Photonics & Fiber Connectivity: Who Gets Paid as AI Networks Expand?
Tenet Research | October 4, 2026 · Deep Dive
Buying more computing power only helps if the machines can exchange information efficiently. The investment opportunity in photonics is to understand which suppliers make that exchange possible, and which can turn rising demand into durable cash flow.
Coverage: LITE, AAOI, MXL, APH, VIAV, FN, CIEN, MRVL, COHR and GLW. Original financial research as of October 4, 2026; market valuations use October 2 closing data. Updated October 5 with AAOI’s ATM completion and a terminology explainer. This is an independent Tenet analysis, not a bank research report or a personalized recommendation.
The words behind optical networking
These terms describe different parts of the same system. Follow a message from the electronic chip to the light signal.
Photonics
The broad field of making, guiding, controlling and detecting light. Lasers, fiber and light-handling chips all belong here.
Think: the technology family.
Optical
Means related to light. An optical link carries information using light; an electrical link uses electrical signals. “Optics” and “photonics” often overlap in industry language.
Think: how the message travels.
Silicon / silicon photonics
Silicon is a semiconductor material. Silicon photonics uses silicon-based chip manufacturing to build light-guiding and light-controlling components. Silicon alone does not tell you whether a chip handles electrical or optical signals.
Think: a material, and one way to build photonic chips.
PIC: photonic integrated circuit
A chip containing several optical functions, such as guiding or modulating light. Silicon photonics is one platform for making a PIC; other material platforms also exist.
Think: the light-handling chip.
CPO: co-packaged optics
The optical engine is placed beside the main electronic chip within the same package. Data travels a shorter electrical path before being converted to light. This can improve bandwidth density and energy efficiency, with tougher packaging and servicing requirements.
Think: where the optical engine sits.
PCB: printed circuit board
The board that supports and electrically connects chips and other components. In a conventional pluggable design, electrical signals travel across the board toward the optical module. CPO brings that conversion point closer to the main chip.
Think: the board underneath the components.
Electronic chip → optical engine converts the data → light travels through fiber → a receiver converts it back to an electrical signal.
A silicon-photonics engine can be used in a pluggable module or in a CPO design. The laser supplying its light commonly uses compound-semiconductor material, such as indium phosphide, and can be external or integrated with the photonic chip. The electronic processors still perform the main computing work in the networking systems discussed here.
TENET RESEARCHtenetresearch.aiHow AI data centers connect
“AI networking” covers several distinct systems. Connections within a tightly coupled compute domain, links across racks and transport between separate sites do not have identical requirements. NVIDIA’s platform materials describe copper inside rack-scale systems alongside optical connectivity beyond them, while Ciena’s coherent-optics material addresses a different range of transport challenges. (NVIDIA; Ciena)
That is why we do not start with the claim that copper is finished. The sensible engineering choice depends on reach, bandwidth density, power, cost and serviceability. Our investment framework is coexistence first, followed by selective displacement as particular architectures justify the extra optical complexity.
The constraint is not a universal distance at which every copper cable stops working. Electrical-channel loss and the signal-conditioning burden depend on the implementation; bringing optics closer can shorten that difficult electrical path. Neither medium is free of power or thermal costs, and a passive cable should not be confused with the active circuitry at its ends. (Broadcom’s electrical-path explanation; NVIDIA’s optical and thermal architecture)
Inside the rack, across the cluster and between sites
Location determines the technology, the deployment cycle and which supplier gets paid.
TENET RESEARCHtenetresearch.aiScale-up links make closely coupled accelerators behave more like a coordinated system. Scale-out networks connect servers and racks into a larger cluster; scale-across or data-center interconnect connects separate facilities. Marvell’s latest call explicitly distinguished the continuing role of pluggables in scale-out from the earlier-stage transition toward optics in scale-up. (MRVL reporting notes)
The practical takeaway is simple: do not use a long-distance coherent-transport product announcement as proof that an in-rack optical architecture is ready, or treat a switch demonstration as evidence that every supplier in the ecosystem is already collecting volume revenue. The customer, topology and deployment milestone must match the thesis.
How information travels as light
An optical link still begins and ends with electronics. Light is generated, modulated to carry information, routed through optical structures and detected at the receiving side, where electronics recover the signal. Integration can combine some of these functions, but it does not eliminate the need to perform them. (NVIDIA architecture discussion; Ciena technical explainer)
How data travels through an optical link
Electronics send the data, light carries it, and a receiver turns it back into an electrical signal.
Electrical data
A driver prepares the signal. A laser supplies the light.
Encode
A modulator changes the light to carry information.
Transport
Waveguides route light on-chip. Fiber carries it between endpoints.
Recover
A receiver detects the light; electronics recover the data.
Silicon photonics is a technology for integrating optical functions. CPO describes where the optics are packaged. A silicon-photonics device can also be used inside a pluggable module.
TENET RESEARCHtenetresearch.aiA laser supplies the optical carrier; a modulator changes a property of that light to encode data. The familiar on/off explanation is only an introduction: modern implementations can use multiple signal levels, while coherent systems exploit amplitude, phase and polarization with substantial digital signal processing. A waveguide is an optical path on a chip; fiber carries light between endpoints, and wavelength multiplexing can place multiple optical channels on one fiber. (NVIDIA silicon photonics; Ciena coherent optics)
The speed labels also need care. A module’s aggregate data rate, the rate of each electrical or optical lane and the capacity of a single wavelength are not the same measurement. A multi-lane transceiver and a coherent modem can share a headline rate while using very different architectures. (NVIDIA lane and engine specifications; Ciena wavelength-capacity explanation)
For investors, the point is not to memorize every acronym. It is to identify the constrained function and ask who supplies it. Faster links can increase the value of a laser, driver, receiver, connector, manufacturing process or test system, but the amount of value captured depends on the actual design.
