Business Overview
7 September 2026 · Built from Coherent's own filings FY2016–FY2026 in the provided research folder, plus independent industry and competitor sources. Not a valuation and not a recommendation.
Coherent grows crystals, cuts laser chips out of them, and sells the chips — mostly inside optical transceivers that let AI datacentres move data as light instead of electricity. It is the only company in its industry that owns every step from the raw wafer to the finished module, and that is not currently what earns the highest margin in the industry.
The economic engine
One 6-inch indium phosphide wafer
The saleable unit is one high-speed optical transceiver, but the unit that governs the economics sits upstream. A transceiver is largely a housing for a laser, and the laser is the part that cannot currently be bought in sufficient quantity. Coherent's chief executive: indium phosphide capacity “continues to be our primary constraint”, while transceiver assembly and test capacity is not. Every dollar of Datacenter & Communications revenue — $5,275M in FY2026, at a 25.2% segment margin — is rate-limited by wafers out of Sherman, Texas and Järfälla, Sweden.
Industry unit price
~$248
Average optical module ASP, from the only competitor that discloses units (Innolight, 21.09M modules, 2025). Coherent discloses none.
What the unit earns
25.2%
Datacenter & Communications segment profit margin, FY2026, up from 24.1% and 19.0%. Excludes stock comp, amortisation, corporate cost.
Headline earnings
$4.12
GAAP diluted EPS FY2026; $5.61 non-GAAP. Contains $124M of divestiture gains, $74M on an equity sale, and a 7% tax rate.
Leverage
$1.24bn
Net debt at 30 June 2026, from $2.78bn a year earlier. No leverage ratio is disclosed. $604M of cash is trapped inside the silicon carbide subsidiary.
Cycle position
+49%
Q4 FY2026 revenue above the prior peak quarter. But gross margin is still 250bp below its 2020 peak.
Reported vs organic
+22.5%
FY2026 reported. Datacom +40%, Industrial −10% mostly on disposals. No organic bridge is disclosed.
Inside an AI datacentre the chips that do the computing talk to each other electrically, and an electrical signal falls apart beyond about one to two metres at modern per-lane speeds. Past that distance the signal has to become light, travel down a fibre, and become electricity again. The device that performs that conversion is an optical transceiver, and Coherent is one of a very small number of companies that can build one starting from the raw crystal.
Everything is sold outright. Revenue is recognised at a point in time, generally on shipment, with 30 to 90 day payment terms; contract liabilities were $63M against $7,118M of revenue, so there is no meaningful recurring or deferred revenue. This is a manufacturer that ships goods and gets paid once. It does not disclose how many units it ships or at what price — the 10-K states plainly that revenue is not disaggregated by product because the chief operating decision maker does not use that view.
From 1 July 2025 the company reports two segments instead of three, recasting prior years. Datacenter & Communications is $5,275M, 74.1% of revenue, up 40%: 800G and 1.6T transceivers, co-packaged optics, optical circuit switches, ZR/ZR+ coherent modules for linking datacentres, and the merchant lasers and detectors underneath. Industrial is $1,844M, down 10%: excimer and CO₂ lasers for displays and wafers, ultrafast lasers for chip inspection, and engineered materials including silicon carbide substrates. Segment profit is the company's own generous measure — the gap between the $1,752M of segment profit and $848M of pre-tax earnings is $905M of amortisation, corporate cost, stock compensation and interest.
Read across the reported results of every listed company in the optical chain and gross margin is highest at the two ends and lowest in the middle. Broadcom, which designs the switch silicon and owns no fabs, earned a 67.8% gross margin and 39.9% operating margin. Fabrinet, which assembles modules to order in Thailand, earned 12.0% gross and 10.0% operating. Coherent, which owns more of the chain than anyone, earned 37.5% gross and 12.6% operating.
