In fiscal 2025 Coherent turned over a record $5.81B (up 23%) and lifted non-GAAP EPS 191% to $3.53[12]. Eight months later NVIDIA bought $2B of its stock and committed to buy billions more[21]. By June 2026, after rising from a $76.88 low into the $400s, it was worth about $74B[33].

Coherent is the company formed when II-VI bought the original Coherent, Inc. in 2022 and took its name. It makes the physical light path of an AI datacenter — the indium-phosphide and gallium-arsenide laser chips, the silicon-photonics engines, and the finished 800G and 1.6T pluggable transceivers — plus industrial lasers and compound-semiconductor materials[24]. The open questions are not whether it is a real business — it books a record $5.81B in revenue and leads its market — but whether its vertical integration holds margin against lower-cost Chinese assemblers[17], whether the Anderson turnaround has truly fixed a debt-laden balance sheet[13], and whether a multiple near 157× earnings is earned on a business this cyclical[33]. The evidence cuts both ways. This study lays out both cases; the verdict is yours.

The decisive questions

Each links to the section that lays out the evidence on both sides.

Six years of revenue

Total revenue, US$B, fiscal years ending in late June. FY2023 jumps on the first full year of the II-VI/Coherent merger; FY2024 is a cyclical down year; FY2026E (~$7.0B) is an estimate implied by trailing-twelve-month revenue of $6.60B at 31 March 2026 plus the Q4 guidance midpoint — not a reported figure.

How to read this

Ten sections, each built the same way: a neutral synthesis, a two-sided case-for / case-against ledger, sourced data and charts, and dated facts. Start with the question that interests you, or read in order from the Overview.

Coherent is the product of a half-century of compound-semiconductor expertise plus two big bets — Finisar (2019) and the $7B Coherent, Inc. deal (2022) — that turned a materials house into a full-stack optics company. The merger brought the technology and the name; it also brought ~$4.1B of debt and a sprawling portfolio that CEO Jim Anderson has been pruning since June 2024[1][15].

What it actually does

Coherent designs and manufactures across the optical stack. At the bottom it grows compound-semiconductor crystals — indium phosphide (InP), gallium arsenide (GaAs) and silicon carbide — and fabricates the laser chips (EMLs, continuous-wave lasers, VCSELs) and silicon-photonics engines that generate and detect light[24]. At the top it sells finished 800G and 1.6T optical transceivers for datacenters and telecom, industrial and fiber lasers, and optical components and instruments[4]. That spread — from raw crystal to finished module — is the company's defining characteristic and the heart of its competitive debate.

How it got here

Milestones from Coherent and NVIDIA releases and reputable trade press; see Sources.

Leadership

The turnaround thesis rests heavily on one person. Jim Anderson joined as CEO on 3 June 2024 from Lattice Semiconductor, where as CEO (2018–2024) he led a turnaround widely credited in coverage — sustained double-digit revenue growth with record gross margins — after earlier senior roles at AMD, Intel and Broadcom[3][5]. He succeeded Dr. Vincent “Chuck” Mattera Jr., the long-tenured executive who engineered the II-VI/Coherent combination. CFO Sherri Luther joined from the same Lattice playbook, reinforcing the new operating team[14].

The number that matters: AI optical-transceiver sales are estimated to jump from ~$16.5B in 2025 to ~$26B in 2026, roughly +57%[7]. The growth is real and steep — but it is tightly coupled to GPU shipments, and a single chip bottleneck (or a GPU air-pocket) ripples straight through to transceiver demand[10].

How big, how fast

Estimates vary by what each analyst counts, but the direction is unanimous. LightCounting sees the broad optical-interconnect market growing ~30–35% a year in 2025 and 2026 before moderating to 15–20% later in the decade[6]. Cignal AI put the datacom optical-component market above $16B in 2025 (up 60%+), with 800G the single fastest-growing segment and 1.6T just beginning to ramp[8]. TrendForce's AI-transceiver-only figure — the slice most relevant to Coherent's datacenter business — is the ~$16.5B→~$26B path below[7].