Why optics are moving closer to the chip
A conventional pluggable optical module converts the signal near the equipment’s front panel. Co-packaged optics brings the optical engine onto a common package with the switch, shortening the high-speed electrical path. Near-packaged optics is an intermediate arrangement that brings optical conversion closer without using the same degree of package integration. (Broadcom CPO architecture; LITE reporting notes; FN reporting notes)
CPO moves the conversion point closer to the switch.
The engineering trade-off: a shorter electrical path versus more demanding integration and servicing.
ASIC
module
The conversion happens at a module near the faceplate. The module is independently replaceable.
ASIC
engine
The electrical hop is shortened. An external laser can remain replaceable even when the optical engine is more deeply integrated.
TENET RESEARCHtenetresearch.aiThe attraction is lower electrical loss and a route to greater bandwidth density, with potentially better power efficiency. Those are reasons to take CPO seriously, but product-specific vendor comparisons are not universal savings estimates for every network. Broadcom and NVIDIA describe different implementations and performance claims; investors should not combine their numbers into one industry-wide promise. (Broadcom; NVIDIA)
Serviceability is more nuanced than “a laser fails, so the whole switch is scrap.” The OIF’s ELSFP approach supports a replaceable external laser source, and NVIDIA describes modular features in its optical implementation. That does not make every integrated optical-engine fault easy to repair, but it shows why laser replacement, engine replacement and whole-system reliability must be evaluated separately. (OIF ELSFP; NVIDIA implementation)
We would assess a new architecture on total economics: acquisition cost, energy, usable bandwidth, manufacturing yield, qualification burden and expected downtime. The technically most integrated solution does not automatically become the fastest commercial winner. A less integrated design can remain attractive if it is easier to deploy and maintain.
Where the economic value can accumulate
Our preferred map follows the customer’s spending decision rather than a rigid stack of “best” companies. A customer may buy a finished system, source components directly or use a manufacturing partner. The same supplier can therefore appear in more than one part of the chain, and the reported revenue pools overlap.
Who supplies each part of the network
LITE · COHR · AAOI
Sell the devices that turn electronic traffic into optical links.
Watch yields, laser supply, customer qualification and price per module.
MRVL · MXL
Provide signal processing, drivers and related connectivity chips.
Watch architectural content shifts, design wins, R&D and customer mix.
GLW · APH
Supply fiber, connectors, cables and broader interconnect systems.
Watch optical content, capacity returns and non-AI business exposure.
FN
Build complex optical products for other companies.
Watch customer concentration, program ramps and factory cash needs.
VIAV
Help qualify, manufacture and monitor high-speed networks.
Watch test intensity, production demand and acquisition-adjusted growth.
CIEN
Connect data centers and networks using coherent optics and systems.
Watch cloud demand, backlog conversion and supply commitments.
TENET RESEARCHtenetresearch.aiQualified light sources: the latest Lumentum and Coherent calls describe tight demand and capacity expansion around lasers and indium-phosphide manufacturing. The analytical attraction is scarcity combined with qualification requirements; the risk is that successful industry expansion eventually normalizes that scarcity. (LITE reporting notes; COHR reporting notes)
Signal-processing and analog content: Marvell and MaxLinear participate in the electronics that make optical links usable. Their calls also show that architectures can change the required mix of DSPs, drivers, TIAs and other devices. The investment task is to track content per link, not assume a rising port count guarantees every component the same growth. (MRVL reporting notes; MXL reporting notes)
Assembly, fiber and validation: Fabrinet, Corning, Amphenol and Viavi participate through manufacturing, physical connectivity or testing rather than a single common product. Their latest reporting supports the breadth of that opportunity, but their margin structures and capital needs differ materially. (FN reporting notes; GLW reporting notes; APH reporting notes; VIAV reporting notes)
We define a bottleneck investment narrowly: a supplier must be difficult to replace, able to expand economically and able to retain enough pricing or content to earn an attractive return. A shortage alone is not a moat. A business can be essential to a system yet earn disappointing shareholder returns if competition, capital intensity or customer bargaining power absorbs the benefit.
How large could the optical networking market become?
Goldman Sachs’ publicly accessible May 2026 overview describes a potential AI-networking addressable-market expansion to $154 billion across configurations including copper cables, pluggable optical modules, CPO and PCB midplanes. The public summary does not disclose a forecast year or enough assumptions to reproduce the model, so we do not present that figure as verified optical-industry revenue by a particular date. (Goldman Sachs Research overview)
The sensible inference is that architectures with more networking content could open substantial new spending pools. The wrong inference is that every dollar belongs to photonics, arrives on schedule or becomes revenue for the companies on this page. Adding laser, module and complete-system markets can also double-count the same value as it moves through the supply chain.
Our framework starts with deployed systems, required connections and realizable content per connection. Then it applies attainable market share, selling-price pressure, manufacturing yield and the cash required to scale. Finally, it asks how much of the resulting value belongs to each diluted share. That is a more useful bridge from technology to investing than a large TAM divided by today’s market capitalization.
The ten-company deep dive
The following assessments are Tenet’s analytical judgments, not formal ratings or price targets. The company cards use consolidated reported revenue, GAAP operating margin and trailing free cash flow, while the reporting notes preserve the distinction between reported results and management expectations.
