 52-week range, and a heavy retail following — borrow may be expensive or unavailable, and squeeze risk in a sold-out capacity cycle is genuine. Anyone who cannot source cheap borrow should express this as an overweight-LITE / zero-AAOI allocation rather than a paired short.

WATCH [AAOI-WATCH-2] Applied Optoelectronics — the level where I become a buyer

• No position at $124.82. I am not short outright and I am not long.
• Buy trigger: $85–$95 combined with a printed gross margin above 32%. Price alone is not enough — the whole objection is margin, so the entry must be margin-conditional. At $90 with 32% margins, AAOI is ~$7.6B on ~$350M of 2026 gross profit, or 22x, which is finally cheaper than Marvell.
• What would make me buy earlier: the ATM being retired substantially undrawn, which would say the capex is now self-funding.
• Timeframe: 3–6 months

8. What I would tell you in one paragraph

Buy Marvell into the 27 August print at half size, because the market charged it 5.6% for a contract whose economic cost is 0.9% and whose consideration is $12.18B of cash, and because on the multiple that survives a shortage ending it is the cheapest name in the group. Own Lumentum as the quality position, but on weakness rather than at the highs, and understand that the −$92.96 EPS on your screen is a convertible-note artifact concealing a 50.4%-margin business whose balance sheet is now clean. Do not own Applied Optoelectronics here: the demand is genuine and the management commentary is the most bullish of the three, but a company burning $1.51 of cash per dollar of revenue at a 28.9% gross margin, funding itself with its third equity raise of the year, has no business trading at a higher multiple of gross profit than either of them. If the shortage ends on the 2027 schedule the supply chain is modelling, that is precisely the ranking that will matter.

Primary sources

• Marvell Technology — Q1 FY2027 results, 8-K Exhibit 99.1 (SEC EDGAR)
• Marvell Technology — Form 8-K filed 19 August 2026 (Google warrant)
• Marvell Technology — Q1 FY2027 press release (Investor Relations)
• Marvell Technology — Q4 and FY2026 results press release
• Applied Optoelectronics — Q2 2026 earnings call transcript, 6 August 2026
• Applied Optoelectronics — SEC filings index, including the 21 August 2026 equity distribution agreement
• Applied Optoelectronics — $600M at-the-market equity distribution programme announcement
• Lumentum — fiscal Q4 2026 results and FQ1 FY2027 guidance
• Lumentum — FY2026 financial statements · Applied Optoelectronics financials · Coherent financials
• Marvell Q2 FY2027 earnings preview — 27 August date, guidance and consensus
• AI data-centre optical component shortage — McKinsey supply-demand gap and Nvidia laser lock-up
• Silicon-photonics foundry capacity, prepayments and the SOI/InP constraint
• Marvell/Google agreement and the read-across to Broadcom's TPU share

Prepared by Daily Analysts, 24 August 2026. This is analysis, not investment advice. Positions described carry risk of total loss. All price references are 21 August 2026 closes unless stated otherwise.