The second comparison is the uncomfortable one. Fabrinet turns 12.0% gross into 10.0% operating because its operating costs are about two points of sales. Coherent turns 37.5% into 12.6% because its operating costs are roughly twenty-five points, and it spends 15.5% of revenue on equipment while Fabrinet builds sheds. On the operating line, the contract manufacturer is within three points of the vertically integrated technology owner — on a fraction of the capital and none of the technology risk.
| Stage of the chain | Company | Gross margin | Bottom-line margin |
|---|---|---|---|
| Switch and DSP silicon (fabless) | Broadcom | 67.8% | 39.9% op. |
| Module assembly, buys chips | Eoptolink | 47.8% | 38.4% net |
| Chips + modules, buys substrate | Lumentum | 41.7% | 17.4% op. |
| Module assembly, buys chips | Innolight | 42.6% | 28.2% net |
| Substrate → epi → chip → module | Coherent | 37.5% | 12.6% op. |
| Industrial laser, integrated | IPG Photonics | 38.0% | 1.3% op. |
| Sub-scale module assembly | Applied Optoelectronics | 30.0% | −12.0% op. |
| Contract assembly | Fabrinet | 12.0% | 10.0% op. |
One does: qualified indium phosphide epitaxy. Two competing issuers independently describe the same shortage — Coherent calls it its primary constraint, Lumentum says it is “effectively sold out for the foreseeable future despite our rapid capacity expansion”. The upstream metal is constrained too. China refines about 70% of world indium, imposed export controls in February 2025, and unwrought indium exports fell 72% year on year to September 2025 (USGS, 2026).
A second is Coherent's own and is measurable: the 6-inch wafer learning curve. The company claims 6-inch InP delivers four times the capacity and a 60% lower die cost than its legacy 3-inch lines, and its CFO says 6-inch yields already exceed 3-inch across all devices with the bulk of the ramp still ahead. The US Commerce Department calls Sherman, Texas the first and largest high-volume 150mm InP facility and issued a letter of intent for up to $50M of CHIPS support in June 2026.
The barriers most often cited do not bind. LightCounting judged Chinese vendors two to three years behind on 100G-per-lane components in January 2024; by 2026 Innolight and Eoptolink held more than 60% of the market above 800G, and Chinese firms held seven of the top ten positions. In the fourth calendar quarter of 2025, Innolight alone booked $1.87bn against Coherent's $1.2bn of datacentre and communications revenue. Hyperscaler qualification slows share shifts by quarters, not years. And capital intensity only bars entry when capital is scarce — Coherent raised $2.0bn from a single buyer in a single day in March 2026.
Coherent makes four claims about its position. Outside evidence supports one and a half.
Vertical integration into InP is a differentiator — partially supported. The rarity is real: Lumentum buys InP substrate externally and said so on its own call; Innolight derives 98% of revenue from module assembly with no substrate or epitaxy operation disclosed. Coherent appears to be the only participant owning substrate through module. The margin evidence contradicts the value of it. The most integrated company in the set earns a lower gross margin than the two least integrated, and the two other chip-owning integrators — Accelink and Ligent — earn the worst margins of any branded module maker. Ligent's own listing prospectus discloses an optical chip segment running a negative 121% gross margin in 2025 because volume did not cover fixed cost. Integration buys supply access, not margin.
Its capacity is scarce — supported, but the scarcity belongs to the industry. Every participant is adding into the same shortage at once: Coherent doubled internal InP output a quarter early and intends to more than double it again by the end of calendar 2027; Lumentum is expanding two Japanese fabs and converting a US one; Applied Optoelectronics targets 650,000 modules a month by the end of 2026; Fabrinet is building a two-million-square-foot plant adding $3.0–3.5bn of revenue capacity. LightCounting expected the shortage to ease from mid-2026.
It has pricing power in AI datacom — contradicted. Coherent's own risk factor concedes that large customers seek price concessions and will continue to. Its defence is long-term agreements — but the chief executive described those as carrying agreed pricing running three years and in some cases through the rest of the decade, which means it cannot reprice upward into a shortage. Coherent's non-GAAP gross margin rose 152 basis points in FY2026. Lumentum's rose 1,130. Innolight's rose 796.
The materials business is a moat — contradicted, and the disclosure has been withdrawn. Materials was Coherent's highest-margin reported segment in FY2024, at $1,017M of revenue and a 21.0% non-GAAP operating margin. It no longer exists as a reported segment; it sits inside Industrial, which fell 10% in the strongest capital-spending year the industry has ever had. A moat that shrinks in a boom, and whose disclosure is then removed, cannot be tested.