Lumentum (LITE)
Lumentum sells both critical light-generating components and more complete optical systems. That gives it exposure to existing transceiver demand and to the possibility that future architectures require different laser sources, rather than forcing investors to depend on one packaging format. (LITE reporting notes)
2026 fiscal Q4 · period ended 2026-06-27 · company-wide results, not an optics-only segment.
What the evidence says: The latest call described growing laser demand, record 800G shipments, initial 1.6T shipments and an optical-circuit-switching ramp. Management also separated current scale-out shipments from a later scale-up opportunity, with timing dependent on customer deployments. (LITE reporting notes)
The investment case: The attractive part of the case is the component position. If qualified laser output remains difficult to expand, additional capacity can support both revenue and pricing. Selling lasers into multiple architectures may be more resilient than relying on a single module configuration. The question is how long that scarcity lasts after new capacity arrives.
What can go wrong: Vertical integration does not remove execution risk. Lumentum still needs substrate supply, manufacturing yield and timely customer qualifications. Its recent GAAP net loss also contains a major noncash convertible-debt charge, so a negative headline earnings figure is not a clean measure of operating performance. Equitization can improve debt exposure while still changing shareholders’ economic ownership. (LITE reporting notes)
Our read: Treat LITE as a laser-and-systems execution story with significant expectations already attached. Monitor incremental cash generation, the economics of expanded capacity and the gap between current shipments and later scale-up revenue.
Coherent (COHR)
Coherent participates across a wider optical product set rather than selling one isolated component. Its manufacturing footprint and internal indium-phosphide capability give it several potential ways to supply next-generation links, while its broader business mix means it is not a pure AI-transceiver company. (COHR reporting notes)
2026 fiscal Q4 · period ended 2026-06-30 · company-wide results, not an optics-only segment.
What the evidence says: Management described an ongoing 800G business, a rapid 1.6T ramp and planned CPO-related contributions. It also described expanding six-inch indium-phosphide production for EMLs, continuous-wave lasers and photodiodes, with additional capacity planned beyond the current ramp. (COHR reporting notes)
The investment case: The investment case is that internal component capability and manufacturing scale can matter more as customers need larger volumes of difficult-to-produce devices. If yields improve while premium products become a larger share of sales, the same factory investment can support revenue growth and better unit economics.
What can go wrong: The cash-flow comparison is the counterweight. Heavy capacity spending has consumed more than operating cash generation over the measured trailing year. That does not prove the investments are poor, but it raises the burden of proof: customer commitments must become shipments, shipments must generate margins, and those margins eventually need to recover the cash invested. (COHR reporting notes)
Our read: Separate the technology advantage from the return on the factory buildout. Strong demand commentary is encouraging; sustained free cash flow after the expansion is the more demanding test.
Applied Optoelectronics (AAOI)
AAOI is a direct module-manufacturing exposure, but its revenue is not entirely advanced AI optics. Data-center products coexist with a substantial cable-television equipment business. The distinction matters because changes in either business can affect consolidated growth and margins. (AAOI reporting notes)
2026 fiscal Q2 · period ended 2026-06-30 · company-wide results, not an optics-only segment.
What the evidence says: The latest reported quarter included 800G deliveries, while management expected initial 1.6T shipments after qualification. Management disclosed very concentrated customers and described financing future manufacturing investment with a combination of cash, operating cash flow, equity sales and additional debt. (AAOI reporting notes)
The investment case: If qualifications convert into large production programs and manufacturing yields improve, AAOI can experience substantial operating leverage from a smaller starting base. A rising mix of higher-speed products can create a powerful revenue ramp without requiring it to dominate the entire optical market.
What can go wrong: Equipment spending, inventory and prepayments require funding before the finished modules generate cash. A delayed qualification or a customer order change can hurt utilization while those cash commitments remain. The completed equity raise strengthens the funding position, but its issued shares still dilute existing ownership. Further financing remains possible if investment needs outpace available cash. (AAOI reporting notes)
Our read: With the announced ATM now completed, follow how effectively AAOI deploys the proceeds: deliveries, manufacturing yields, margins and cash conversion. The new funding gives the expansion more financial support; returns per share will depend on execution and the enlarged share count.
Marvell Technology (MRVL)
Marvell provides the electronic intelligence around optical connections, including optical DSPs, analog components and switching. It also participates in custom computing, so neither total company revenue nor its data-center segment should be labeled optical revenue. (MRVL reporting notes)
2027 fiscal Q2 · period ended 2026-08-01 · company-wide results, not an optics-only segment.
What the evidence says: Management described strong 800G demand and a rapidly ramping 1.6T business. It expected pluggables to remain central to scale-out, while customers evaluated near-packaged and co-packaged approaches for scale-up. Those architectures represent a range of possible implementations, not one locked-in industry design. (MRVL reporting notes)
The investment case: The breadth of Marvell’s connectivity portfolio can help it participate even when the physical layout changes. A customer moving the optical engine closer to the processor still needs an electrically and optically functional link. The relevant question is which Marvell content remains, which changes and which expands.
What can go wrong: Architectural transitions can redistribute value. Removing or simplifying a discrete DSP in a particular link is not irrelevant simply because total bandwidth rises. Custom-compute mix can also change consolidated margins, and stock-based compensation affects the conversion of business growth into per-share ownership. A strong company can still be a demanding investment at a high multiple of trailing sales. (MRVL reporting notes)
Our read: Track connectivity growth separately from custom compute, then assess the combined cash-generation and dilution picture. Do not assume every dollar of AI-related revenue carries the same margin or valuation.
