A map of photonics stocks across six layers — substrates, lasers, photonic ICs, optical engines, transceivers and test — with $COHR, $LITE, $FN, $AAOI, $POET, $AVGO and who wins at each level.
More Macroplane supply-chain guides · Baskets · Investment theses · Macro trends
Photonics stocks are the companies that move data with light instead of electricity, and in 2026 the demand driver is one thing: AI clusters have outgrown copper. Above roughly a metre, at 1.6 terabits per second, electrical signalling costs too much power and too much signal integrity to keep scaling. Everything optical downstream of that constraint is now a supply chain with real revenue, and it splits into six layers that behave nothing alike as investments.
👉 See it live: the Photonics / CPO basket tracks this supply chain on Macroplane, alongside the Silicon Photonics & Optical Interconnects trend and the InP & Substrates basket underneath it.
Three things landed in the last twelve months that turned photonics from a component category into an architecture.
Co-packaged optics started shipping. Broadcom's Tomahawk 6 "Davisson" reached customers with 102.4 Tbps of switching and a claimed 70% cut in optical interconnect power against pluggable transceivers, built on TSMC's COUPE optical engine platform. NVIDIA began shipping Spectrum-X co-packaged switches to partners in June 2026.
The industry started standardising it. On 13 August 2026 a 19-company coalition formally launched "Open Silicon Photonics for AI Systems" as an official Open Compute Project workstream, publishing a roughly 300-page architectural blueprint and targeting first specifications in Q4 2026. The stated goal is scaling OCP racks from 72 nodes to more than 1,024 with multi-vendor interoperability. Named participants include Celestica, Dell, Flex, Foxconn Interconnect Technology, Global Unichip, Hyve Solutions, Keysight, Lightmatter, Qualcomm and Quanta Cloud Technology.
That second point is what changes the investment case. Until now, co-packaged optics was three vertically integrated programmes at NVIDIA, Broadcom and their foundry partners. A multi-vendor specification creates a merchant market, and merchant markets have suppliers you can buy.
The bottleneck moved to the light source. Once you put the optical engine next to a switch ASIC dissipating a kilowatt, laser efficiency at temperature becomes the limiting factor. That has pulled a set of laser and amplifier specialists into a market they were not previously in.
Every laser and photodetector in the stack starts as an indium phosphide or gallium arsenide wafer. This layer is small, concentrated, and geopolitically exposed given China's gallium and germanium export controls.
| Company | Ticker | Position |
|---|---|---|
| AXT Inc | $AXTI | InP and GaAs substrates, gallium processing |
| IQE plc | $IQEPY | Epitaxial wafers and laser diode epi |
| Soitec | $SOI | Engineered substrates, SOI |
| LandMark Optoelectronics | 3081.TW | Epi wafers and laser chips for optical comms |
| Dowa Holdings | $5714.T | Compound semiconductor materials |
These are cyclical, low-margin businesses with high operating leverage to an optical volume ramp. They move violently on transceiver order cycles.
The highest-value layer per unit, and the one the CPO transition has most changed.
Lumentum ($LITE) and Coherent ($COHR) are the incumbents, supplying distributed feedback (DFB) laser arrays and the full transceiver stack around them. Both have substantial existing data centre revenue, which makes them the lowest-risk way to own the theme.
MACOM ($MTSI) sells the electrical half: transimpedance amplifiers, laser drivers and PIN diodes that sit alongside the optics.
The interesting entrants are smaller. Sivers Semiconductors ($SIVE.ST) builds high-power DFB lasers and optical amplifiers. Aeva Technologies ($AEVA) spent five years building semiconductor optical amplifiers for FMCW lidar and in August 2026 launched an Optical Connectivity segment aimed at AI data centres, with a joint development agreement tied to a top-four hyperscaler. We covered that pivot in detail in the Aeva teardown.
This is where the optical functions get integrated onto a chip. It is also where most of the speculative capital sits, because several of these companies have platforms and no volume revenue.
| Company | Ticker | Platform |
|---|---|---|
| POET Technologies | $POET | Optical interposer |
| Lightwave Logic | $LWLG | Electro-optic polymer modulators |
| Tower Semiconductor | $TSEM | Silicon photonics foundry (OpenLight) |
| GlobalFoundries | $GFS | GF Fotonix silicon photonics platform |
| Marvell | $MRVL | Silicon photonics and optical DSP |
GlobalFoundries deserves a specific mention: its Fotonix platform is what Lightmatter's Passage M1000 photonic superchip is built on, a 4,000+ mm² multi-reticle active photonic interposer claiming 114 Tbps of optical bandwidth. Lightmatter is private, so $GFS and packaging partner Amkor ($AMKR) are the listed proxies.
The system owners. This is the layer capturing the most value and the least available to a pure-play investor, because the leaders are enormous companies where photonics is one line item.
If you want the CPO transition without single-name risk, this layer plus layer 5 is the practical expression.
The volume layer. Today's revenue lives here, in 800G and 1.6T pluggable modules, and the open question is how quickly co-packaged optics erodes it.
| Company | Ticker | Role |
|---|---|---|
| Coherent | $COHR | Transceiver OEM, vertically integrated |
| Lumentum | $LITE | Transceiver OEM |
| Fabrinet | $FN | Contract optical assembly for the OEMs |
| Applied Optoelectronics | $AAOI | Transceivers, data centre and cable |
| Zhongji Innolight | $300308.SZ | China's largest AI transceiver supplier |
| Eoptolink | $300502.SZ | 800G/1.6T volume supplier |
| Accelink | $002281.SZ | Optical components and modules |
The bear case on this layer is that CPO deletes the module. The bull case is that CPO takes years to reach the majority of ports, pluggables keep scaling to 3.2T, and the same companies supply the optical engines inside co-packaged designs anyway. Both are partly true, and the split matters more for $AAOI and the China trio than for $COHR, which supplies components either way.