The most concrete threat is a customer moving upstream. At OFC in March 2026 Broadcom announced not only a 400G-per-lane optical DSP but “first-to-market 400G electro-absorption modulated laser and photodiodes”, alongside 200G-per-lane VCSEL, EML and co-packaged optics technologies. Those are Coherent's core categories, entered by a company with a 67.8% gross margin that can fund the attempt out of petty cash. Broadcom does not disclose optical revenue, so the scale is unknown — but a fabless business with two-thirds gross margins subsidising entry into a capital-intensive one with a third is the textbook shape of a profit pool migrating.
On the evidence, winners here own a scarce input the buyer cannot second-source, or sit where their part is a small share of system cost but determines system performance, or are the low-cost producer at scale. Coherent qualifies on the first and is trying to on the third. It does not qualify on the second, and LightCounting's account of why explains most of the industry's history: optics is “a tiny part of a massive industry” yet an “ever-larger portion of the bill of materials” — big enough to be worth squeezing, small enough to have no leverage. Coherent is a good version of a structurally difficult business.
FY2026 revenue grew 22.5%. That is Datacenter & Communications up 40% and Industrial down 10%, and the Industrial decline is attributed primarily to two disposals — so reported growth understates organic growth this year, the mirror image of the usual problem. The reverse distortion sits in the history and matters more.
| Year | Reported growth | What caused it |
|---|---|---|
| FY2020 | +74.7% | Finisar closed Sept 2019 for $2,909M and contributed $938M of revenue in nine months, with a $94.6M net loss. Almost entirely acquired. |
| FY2023 | +55.6% | Coherent, Inc. closed 1 July 2022 for $7,096M and contributed $1,469M of revenue, with a $412M net loss. Almost entirely acquired. |
| FY2024 | −8.8% | Organic. Declines in all four then-reported markets: electronics −43%, instrumentation −17%, industrial −5%, communications −1%. |
| FY2025 | +23.4% | Organic. Communications +51% on AI datacom. |
| FY2026 | +22.5% | Organic and understated. D&C +40%; Industrial −10% on the aerospace & defence and Munich disposals. |
Over the decade revenue rose from $827M to $7,118M — but diluted shares went from 62.9 million to 195.4 million and total debt from $235M to $3,222M. A long-run revenue chart for this company is substantially a chart of acquisitions.
Structural in direction, cyclical in amplitude
1. AI datacentre transceiver volume
Hyperscaler capital spending is the entire mechanism. Amazon's trailing-twelve-month capex to June 2026 was $169.0bn, up 64%; Microsoft spent $115.9bn in FY2026, up 80%; Meta guided calendar 2026 to $130–145bn; Alphabet spent $80.6bn in the first half of 2026 alone. LightCounting puts optics at 2.7% of top-five cloud capex in 2025, rising to 3.1% in 2026. The direction is structural because copper does not reach. The amplitude is a spending cycle set by four balance sheets.
Management-driven
2. Indium phosphide output and 6-inch yield
Coherent produced roughly 80% more InP lasers in the June 2026 quarter than a year earlier and doubled internal capacity a quarter ahead of plan. This is the driver most within management's control, and the one that converts scarcity into Coherent's revenue rather than someone else's.
Management-driven
3. Mix shift out of Industrial
Datacenter & Communications went from 55.9% of revenue in FY2024 to 74.1% in FY2026, partly by growing and partly by selling the other side. Because D&C earns a 25.2% segment margin against Industrial's 22.9%, and Industrial carries 58% of group depreciation and amortisation on 26% of revenue, the mix shift moves reported margin on its own.
Temporary
4. Pricing optimisation
Management names pricing as one of five gross-margin drivers, without quantifying it. Current pricing strength is a scarcity artefact. The last time this industry had one, in 2016–17, price erosion resumed and then accelerated within two quarters of the shortage ending.
Structural, early
5. Datacentre interconnect and telecom transport
Elevated demand for new ZR and ZR+ transceivers, plus sustained telecom transport growth. Dell'Oro forecasts the optical transport equipment market to grow 16% in 2026 and pass $18bn for the first time since 2000, and notes lengthening lead times.