MaxLinear (MXL)
MaxLinear supplies high-speed signal-processing and analog technology alongside broadband and connectivity products. In optics, its opportunity is to win silicon content in modules and related architectures rather than to manufacture the entire fiber network. (MXL reporting notes)
2026 fiscal Q2 · period ended 2026-06-30 · company-wide results, not an optics-only segment.
What the evidence says: Management described Keystone moving through high-volume 400G and 800G programs. It placed Rushmore, Annapurna and newer TIA products at sampling or qualification stages, with initial revenue expected later. Its optical revenue outlook is therefore useful context, but it should not be confused with reported results. (MXL reporting notes)
The investment case: A smaller revenue base means a successful hyperscaler program can change the financial profile quickly. Additional products may increase the amount of content MaxLinear can offer, and participation in both optical and electrical connectivity reduces the need to frame the thesis as a simple copper-versus-fiber bet.
What can go wrong: The latest consolidated operating result remains much less mature than the headline growth rate might suggest. R&D, stock-based compensation and supply commitments matter. A design win can require substantial engineering work before generating material revenue, and a product-generation transition can delay the point when those expenses produce cash. (MXL reporting notes)
Our read: MXL is a qualification-and-execution case, not a smaller interchangeable version of MRVL. The evidence to watch is production revenue from named product families, cash conversion and whether per-share economics improve alongside sales.
Corning (GLW)
Corning supplies the physical optical infrastructure that carries and connects light. It also has significant businesses outside optical networking, so its consolidated results reflect more than the AI infrastructure cycle. (GLW reporting notes)
2026 fiscal Q2 · period ended 2026-06-30 · company-wide results, not an optics-only segment.
What the evidence says: The latest call described strong Optical Communications and enterprise demand. Importantly, management said the current enterprise growth was scale-out, while the newer scale-up and photonics opportunities had not yet entered those results. Corning also distinguishes its core and segment reporting from consolidated GAAP reporting. (GLW reporting notes)
The investment case: Fiber density, connection complexity and deployment requirements can raise the value of engineered connectivity beyond the price of raw glass. A supplier that helps customers fit more connections into constrained space can benefit even without owning the switch ASIC or the optical DSP.
What can go wrong: Large future optical-content opportunities still require qualification, adoption and acceptable returns on capacity. Other end markets can offset optical strength, while customer-supported expansion does not eliminate demand risk. A company-wide sales multiple should not be applied mentally to the optical segment as though the rest of the business did not exist. (GLW reporting notes)
Our read: GLW provides established fiber-and-connectivity exposure with future optical scale-up optionality. Keep existing segment performance separate from management’s longer-term aspirations and use consistent GAAP or core definitions when comparing margins.
Amphenol (APH)
Amphenol sells the connections that make equipment work across many end markets. Within AI infrastructure, that includes high-speed copper, fiber-optic and power-interconnect solutions. It is therefore positioned differently from a company whose investment case requires copper to disappear. (APH reporting notes)
2026 fiscal Q2 · period ended 2026-06-30 · company-wide results, not an optics-only segment.
What the evidence says: Management described strong IT-datacom demand and a mix of optical, copper and power opportunities. It did not provide a precise optical revenue split. Reported growth also differs from organic growth because acquisitions contribute to the business. (APH reporting notes)
The investment case: The appeal is architectural breadth. Customers can change the balance between copper and optical links while still needing connectors, cable assemblies and power connections. The established operating-margin and cash-generation profile gives investors a different starting point from a manufacturer still funding its first major ramp.
What can go wrong: Diversification is not immunity. AI-related expectations can still become excessive, and acquisitions create integration and financing obligations. A lower sales multiple than a laser specialist is not proof of undervaluation because the growth mix, margins, capital needs and optical exposure are different. (APH reporting notes)
Our read: APH is a broad connectivity compounder candidate rather than a pure photonics bet. Watch organic growth, acquisition returns and capital allocation alongside the AI narrative.
Fabrinet (FN)
Fabrinet manufactures sophisticated products for other companies. Its value proposition is process execution, optical assembly and the ability to bring complex customer designs into reliable production. That is a different business model from owning the core semiconductor or laser intellectual property. (FN reporting notes)
2026 fiscal Q4 · period ended 2026-06-26 · company-wide results, not an optics-only segment.
What the evidence says: Management described growth across transceivers, data-center interconnect and high-performance-computing programs. Fabrinet was building CPO devices but not yet at full-scale volumes. Its disclosed large customers also illustrate how a diversified program pipeline can still coexist with material customer concentration. (FN reporting notes)
The investment case: A technology transition can increase assembly difficulty, making a proven manufacturing partner more valuable. Fabrinet does not need to own every winning architecture if multiple customers use its capabilities to build their products. Reliable yield and program execution can be commercially important even without a monopoly on the design.
What can go wrong: Manufacturing revenue carries different margins from chip IP or scarce lasers. Capacity construction also consumes cash before the new lines are fully productive. Recent trailing free cash flow is small despite healthy accounting profitability, and customers may change program timing or manufacturing allocation. (FN reporting notes)
Our read: FN is a way to study the production layer rather than make a single-component bet. The key indicators are customer program diversity, factory utilization, capital returns and the recovery of cash flow after expansion.
Viavi Solutions (VIAV)
Viavi helps equipment makers and operators determine whether high-speed networks meet their performance requirements. It participates in laboratory development, production testing and field operations, with other businesses outside the AI data-center ecosystem. (VIAV reporting notes)
2026 fiscal Q4 · period ended 2026-06-27 · company-wide results, not an optics-only segment.