Third-party estimate; AI-transceiver-only scope. Other providers count different segment boundaries — treat as directional, not precise.

The value chain — and where the money sits

An optical module is a small stack: a compound-semiconductor laser chip (an EML or CW laser at the high end, a VCSEL for shorter reach) or a silicon-photonics engine; a DSP that conditions the signal; and the assembly that packages it into a pluggable transceiver. Value concentrates at the ends. Analysts estimate the DSP is roughly 30% of a module's bill of materials ($80–90 at 800G, $150+ at 1.6T) and is a Broadcom/Marvell duopoly, while scarce high-speed EML lasers are controlled by a handful of makers including Coherent and Lumentum[11]. The assembly step, where Chinese firms are strongest, earns thinner margins that depend on the current laser-chip shortage holding[11]. Coherent's strategy is to own as much of both ends as possible.

The architecture question hanging over it

The market's shape in five years is genuinely contested. Today the money is in pluggable transceivers. Two shifts could move it: co-packaged optics (CPO), which puts the optical engine onto the switch package, and linear-drive pluggable optics (LPO), which strips the DSP out of the module. Yole expects large-scale CPO only in 2028–2030[9], and Cignal AI sees “no material impact to pluggable shipments” for three years[8] — so the disruption is real but not imminent. How fast it arrives decides how much of this market stays in the products Coherent sells today.

The business has been re-pointed at one engine. In Q3 FY2026, Datacenter & Communications was about 75% of revenue and grew +40.5% to $1.36B, while the Industrial segment shrank 16% to $444M[13]. The AI mix shift is lifting margins — non-GAAP gross margin reached 39.6% — but it also concentrates the company on a single demand driver.

How the segments changed

Read Coherent's numbers carefully, because the reporting structure moved. Through FY2025 the company reported three legacy segments — Networking ($3.42B), Lasers ($1.44B) and Materials ($0.95B)[12]. Under Anderson it collapsed these into two market-facing segments effective FY2026: Datacenter & Communications and Industrial[13]. The two charts below are not directly comparable — they are snapshots of the old and new lenses — but together they show where the weight has moved.

Left: full-year FY2025 under the old three segments. Right: a single quarter (Q3 FY2026) under the new two. Scopes and periods differ — do not blend them. Source: Coherent results releases.

The vertical-integration bet

Most optical-module makers buy their laser chips and DSPs on the merchant market and assemble them. Coherent instead grows its own indium-phosphide and gallium-arsenide crystals and fabricates the EML/CW lasers, VCSELs, silicon-photonics PICs and photodiodes that go inside its transceivers, outsourcing mainly the DSP[24]. The thesis: when laser chips are scarce — as EMLs and CW lasers have been, with demand running an estimated ~30% above supply[10] — owning that supply is both a cost advantage and a security-of-supply advantage. At its May 2025 Investor Day management framed a long-term model of double-digit revenue growth with expanding gross margin and EPS, built on exactly this integration[14].

The counter-argument is that integration is expensive and that breadth dilutes focus. An independent analysis estimates Coherent's module costs run ~10–20% aboveChina's InnoLight, the lowest-cost scaled assembler[11]— so the integration premium only pays off if the laser-chip shortage and Coherent's technology lead persist. And the legacy of the merger is a portfolio spanning “disparate industries,” which is precisely why management has been selling pieces off[15].

By one estimate Coherent led the transceiver market with ~22% share in 2025, the top five at ~72%[16]. But in the part that matters most — 800G modules to NVIDIA — Chinese assemblers InnoLight and Eoptolink supply ~60%, leaving Coherent and Lumentum to split the rest[17]. Leadership of the whole market and leadership of the AI-volume front are two different things.