Cyclical
6. Semiconductor capital equipment
The one part of Industrial that management describes as strong. It is a capital-equipment cycle and it behaves like one.
On 2 March 2026 NVIDIA bought 7,788,161 shares at $256.80 for $2.0bn, alongside a multi-year agreement carrying a “multi-billion-dollar purchase commitment”, future access and capacity rights, and access to five additional Coherent product families related to co-packaged optics. The dollar value is not disclosed. Coherent also entered a capacity agreement that “may require incremental investments in equipment, labor, and working capital… through 2030” and which “may result in material future cash requirements and could affect revenue concentration, gross margin, and capital expenditures as volumes ramp”. The equity funded the capital expenditure; the commitment is the reason for it. Whether the arrangement is worth more than it costs cannot be answered from what has been disclosed.
Gross margin rose 233 basis points to 37.5%. Management attributes it to five things: lower input costs, better cycle times, yield improvement in Datacenter & Communications, pricing optimisation, and lower intangible amortisation — none quantified. Two of the five are mechanical rather than operational: amortisation fell from $303M to $280M because acquired assets age, and the Industrial benefit is a subtraction of bad margin rather than an addition of good.
Revenue and GAAP gross margin, FY2016–FY2026. The two step-ups in revenue are acquisitions, not organic growth: Finisar in FY2020 and Coherent, Inc. in FY2023 — and both arrived with purchase-accounting inventory step-ups that drove gross margin down, $87.7M in FY2020 and $158M in FY2023. Margin has recovered to 37.5% but remains below the 39.9% of FY2017 and the 39.1% of FY2021, on more than five times the revenue.
Reported earnings of $805M and $4.12 of diluted EPS contain a large amount that will not repeat. A $124M gain on the sale of businesses appears as a new income-statement line, zero in both prior years. A $74M gain on an unnamed equity investment sits in other income. The effective tax rate was 7%, against 68% in FY2025 and a 21% statutory rate, on discrete items — a $62.8M release of unrecognised tax benefits and a $47.5M German tax-law change. At the statutory rate the charge would have been roughly $178M rather than $61M. Restructuring also fell $97M year on year, worth about 1.4 points of margin, and management expects the programme substantially complete by the end of FY2026, so that year-on-year benefit stops as well. Strip the gain and add back the held-for-sale impairment and operating income is $838M, an 11.8% margin rather than 12.6%.
Operating cash flow against capital expenditure, FY2016–FY2026. In FY2026 the two lines cross for the first time and cross hard: $80M of operating cash flow against $1,103M of capex, for free cash flow of −$1,023M. The gap was funded by NVIDIA's $2.0bn equity purchase, $437M of divestiture proceeds and $89M from an equity sale. For contrast, Lumentum — running the same demand at 42% of Coherent's revenue — generated $751M of operating cash flow and $300M of free cash flow in the same fiscal year.
The mechanism is a working-capital build of extraordinary size. Inventories rose 79.5% to $2,581M against 22.5% revenue growth, with work in progress alone up 89%; receivables rose 39.3%. Days of inventory went from 139 to 212. Against that, payables rose 125% to $1,905M and days payable stretched from 82 to 156 — without that extension operating cash flow would have been sharply negative, and $372M of the payables balance is unpaid capital expenditure rather than trade credit. Behind it sit $11.8bn of purchase commitments to vendors, $3.4bn of them falling due in FY2027 against $1,092M a year earlier. Depreciation actually fell in FY2026 despite a 60% rise in net property, because most of the new capacity is not yet in service. It will step up sharply.
| US$ millions | FY2019 | FY2023 | FY2024 | FY2026 |
|---|---|---|---|---|
| Revenue | 1,362 | 5,160 | 4,708 | 7,118 |
| Gross margin (GAAP) | 38.3% | 31.4% | 30.9% | 37.5% |
| Operating margin (derived) | 10.9% | (0.7)% | 2.0% | 12.6% |
| Operating cash flow | 178 | 634 | 546 | 80 |
| Capital expenditure | 137 | 436 | 347 | 1,103 |
| Total debt | 467 | 4,310 | 4,100 | 3,222 |
Ranked: acquisitions $5,489M — effectively all of it the Coherent, Inc. close on the first day of FY2023; gross debt repayment $4,466M, of which $1,269M was voluntary prepayment; capital expenditure $2,327M, front-loaded into FY2026; dividends $50M, all on preferred stock, with no common dividend ever paid; and share repurchases of exactly zero, with no programme in effect in FY2025 or FY2026.