What the evidence says: Management linked demand to the data-center ecosystem and the transition toward higher-speed products. It also discussed CPO testing and the contribution of acquired product lines. The data-center exposure cited on the call refers to a portion of NSE, not the same percentage of consolidated company revenue. (VIAV reporting notes)
The investment case: Greater optical complexity can increase the importance of testing before an expensive assembly is completed. The attraction is that validation is needed across competing architectures, including those that ultimately replace today’s favored design. A supplier can benefit from technical change without manufacturing every module.
What can go wrong: Test intensity is not a guarantee of recurring revenue growth. Customers can improve test efficiency, delay factory spending or reuse equipment. Reported growth needs to be separated from acquired growth, and capital-structure changes can dilute shareholders even while reducing financing pressure elsewhere. (VIAV reporting notes)
Our read: VIAV is a test-demand and operating-leverage story, not an automatic toll on every optical bit. Watch organic lab-and-production demand, the mix of newer products and free cash flow per share.
Ciena (CIEN)
Ciena connects networks and data centers over optical transport infrastructure. Its coherent technology and line systems address a different part of the problem from the short electrical hop inside a compute rack. (CIEN reporting notes)
2026 fiscal Q3 · period ended 2026-08-01 · company-wide results, not an optics-only segment.
What the evidence says: Management reported strong optical-networking and interconnect demand, including growth in coherent pluggables. It distinguished current product shipments from future Hyper-Rail and Vesta CPX contributions. Supply agreements and capacity support are part of converting that demand into revenue. (CIEN reporting notes)
The investment case: When computing is distributed across sites, the network between those sites becomes part of the system rather than an afterthought. Higher capacity per wavelength and better use of installed fiber can have economic value. Ciena’s existing transport business means the case does not require immediate universal CPO adoption.
What can go wrong: Backlog is not the same as collected cash. Supplier commitments and customer timing can create cash-flow pressure before revenue arrives. Carrier spending remains relevant, competitors can improve their coherent products, and newer short-reach opportunities still need to progress beyond samples. (CIEN reporting notes)
Our read: CIEN is principally a coherent-networking and deployment-execution case, with additional optionality. Follow backlog conversion, supply prepayments, cloud/customer mix and the distinction between product acceptance and commercial scale.
The financial comparison: growth is only the first filter
A sector can have a strong demand environment and still contain very different investments. The comparable data show why we look at growth alongside operating profitability and cash requirements, rather than interpreting a green stock-price screen as a business-quality ranking.
Revenue growth across the ten companies
Latest reported quarterly revenue growth versus the same fiscal quarter a year earlier.
TENET RESEARCHtenetresearch.ai| Company / period end | Revenue | YoY | Operating margin |
|---|---|---|---|
| LITE2026 Q4 · 2026-06-27 | $1.01B | 109.3% | 27.8% |
| AAOI2026 Q2 · 2026-06-30 | $191.9M | 86.4% | -12.9% |
| MXL2026 Q2 · 2026-06-30 | $168.8M | 55.2% | -2.5% |
| APH2026 Q2 · 2026-06-30 | $8.76B | 55.0% | 29.5% |
| VIAV2026 Q4 · 2026-06-27 | $443.1M | 52.5% | 13.8% |
| FN2026 Q4 · 2026-06-26 | $1.32B | 44.6% | 10.2% |
| CIEN2026 Q3 · 2026-08-01 | $1.67B | 37.0% | 18.0% |
| MRVL2027 Q2 · 2026-08-01 | $2.74B | 36.5% | 16.8% |
| COHR2026 Q4 · 2026-06-30 | $2.05B | 33.7% | 12.4% |
| GLW2026 Q2 · 2026-06-30 | $4.50B | 16.6% | 15.5% |
Read the periods before comparing the numbers. A fiscal quarter label is not a universal calendar date, and these companies do not all report on the same day. The table deliberately uses GAAP operating margins throughout; it does not quietly mix adjusted margins from earnings presentations with reported margins from financial statements.
Operating margin is useful here because net income can be distorted by financing transactions, investment remeasurement or unusual tax items. It is still not cash flow and it is not a replacement for net income. Lumentum’s convertible-debt charge and Fabrinet’s investment remeasurement and tax items are reminders to inspect the reconciliation before drawing conclusions from a headline earnings multiple. (LITE reporting notes; FN reporting notes)
Free cash flow across the ten companies
Trailing-twelve-month free cash flow as a percentage of consolidated revenue.
TENET RESEARCHtenetresearch.aiThe trailing cash picture is intentionally not presented as a “best stocks” leaderboard. Low free cash flow may reflect attractive expansion investment, but that explanation has to be earned through future returns. Conversely, positive free cash flow can benefit from working-capital timing, customer funding or deferred spending and should not automatically be extrapolated.
Stock-based compensation also deserves its own check. It is added back in operating cash flow under the indirect method, but it can still transfer ownership from shareholders. Rather than treating it as a literal cash expense twice, we monitor the expense, the share count and the cash spent offsetting dilution. A company can grow total free cash flow while creating much less value per share.