Who Coherent competes with

The field splits by layer. In datacom modules, the volume leaders are now Chinese: InnoLight (Zhongji Innolight) is the global #1 (~$3.3B 2024 revenue) and Eoptolink is #3 — seven of the top-ten module makers are Chinese, up from two a decade ago[17]. Among Western peers, Lumentum is the closest pure-play rival and Fabrinet is the dominant contract manufacturer (both a partner and a competitor). At the DSP / silicon-photonics layer sit Broadcom and Marvell, far larger and increasingly integrated. In industrial lasers, Coherent meets IPG Photonics, MKS and TRUMPF, where Chinese makers like Raycus and Maxphotonics are compressing prices[20].

The estimated 800G split to NVIDIA

Directional industry estimate (PhotonCap), not a disclosed figure. Coherent is relatively stronger in multimode/VCSEL and weaker in single-mode, where Chinese assemblers lead[17].

Porter's Five Forces

Click a force to see the rating and the sourced basis.

Positioning

Two axes that actually differentiate this market: vertical integration (assembler → chips-and-materials) and AI-datacenter exposure. Hover a point for the sourced basis.

Datacenter & Communications revenue grew +40.5% year-on-year to $1.36B in Q3 FY2026, the engine of the whole company[13]. NVIDIA's $2Bequity stake and multi-billion-dollar purchase commitment is the most direct external endorsement Coherent has received — and, read the other way, a measure of how concentrated Coherent's fate now is on AI capex[21].

The growth, in one chart

Source: Coherent Q3 FY2026 results[13].

The NVIDIA partnership

On 2 March 2026, NVIDIA and Coherent announced a multi-year, non-exclusive partnership: NVIDIA would invest $2Bto fund Coherent's R&D, capacity and US manufacturing, alongside a multi-billion-dollar purchase commitment and future capacity/access rights, expanding a roughly 20-year supplier relationship[21]. The 8-K records the mechanics: 7,788,161 shares at $256.80, about $2B in cash and roughly 4% dilution[22].

The technology Coherent is bringing

At OFC 2026, Coherent demonstrated multiple 1.6T transceivers spanning a silicon-photonics PIC, a high-power InP CW laser, a 200G InP EML and a 200G GaAs VCSEL, plus pieces of a 3.2T / 12.8T roadmap[23]. The breadth is the point: rather than betting on one optical technology, Coherent is fielding all of them — and supplying the laser chips inside its own modules from a ramping 6-inch indium-phosphide platform[24].

The bear read on the same facts

Skeptics turn each strength around. Vertical integration across silicon photonics, InP and VCSEL can read as generalism — one analysis argues Coherent still sources some lasers from Lumentum (whose epitaxy “appears to be ahead”) and lacks Broadcom's integrated laser-plus-DSP system depth[25]. In volume modules, Chinese assemblers led by InnoLightwere first through NVIDIA's 1.6T qualification and are projected to take ~50–60% of the 1.6T module market[26]. And LPO— which removes the DSP, cutting a 1.6T module's power from 30W+ to ~10W and cost 20–30% — is entering hyperscale qualification, threatening the economics of DSP-bearing modules[26].

The merger left Coherent with about $4.1Bof debt and a portfolio spanning “disparate industries”[15]. Since 2023 the company has monetized a SiC stake, sold two non-core units, and repaid debt from cash flow — cutting net leverage from ~2.3× to ~1.7× and total debt to ~$3.2B[30][13]. Real progress; not yet a clean balance sheet.

The deleveraging path

Post-merger peak is the high cited in coverage; FY2025 and Q3 FY2026 are reported balance-sheet debt. Source: Coherent results and trade coverage[13][30].

The portfolio moves

Sources: Coherent and Semiconductor Today / StockTitan releases[27][28][29].