The pattern is legible. Management bought two large businesses on debt, spent four years paying it down ahead of schedule, refinanced repeatedly to cut the spread — the term loan margin came down from 2.75% to 2.50% in April 2024, to 2.00% in January 2025 and to 1.75% in September 2025 — and then, once the balance sheet was repaired, turned the capital taps on. Total debt fell from $4,310M to $3,222M while interest expense fell from $288M to $190M. The equity side went the other way: the $2.15bn of Series B preferred issued to Bain Capital converted into 30.1 million common shares in December 2025 after the holder waived its dividend rights, and NVIDIA bought 7.8 million more. Shares outstanding rose 25.8% in one year. This is management that has consistently chosen dilution over leverage, and growth over returning cash.
Post-dating the reported figures: a ~$945M unsecured subsidiary facility signed 12 August 2026, undrawn at the filing date; and $100M of one-time performance share units approved 27 August 2026 across five executives, $50M of it to the chief executive, vesting only against 10–25% absolute share-price CAGRs over four years.
At the top on volume; not at the top on margin. Fourth-quarter FY2026 revenue of $2,045M is 49% above the previous peak quarter of $1,370M in December 2022 and 94% above the trough of $1,053M in September 2023. Full-year revenue is 38% above the prior peak year. On gross margin the picture differs: the quarterly GAAP figure of 38.5% is still 250 basis points below the all-time peak of 41.0% set in December 2020, and annual GAAP operating margin of 12.6% is 30 basis points below the FY2021 peak — 11.8% excluding the one-off gain and impairment. Peak volume at below-peak margin is what a business looks like when it is growing into a cost base rather than harvesting one.
| Exposure | FY2026 | FY2025 | Direction |
|---|---|---|---|
| Datacenter & Communications | 74.1% | 64.6% | Growing, +40% |
| Industrial | 25.9% | 35.4% | Shrinking, −10% on disposals |
| North America (customer HQ) | 65.1% | 61.4% | Rising with hyperscaler mix |
| China (revenue) | 11.4% | 11.7% | +19.6% in dollars |
| China (long-lived assets) | $969M | $403M | +140% — asset exposure rising far faster than revenue |
| Largest customer | 20% | 10% | Doubled in one year; a second at 12% |
The cleanest evidence is Lumentum, whose mix made it the pure-play victim of the last cycle. Revenue fell 23.1% from FY2023 to FY2024; GAAP gross margin fell 1,370 basis points to 18.5% and non-GAAP gross margin 1,020 basis points; GAAP operating margin went from −6.5% to −31.9%. Roughly 44 basis points of gross margin was lost for every 1% of revenue decline. Coherent's more diversified mix cushioned it to about 27 basis points per 1%. Apply the milder of those elasticities to a company whose depreciation base is about to step up materially, and the outcome is worse than FY2024, not better. This does not require an AI spending pause. It requires only that the shortage ends — which is what the industry's own tracker expects, and what every competitor's capacity expansion is designed to cause.
The base rate is the other half. Average optical component supplier margins ran 9% in 2016–17, 4% in 2018 and 1% in 2019, and LightCounting attributed the collapse to price erosion that accelerated as volumes fell. There is no episode in this dataset where volumes declined and price held.
Concentration compounding with commitment. One customer is 20% of revenue and a second is 12%. The filing concedes such customers seek price concessions and may impose most-favoured-customer or exclusivity terms that foreclose selling to others. If forecast orders do not arrive, the 10-K names the consequences precisely: excess or obsolete inventory, underutilised capacity, liabilities under supplier arrangements, reimbursement obligations for supplier capital expenditure, and non-cancellable purchase commitments. The exposure is $2,581M of inventory at 212 days and $11.8bn of vendor commitments. One customer deferring a ramp produces a cost-of-goods charge and a balance-sheet write-down at the same time.