Valuation: a strong theme can still be an expensive trade
| Company | Market cap | Price / sales | TTM FCF | TTM SBC / sales |
|---|---|---|---|---|
| LITE | $84.45B | 28.0× | $300.1M | 5.6% |
| AAOI | $9.28B | 15.6× | −$600.8M | 2.6% |
| MXL | $9.61B | 16.9× | $2.4M | 15.6% |
| APH | $214.44B | 7.4× | $4.71B | 0.5% |
| VIAV | $11.62B | 7.7× | $82.8M | 3.6% |
| FN | $16.61B | 3.6× | $4.2M | 0.7% |
| CIEN | $55.40B | 9.2× | $814.4M | 3.5% |
| MRVL | $238.46B | 25.2× | $1.73B | 8.8% |
| COHR | $65.94B | 9.3× | −$1.02B | 2.6% |
| GLW | $141.41B | 8.3× | $2.74B | 2.5% |
Price-to-sales is a diagnostic, not a fair-value model. It does not account for debt, cash, margins or future dilution, and it uses total company sales rather than an imagined optical-only revenue base. A contract manufacturer can legitimately trade at a different sales multiple from a semiconductor-IP business; the lower number is not automatically the cheaper investment.
We do not use negative P/E ratios as meaningful valuation comparisons. We also do not provide enterprise-value multiples from incomplete debt fields or substitute management’s optimistic future run rate for reported trailing revenue without saying so. The objective is to expose the assumptions embedded in a stock, not manufacture a precise-looking ranking.
A useful pressure test is to ask what happens if revenue keeps growing but the valuation multiple falls. If the investment only works with flawless execution and an unchanged premium multiple, the margin of safety may be smaller than the industry narrative suggests. Growth can offset multiple compression, but it does not promise to do so on an investor’s preferred timetable.
The next stages of adoption
There is no single industry switch-over date. Lumentum and Coherent described current high-speed module activity, MaxLinear distinguished shipping products from future product ramps, Fabrinet described early CPO production rather than full-scale volumes, and Ciena separated coherent shipments from newer CPX opportunities. Those are overlapping adoption paths. (LITE reporting notes; COHR reporting notes; MXL reporting notes; FN reporting notes; CIEN reporting notes)
- Near-term evidence: shipped products, recognized revenue, qualified capacity and repeat orders. Our focus is whether higher-speed demand improves operating earnings and cash conversion, not merely reported backlog.
- Next-stage evidence: customer acceptance of newer modules, near-packaged designs and deeper optical integration. We look for deployment milestones and a clear supplier bill of materials rather than assigning an entire future market to today’s winners.
- Longer-term evidence: broader optical scale-up, more integrated packaging and expansion into additional workloads. We treat these as conditional opportunities until adoption, manufacturing economics and customer commitments become visible.
Announcement language requires scrutiny too. Broadcom’s original Davisson release used “now shipping” while its availability section described sampling to early-access customers and partners. That is evidence of product progress, but not a disclosed industry-wide volume ramp. (Broadcom announcement)
September product announcements
Coherent launched PhotonLink at ECOC on September 21, 2026, bringing together components, optical assemblies, manufacturing and testing for CPO, NPO and emerging chip-to-chip connectivity. The announcement supports the breadth of its product strategy, but a platform launch does not itself establish the size of the resulting commercial business. (Coherent PhotonLink announcement)
Lumentum’s September 21 announcement described a DWDM external-laser-module demonstration and expected initial availability in the first half of 2027. The underlying laser platform was already in production; the new module’s availability was still forward-looking. Those two statements should not be collapsed into “the entire new product is shipping today.” (Lumentum ELSFP announcement)
A separate Lumentum, Qualcomm and Corning announcement described a proof-of-concept optical die-to-die showcase combining interface IP, VCSEL technology and multimode-fiber connectivity. This is useful evidence that multiple technical approaches are being explored, not proof that one universal architecture has won or that volume revenue has already arrived. (Joint optical die-to-die announcement)
Our takeaway is constructive but selective. The opportunity is expanding beyond one laser type or packaging choice, which can help suppliers with broad capabilities. It also makes simplistic rules such as “all optical links require the same scarce component” less reliable.
Bull, base and bear cases
Bull case: demand and economics advance together
High-speed deployments broaden, qualified supply expands without destroying pricing and optical integration opens additional connections. Cash generation catches up with capital spending. In this scenario, the strongest outcomes should come from suppliers that combine technical relevance with improving returns on capital, not simply the fastest announced capacity additions.
Base case: several architectures coexist
Pluggables keep shipping, copper remains important at short reach and newer packaging ramps selectively. Revenue growth is uneven across suppliers and customers. Stock selection becomes a question of execution, valuation and per-share cash generation rather than one sector-wide adoption date.
Bear case: the industry grows, but shareholders disappoint
Customer deployments slip after suppliers commit capital, qualification takes longer than expected and pricing falls as capacity catches up. Revenue may still rise, but weaker cash conversion and dilution reduce shareholder participation. Premium multiples compress before the longer-term opportunity has time to arrive.
These are qualitative Tenet scenarios, not probability-weighted forecasts or price targets. The most important difference between them is whether industry spending becomes durable, cash-generating revenue at the supplier level.
What could break the thesis?
- Customer concentration: one deployment change can overwhelm an otherwise positive industry trend. Diversifying across suppliers does not necessarily diversify exposure to the same hyperscaler.
- Capacity before demand: fabs, equipment, inventory and prepayments require cash before the customer pays. Expansion backed by optimistic forecasts is different from expansion supported by enforceable commitments.
- Qualification and yield: an impressive demonstration is not proof of economical high-volume production. Small yield problems can become expensive when integrated assemblies contain valuable components.
- Architecture and content shifts: a new design may remove one supplier’s component while creating another supplier’s opportunity. More bandwidth does not imply more revenue for every incumbent.
- Pricing and competition: shortages can support price increases, but customers have reasons to qualify alternatives. Capacity additions that solve the industry’s bottleneck can also weaken the bottleneck supplier’s economics.