The Anderson playbook

The bet on the new CEO is a bet on a repeat. At Lattice Semiconductor, Anderson took a slow-growing chip company and, over 2018–2024, delivered sustained double-digit revenue growth with record gross margins — a turnaround built on portfolio focus, margin discipline and a clearer product roadmap[5]. At Coherent he has applied the same template: a two-segment structure, divestitures of non-core units, and an Investor-Day model promising double-digit growth and gross-margin expansion[14]. The early numbers — non-GAAP gross margin up to 39.6%, GAAP earnings turning positive in Q3 FY2026 — are consistent with the script[13].

What the skeptics still see

The balance sheet is better, not fixed. Coherent still carried ~$3.19B of total debt at 31 March 2026 and large merger goodwill; FY2025 free cash flow was thin (~$193M) and nine-month FY2026 operating cash flow was near zero against heavy capex; and the $2B NVIDIA equity raise, while strategic, diluted existing holders[31][34]. A turnaround that depends on a sustained AI up-cycle to keep generating the cash that pays down the debt is more fragile than one funded by a steady base business.

FY2025 revenue hit a record $5.81B (+23%) and non-GAAP EPS rose 191% to $3.53; by Q3 FY2026 GAAP earnings had turned positive and non-GAAP gross margin reached 39.6%[12][13]. Yet FY2025 free cash flow was only ~$193M, and the stock trades at a trailing P/E near 157× — the earnings have to keep compounding for the price to make sense[34][33].

The numbers

FY2024 figures approximate from results coverage; FY2025 and Q3 FY2026 from Coherent results releases. Revenue series: FY2021 $3.11B → FY2022 $3.32B → FY2023 $5.16B → FY2024 $4.71B → FY2025 $5.81B; TTM at 31 Mar 2026 $6.60B[32][12][13].

Cash generation is the soft spot

Profit is recovering faster than cash. In FY2025 operating cash flow was roughly $634M against about $441M of capex — free cash flow of only ~$193M against a multi-billion debt load[34]. The first nine months of FY2026 were tighter still: operating cash flow of only ~$10M against ~$547M of capex, as the company pours money into AI capacity[34]. The bull reading is that this is growth capex that converts to cash later; the bear reading is that a richly-valued company is not yet self-funding its own expansion.

The valuation

COHR re-rated dramatically — from a 52-week low of $76.88 into the $400s, closing around $377 on 5 June 2026 for a market cap near $74B[33]. At a trailing P/E around 157× (forward ~51×), the stock prices in years of continued AI-driven growth[33]. Whether that is a re-rating of a structurally better business or late-cycle optimism is the question the Risks section takes up.

Coherent is the largest pure-play at $5.81B FY2025 revenue, well ahead of Lumentum (~$1.65B), Fabrinet (~$3.4B) and IPG (~$1.0B)[36]. By market value (~$74B) it sits just above Lumentum (~$67B) — but both are NVIDIA investees, and the DSP giants Broadcom (~$1.83T) and Marvell (~$231B) are in a different league entirely[33][39].

Revenue — optics & laser pure-plays (most recent FY, US$B)

Broadcom ($63.9B) and Marvell ($5.77B) are shown in the table, not the bar — Broadcom would flatten the others. Fiscal years differ; currency conversions approximate.

Market capitalization (~5 June 2026, US$B)

Pure-play and adjacent names only; the DSP mega-caps are excluded from the bar and noted in the table below.

Side by side

Revenues are reported fiscal-year figures; market caps are point-in-time (~5 June 2026) and approximate. Sources on the Sources page.

The single biggest risk is concentration. With Datacenter & Communications at ~75% of revenue and Industrial shrinking, one analysis warns Coherent has “no safety net” if AI demand softens[42]. Layered on top: CPO/LPO disruption[44], Chinese module competition[17], and a trailing P/E near 157×[43].