$3.1bn of goodwill in the shrinking half, with an 8% cushion. At the 1 April 2026 test the Lasers reporting unit's fair value exceeded carrying value by approximately 8% on an 11.0% discount rate, and the company states the unit “remains sensitive to changes in assumptions and future operating performance”. That unit sits inside Industrial, whose revenue fell 10%. Goodwill and intangibles together are $7.26bn — 39.7% of assets and 64.6% of equity.
An unreserved export-control contingency. In January 2025 Coherent received a Bureau of Industry and Security inquiry about past sales to Huawei. It has stopped shipping to Huawei, is cooperating, and states it cannot determine an estimate or range of loss. There is no update in the FY2026 10-K.
A concentrated maturity. $2,136M of principal matures in FY2030 and $986M in FY2031 — 96% of the total in a two-year window, including $990M of 5.00% notes due December 2029. Most debt floats; 100 basis points is $25M of interest, and the cap that saved $17M in FY2026 has decayed in value from $17M to $3M.
The thesis is falsifiable and dated. The case rests on InP scarcity being structural rather than cyclical, and on Chinese module supply being restricted. As of September 2026 no US rule in force restricts imports of Chinese optical transceiver modules: the Section 301 semiconductor action carries a 0% additional rate until June 2027 and does not name modules, and the reported FCC action remains a draft with no primary document located.
Transceiver-only revenue, unit volumes and average selling prices — the company states it does not disaggregate by product. · The organic versus divested revenue bridge for Industrial, and the revenue of the businesses sold. · The identity of the two customers at 20% and 12%, and whether NVIDIA is one of them; the 10-K does not say and the connection should not be inferred. · The dollar value of the NVIDIA purchase commitment. · The split of capital expenditure between growth and maintenance, and any FY2027 capex plan. · Backlog and book-to-bill, neither of which is disclosed. · Standalone silicon carbide economics, invisible since Materials folded into Industrial, against an implied $4.0bn valuation set in 2023 and $604M of restricted cash inside the subsidiary. · Whether optical transceivers fall inside the Section 232 covered-product annex. · Per-vendor 800G and 1.6T market share, which every independent tracker places behind subscription.
| Indicator | Published in |
|---|---|
| Whether InP is still called the primary constraint, and the stated reason for gross-margin change — yield and mix, or price | Coherent quarterly earnings release and call, roughly November, February, May and August |
| Operating cash flow against net earnings, and inventory days | Coherent 10-Q and 10-K cash flow statement and balance sheet, SEC EDGAR CIK 0000820318 |
| Whether unit growth and revenue growth diverge — the divergence is price | LightCounting Quarterly Market Update |
| Chinese competitors' revenue and gross margin — the share clock and the price tell | Zhongji Innolight (300308.SZ) and Eoptolink (300502.SZ) filings on CNINFO |
| Whether a US import restriction on Chinese optical modules is actually enacted | Federal Register; FCC docket notices |
| Section 301 rate step-up on HTS 8541 and 8542, scheduled 23 June 2027 | Federal Register; USTR Section 301 notices |
| Indium supply, Chinese share, price and export volumes | USGS Mineral Commodity Summaries, indium chapter, each January |
| Broadcom's advance into EMLs, photodiodes and VCSELs | Broadcom OFC product announcements each March |
| Lumentum and Fabrinet results as an independent read on the same demand | Lumentum and Fabrinet quarterly releases, both on a June fiscal year |
What this business really is. A vertically integrated component manufacturer that has become, for now, an AI infrastructure supplier — three-quarters of revenue is Datacenter & Communications, sold outright with no recurring revenue.
The economic engine. The indium phosphide wafer. Owning qualified 6-inch InP capacity is what lets Coherent ship transceivers when the industry cannot, and the 6-inch yield curve is the one genuine cost advantage the evidence supports.
The main growth lever. Hyperscaler capital spending, converted through wafer output. Everything else — mix, pricing, restructuring — is second order or non-repeating.
What breaks the story. The shortage ending. Prices here have never held when volumes fell, Coherent has fixed its selling prices under multi-year agreements, and it has taken capex to 15.5% of sales into a shortage every competitor is spending to eliminate.
What to monitor. The gap between earnings and cash. FY2026 produced $805M of net earnings and $80M of operating cash flow; until that closes, the reported recovery is an accrual, not a cash, event.