- Power, construction and deployment timing: networking orders cannot be evaluated independently of the systems and facilities that will use them. A ready component is not the same as a ready customer installation.
- Financing and dilution: rising sales can coexist with growing share counts, debt obligations and weak cash generation. Investors need to follow ownership and funding, not just revenue.
- Geopolitical and supply-chain exposure: concentrated materials, manufacturing or customers can create interruption risk. We would evaluate issuer-specific disclosures rather than assume one generic tariff or export-control outcome for the whole group.
- Workload change: training, inference and distributed workloads do not create identical network demand. An architecture optimized for one workload may not capture the same spending if the mix changes.
- Valuation: even excellent execution can disappoint investors when the purchase price already assumes more. The industry thesis and the stock-entry decision must remain separate.
This is our risk checklist, not a claim that each risk is currently worsening. The company-specific examples in the dossiers and reporting notes show where the evidence is strongest and where uncertainty remains.
How we would follow the group
For established breadth, we would study APH and GLW as different ways to participate without relying entirely on one optical-module design. For direct optical execution, LITE and COHR offer broad product and manufacturing cases, while AAOI adds a more concentrated and funding-sensitive ramp. MRVL and MXL require close attention to silicon content and product qualifications; FN, VIAV and CIEN provide manufacturing, test and transport angles respectively.
That is a research map, not a model portfolio. We would not call one name the “safest” solely because it is diversified, or another the “best upside” solely because it is smaller. Positioning should reflect valuation, business uncertainty and how much common customer exposure already exists elsewhere in a portfolio.
For traders, a long-duration industry thesis does not invalidate a failed short-term setup. Earnings dates, customer news and crowded positioning can change the stock’s risk before the business thesis changes. For investors, a price decline is not automatically an opportunity unless the evidence still supports future cash generation at the new price.
A short glossary
- Photonics
- Technology that generates, guides, modulates or detects light.
- Transceiver
- A device with both a transmitter and a receiver.
- DSP / SerDes
- Digital signal processing / serializer-deserializer circuitry used to process and move high-speed data.
- EML / CW laser
- An electro-absorption modulated laser / a continuous-wave light source used with a separate modulation function.
- InP
- Indium phosphide, a compound semiconductor used in many optical sources and receivers.
- Silicon photonics
- Integration of optical functions using a silicon-based photonic platform; a light source is still required.
- CPO / NPO
- Co-packaged / near-packaged optics: different positions and integration levels for the optical engine.
- Coherent optics
- Transmission and detection using additional optical dimensions, supported by a receiver laser and DSP.
- WDM
- Wavelength-division multiplexing: multiple optical wavelength channels on a fiber.
- OCS
- Optical circuit switching: routing optical connections rather than requiring the same electrical conversion at every switching point.
- Design win
- Selection for a customer program; not a guarantee of production volume or revenue.
- FCF
- Free cash flow. In this article, operating cash flow less PP&E purchases, with limitations described below.
Technical definitions draw on the NVIDIA architecture explanation, Ciena technical guide, OIF specification overview and the issuer discussions below. The glossary simplifies the terminology rather than specifying every implementation.
Sources, definitions and reporting notes
This is a dated editorial analysis, not a live valuation dashboard. Market capitalization is the October 2, 2026 closing snapshot. Standardized financial statements and selected company earnings-call evidence were retrieved on October 4, 2026. Each company’s latest reported fiscal period and actual period-end date are shown above.
Quarterly revenue growth compares the latest quarter with the corresponding fiscal quarter a year earlier. Operating margin is GAAP operating profit divided by consolidated revenue. Trailing revenue, cash flow and SBC sum the latest four reported quarters. FCF adds signed PP&E purchases to operating cash flow. It does not deduct acquisitions or SBC and may exclude equipment prepayments classified elsewhere. Price-to-sales is market capitalization divided by trailing consolidated revenue; it is not enterprise value-to-sales. Missing debt values were not treated as zero.
Acquisitions, restructuring, capital spending and working-capital movements affect comparability. APH and VIAV reported growth includes acquired businesses. Corning’s core/segment measures differ from consolidated GAAP figures; the comparable tables use the latter. Lumentum’s standardized gross-profit field conflicted with its call, so gross margins were excluded rather than blended. Financing and unusual accounting items are discussed qualitatively rather than normalized through an unsupported estimate.
Reporting links lead to the companies’ own websites; the notes below identify the exact call date, speaker and timestamp used. Management statements about adoption, capacity and future revenue remain management expectations, not Tenet guarantees. Source data were obtained through the financial-data research service; calculations and all visuals are original Tenet work.
LITE · 2026 Q4 · period ended 2026-06-27
Company materials: Lumentum. Standardized income statement and cash flow for the five quarters ending 2026-06-27; the latest four are used for trailing measures.
- Lumentum reported that the bulk of cloud transceiver shipments were 800G and that it had begun shipping 1.6T transceivers. Call 2026-08-11; Michael Hurlston, 12m 49s.
- Management attributed the quarter’s large GAAP net loss to a one-time noncash charge from equitizing convertible notes, rather than an equivalent operating cash loss. Call 2026-08-11; Wajid Ali, 16m 6s.
- Management expected an ultra-high-power laser demand ramp in the second half of calendar 2027 ahead of customer scale-up deployments in 2028. Call 2026-08-11; Michael Hurlston, 5m 36s.
AAOI · 2026 Q2 · period ended 2026-06-30
Company materials: Applied Optoelectronics. Standardized income statement and cash flow for the five quarters ending 2026-06-30; the latest four are used for trailing measures.