Ciena ($CIEN) and Nokia ($NOK), including the acquired Infinera optical portfolio, own coherent optical transport between data centres rather than inside them. Arista ($ANET) and Cisco ($CSCO) build the switch systems.
The layer investors most often miss is test. Co-packaged optics is much harder to test than a pluggable module, because you cannot swap out a failed optical engine after the package is sealed. That pushes test and validation upstream into the manufacturing flow. Keysight ($KEYS) is a founding member of the OCP silicon photonics workstream for exactly this reason. Celestica ($CLS) took a hyperscaler CPO switch design and manufacturing award in early 2026, ramping in 2027.
| Layer | Risk profile | Revenue today |
|---|---|---|
| Substrates | Cyclical, small cap | Yes, cyclical |
| Lasers and amplifiers | Mixed: incumbents plus options | Yes for $LITE/$COHR |
| Photonic ICs | Speculative, platform-stage | Mostly no |
| Optical engines | Mega cap, diluted exposure | Yes, large |
| Transceivers and assembly | Volume, CPO-disruption risk | Yes, large |
| Systems and test | Steady, late-cycle | Yes |
The honest summary: layers 4 and 5 have the revenue, layer 3 has the upside and most of the ways to lose money, and layers 1 and 6 are the picks-and-shovels positions that work regardless of which architecture wins.
A concentrated photonics position is a bet on the pace of the copper-to-light transition. A spread across the six layers is a bet that it happens at all. The second bet is much easier to win.
Photonics stocks are publicly listed companies that make components and systems for transmitting and processing data with light: laser and optical amplifier makers, photonic integrated circuit designers, optical transceiver OEMs, compound semiconductor substrate suppliers, and the switch vendors integrating optics into their packages. The largest demand driver in 2026 is AI data centre interconnect.
There is no single answer, because the six layers have different risk profiles. Coherent ($COHR) and Lumentum ($LITE) are the largest pure-ish plays with real data centre revenue. Broadcom ($AVGO) and NVIDIA ($NVDA) capture the most value but photonics is a small part of each. POET ($POET) and Lightwave Logic ($LWLG) are platform-stage and speculative. Spreading across layers is the standard institutional approach.
The demand case is strong: co-packaged optics is shipping in volume from Broadcom and NVIDIA, and the Open Compute Project workstream launched in August 2026 targets first multi-vendor specifications in Q4 2026. The risk is that most listed pure-plays are small caps with platform technology and little revenue, so the sector's returns are concentrated in a few large caps where photonics is diluted by everything else they sell.
CPO moves the optical engine onto the switch package, which eliminates the pluggable module. That is a direct threat to transceiver OEM unit volumes over time. The offset is that the same companies often supply the optical engines and components inside co-packaged designs, and pluggables continue scaling in parallel for years. Companies with component and materials depth, like Coherent, are better hedged than pure module assemblers.
Several of the most-discussed names have no listed equity: Lightmatter (photonic interposers, chairs the OCP workstream), Ayar Labs (optical I/O chiplets), and Celestial AI (photonic fabric). Listed proxies for their supply chains include GlobalFoundries ($GFS) and Amkor ($AMKR).
This is research and education, not financial advice. Several companies named here are pre-revenue small caps. Do your own work.
Photonics stocks are publicly listed companies that make components and systems for transmitting and processing data with light: laser and optical amplifier makers, photonic integrated circuit designers, optical transceiver OEMs, compound semiconductor substrate suppliers, and the switch vendors integrating optics into their packages. The largest demand driver in 2026 is AI data centre interconnect.
There is no single answer, because the six layers have different risk profiles. Coherent ($COHR) and Lumentum ($LITE) are the largest pure-ish plays with real data centre revenue. Broadcom ($AVGO) and NVIDIA ($NVDA) capture the most value but photonics is a small part of each. POET ($POET) and Lightwave Logic ($LWLG) are platform-stage and speculative. Spreading across layers is the standard institutional approach.
The demand case is strong: co-packaged optics is shipping in volume from Broadcom and NVIDIA, and the Open Compute Project workstream launched in August 2026 targets first multi-vendor specifications in Q4 2026. The risk is that most listed pure-plays are small caps with platform technology and little revenue, so the sector's returns are concentrated in a few large caps where photonics is diluted by everything else they sell.
CPO moves the optical engine onto the switch package, which eliminates the pluggable module. That is a direct threat to transceiver OEM unit volumes over time. The offset is that the same companies often supply the optical engines and components inside co-packaged designs, and pluggables continue scaling in parallel for years. Companies with component and materials depth, like Coherent, are better hedged than pure module assemblers.
Several of the most-discussed names have no listed equity: Lightmatter (photonic interposers, chairs the OCP workstream), Ayar Labs (optical I/O chiplets), and Celestial AI (photonic fabric). Listed proxies for their supply chains include GlobalFoundries ($GFS) and Amkor ($AMKR). This is research and education, not financial advice. Several companies named here are pre-revenue small caps. Do your own work.