The four live risks

1. Concentration. A business that is ~75% AI-datacenter optics has enormous upside and no shock absorber: if hyperscaler or NVIDIA capex pauses, Industrial — itself down 16% year-on-year — cannot cushion the fall[42].
2. Technology disruption.If optical engines move onto the switch package (CPO) or DSPs are stripped out (LPO), the pluggable market where most revenue sits “could shrink structurally” — and capacity hinges on production-grade 6-inch InP yields the industry “has not yet completed at volume”[44].
3. Chinese competition. InnoLight and Eoptolink already supply ~60%of NVIDIA's 800G volume at lower prices and follow on each speed tier within 12–18 months[17].
4. Valuation.Skeptics argue the market “overpays for the bottleneck right before capacity catches up, margins normalize, or hyperscaler capex pauses”[43].

The offsets

None of this is one-sided. NVIDIA's $2B stake and multi-billion purchase commitment give multi-year demand visibility, and Q3 FY2026 showed accelerating growth, expanding margins and improving profitability[46]. On materials-side China risk specifically, when Beijing restricted gallium and germanium exports in 2023 Coherent said the impact would be “minimal to no impact” given ample inventory and the ability to source outside China[45]. The disruption threats, by most estimates, are years rather than quarters away[44].

SWOT

Applied even-handedly — weaknesses and threats get the same weight as strengths.

Method

Research proceeded by fan-out web search and direct fetching of primary and reputable secondary sources. Every URL cited was opened and read during the run; each claim was transcribed into a structured manifest tagging it with a source tier, a confidence level, and a stance (supporting / critical / neutral). The load-bearing figures here — Coherent's FY2025 revenue, margins and EPS, and its Q3 FY2026 segment results and debt — rest on the company's own results releases and newsroom[12][13][35], with the NVIDIA partnership taken from NVIDIA's newsroom and the related 8-K[21][22]. Market-size, market-share, peer-cap and valuation-multiple figures come from third-party data providers and are labeled as estimates[7][16][33].

Frameworks used

The analysis applies the Pyramid Principle for the answer-first executive summary (leading with the balanced state of the debate, not a verdict), Porter's Five Forces for the wafer-to-module competitive landscape with each force rated against a sourced basis, a 2×2 positioning map of vertical integration versus AI-datacenter exposure, a peer-comparables benchmark across the optics and laser group, and a SWOT applied even-handedly so weaknesses and threats get the same weight as strengths. A formal unit-economics teardown and BCG/Ansoff portfolio grids were deliberately skipped: Coherent does not disclose module-level margins or per-line profitability, and an empty framework filled with guesses is worse than none.

Disclosed vs. estimated

Disclosed, high-confidence figures — FY2025 revenue ($5.81B), gross margins, non-GAAP EPS, the Q3 FY2026 segment split, total debt, and the $2B NVIDIA share issuance — come from Coherent's reported results and filings. Treat as estimates: the AI optical-transceiver market size ($16.5B→$26B, TrendForce), the ~22% transceiver-share and the ~60% Chinese 800G-module share (Global Market Insights, PhotonCap and module-market analysts), the FY2026 full-year revenue (~$7.0B, implied from trailing-twelve-month revenue plus Q4 guidance, not a company number), the ~$4.2B post-merger debt peak, and the Investor-Day long-term margin targets (from the slide deck). Peer revenues mix fiscal-year ends; peer market caps and valuation multiples are point-in-time (~5 June 2026) and change daily.

Neutrality & independence

This is a compilation, not an argument. Every section pairs the case for and against with sourced evidence; the Executive Summary frames open questions rather than selling a verdict, and the Risks section presents the bull and bear views side by side rather than making a buy/sell call. The source base is tagged by stance to keep the compilation balanced rather than advocating (see the Sources page for the achieved mix). Where we interpret, we say so and show the basis. The judgment is left to you.

Cross-checked at build time by an automated link checker. Some primary filings (SEC EDGAR) and a few news sites bot-wall automated fetchers; where that occurred, the equivalent figures here are taken from Coherent's own newsroom / GlobeNewswire releases, NVIDIA's newsroom, and reputable press that were fetched and read. See Methodology & Limits.