- In Q2 2026, 56% of revenue came from data-center products and 42% from CATV. Call 2026-08-06; Stefan Murry, 16m 46s.
- Three customers accounted for 42%, 26% and 24% of total revenue in Q2 2026. Call 2026-08-06; Stefan Murry, 20m 33s.
- Management said planned manufacturing investment would be financed through cash, operating cash flow, equity sales and additional debt. Call 2026-08-06; Stefan Murry, 24m 48s.
MXL · 2026 Q2 · period ended 2026-06-30
Company materials: MaxLinear. Standardized income statement and cash flow for the five quarters ending 2026-06-30; the latest four are used for trailing measures.
- Rushmore, Annapurna and the 200G-per-lane TIAs were sampling or in customer qualification, with revenue expected to begin in 2027. Call 2026-07-23; Kishore Seendripu, 17m 55s.
- Management guided calendar-2026 optical data-center revenue to $210 million–$230 million; this was a forecast, not reported full-year revenue. Call 2026-07-23; Kishore Seendripu, 4m 37s.
APH · 2026 Q2 · period ended 2026-06-30
Company materials: Amphenol. Standardized income statement and cash flow for the five quarters ending 2026-06-30; the latest four are used for trailing measures.
- IT datacom represented 43% of quarterly sales and grew 89% reported versus 63% organically. Call 2026-07-29; Adam Norwitt, 19m 5s.
- Management described significant high-speed copper, optical and power-interconnect businesses but did not provide a precise optical revenue breakout. Call 2026-07-29; Adam Norwitt, 30m 19s.
VIAV · 2026 Q4 · period ended 2026-06-27
Company materials: Viavi Solutions. Standardized income statement and cash flow for the five quarters ending 2026-06-27; the latest four are used for trailing measures.
- Management said the data-center business was about half of NSE revenue, not half of all company revenue. Call 2026-08-05; Oleg Khaykin, 17m 58s.
- Viavi reported issuing about 12.78 million shares in a follow-on offering for $575 million of gross proceeds. Call 2026-08-05; Ilan Daskal, 2m 41s.
FN · 2026 Q4 · period ended 2026-06-26
Company materials: Fabrinet. Standardized income statement and cash flow for the five quarters ending 2026-06-26; the latest four are used for trailing measures.
- Fabrinet was building CPO devices for a handful of customers, but not yet at full-scale volumes. Call 2026-08-17; Seamus Grady, 27m 13s.
- Fiscal-2026 revenue concentration was Cisco 20%, NVIDIA 16%, Nokia 11% and Amazon 11%. Call 2026-08-17; Csaba Sverha, 13m 46s.
- Fabrinet described a noncash investment remeasurement gain and a Thailand top-up tax provision in explaining the quarter’s GAAP and non-GAAP results. Call 2026-08-17; Csaba Sverha, 12m 51s.
CIEN · 2026 Q3 · period ended 2026-08-01
Company materials: Ciena. Standardized income statement and cash flow for the five quarters ending 2026-08-01; the latest four are used for trailing measures.
- Management said 800ZR pluggable shipments more than doubled sequentially while Hyper-Rail material revenue was expected during 2027. Call 2026-09-03; Gary Smith, 10m 44s.
- Ciena reported sample orders for Vesta CPX, with revenue expected to begin in 2027 and ramp during 2028. Call 2026-09-03; Gary Smith, 11m 50s.
MRVL · 2027 Q2 · period ended 2026-08-01
Company materials: Marvell Technology. Standardized income statement and cash flow for the five quarters ending 2026-08-01; the latest four are used for trailing measures.
- Management expected pluggable modules to remain the primary form factor for scale-out networks. Call 2026-08-27; Matt Murphy, 8m 23s.
- Data center represented 79% of revenue; that category includes more than optical connectivity. Call 2026-08-27; Dan Durn, 18m 0s.
COHR · 2026 Q4 · period ended 2026-06-30
Company materials: Coherent. Standardized income statement and cash flow for the five quarters ending 2026-06-30; the latest four are used for trailing measures.
- Management expected both 800G growth and a rapid 1.6T ramp, with CPO contributing to revenue growth in fiscal Q2 2027. Call 2026-08-12; Jim Anderson, 5m 47s.
- Management described six-inch InP lines producing EMLs, CW lasers and photodiodes, and planned to more than double internal output capacity again by the end of calendar 2027. Call 2026-08-12; Jim Anderson, 6m 42s.
GLW · 2026 Q2 · period ended 2026-06-30
Company materials: Corning. Standardized income statement and cash flow for the five quarters ending 2026-06-30; the latest four are used for trailing measures.
- Corning said the current enterprise growth was scale-out, with scale-up and its new photonics opportunity not yet contributing to those results. Call 2026-07-28; Wendell Weeks, 4m 55s.
- Optical Communications sales were $2.07 billion, up 32% year over year, on Corning’s segment-reporting basis. Call 2026-07-28; Ed Schlesinger, 27m 39s.
Technical references: NVIDIA co-packaged-optics architecture; NVIDIA Spectrum-X; Broadcom Davisson announcement; Ciena coherent-optics explainer; OIF external-laser form factor. Market-opportunity context: Goldman Sachs’ public May 2026 overview. No third-party images, rankings or chart designs have been reproduced.
Educational research only. Not personalized investment advice, a recommendation to transact or a promise of performance. Tenet portfolio holdings, if any, should be verified in the current portfolio view; this article does not represent a live holdings disclosure. Technology roadmaps and forward-looking statements can change.