Top Silicon Photonics Stocks 2026: The Optical Computing Boom

Here's the problem every AI data center has right now.

The GPUs got fast. Really fast. But the wires between them didn't. Copper can push data about a meter before the signal falls apart, and past 1.6 terabits per second it basically gives up. So you've got a $3 million rack of chips waiting around for data like a Ferrari stuck behind a tractor.

The fix is light. Swap the copper for lasers and glass, and you get more bandwidth, longer reach, and (this is the big one) way less power. Nvidia says its new optical switches cut network power 5x.^[1]

That's photonics. And Wall Street has finally noticed: Bank of America called it the best-performing investment theme on Earth this year, with the group up 113% versus 14% for global stocks.^[2] Roundhill launched a dedicated ETF (LYTE) on August 6. Four different issuers now sell 2x leveraged Lumentum products.

So is it late? Kind of yes, kind of no.

Yes, because the easy money in optical transceivers (the plug-in modules hyperscalers buy by the million) has been made. No, because the second act just started: moving the optics inside the chip package. That's called co-packaged optics, or CPO, and Nvidia's first CPO Ethernet switch entered full production this month with shipments starting this fall.

Copper has no roadmap past 448G. Light does.

We're slicing this theme by where the light gets made and used:

• Lasers & Modules – the companies making the actual photons. Highest purity, highest volatility.
• Integrators – chip giants bolting optics onto their switches and accelerators. Lower purity, lower risk.
• Picks & Shovels – whoever builds and feeds the factories.
• Moonshots – tiny companies with big promises.
• Private Bellwethers – the startups to watch; potential IPOs.

These are the pure-plays. They make the lasers, the modulators and the transceivers that physically turn electrons into photons. When AI capex rises, these stocks go vertical. When Jensen Huang says "copper still matters" at a conference, they drop 3% before lunch. That happened in March.

Lumentum NASDAQ: LITE

If you only remember one name from this list, make it Lumentum.

Fiscal Q4 revenue hit $1.01 billion, up 109% in a year. Gross margin crossed 50%, a level management thought needed far more revenue to reach. And then they raised the bar: 2028 earnings power of $40 per share.

Three businesses are stacking on top of each other. First, the EML lasers inside 800G and 1.6T transceivers, where Lumentum is one of a handful of suppliers on the planet. Second, optical circuit switches (OCS): mirrors that route light around a data center with no electronics in the path. OCS shipments doubled last quarter and should top $100 million next quarter. Third, and this is the 2027-2028 story, the ultra-high-power laser chips for CPO. Nvidia named Lumentum as the laser fab in its Spectrum-X Photonics supply chain. Hard to get a bigger endorsement.

Management has said OCS is its hardest manufacturing ramp, and two customers make up most of the volume.

Coherent NYSE: COHR

Coherent is the bigger, broader, less pure version of Lumentum. Fiscal Q4 revenue was $2.05 billion, up 34%, but look inside: data center and communications grew 59% while industrial fell 16%. The AI business is now 79% of the company and dragging the old laser-cutting business behind it.

What matters for the next 12 months: Coherent expects its first CPO revenue in the December quarter, and it's launching a "one-stop shop" integrated optics platform called PhotonLink on September 21. It also has more than 20 US production facilities, which suddenly matters a lot (see the China section below). Management is targeting a $3 billion quarter by the end of fiscal 2027.

The trade-off versus Lumentum: you're paying for a less concentrated bet with lower margins and a slower-growing industrial anchor.

Applied Optoelectronics NASDAQ: AAOI

AOI is the small, scrappy transceiver maker that finally made it. Q2 was its fifth straight record quarter and it returned to non-GAAP profitability. 800G revenue more than doubled sequentially, and Q3 guidance of $255-290 million implies roughly 130% growth.

The number to watch is capacity. AOI is at about 200,000 units per month today and says it'll hit 650,000 800G and 1.6T units per month by year-end. Demand outstrips supply through mid-2027, they say.

This is the highest-beta name on the list. It dropped 14% on one August afternoon for no company-specific reason. If you buy it, size accordingly.

The chip giants sell switches and accelerators that happen to have photons inside. You get the theme with a seatbelt.

Nvidia NASDAQ: NVDA

Nobody buys Nvidia for photonics. But Nvidia is the reason photonics is a theme at all. Its Quantum-X Photonics InfiniBand switch is already in production at Lambda, and the Spectrum-X Ethernet Photonics switch, with 4x fewer lasers and 5x lower power than pluggables, entered full-scale manufacturing this month to feed the Vera Rubin platform shipping this fall. CoreWeave, Microsoft, Oracle, Meta and Nebius are early adopters.

Think of Nvidia here as the customer whose purchase orders make everyone else on this list go up.

Broadcom NASDAQ: AVGO

Broadcom beat Nvidia to CPO. Its Tomahawk 6 "Davisson" 102.4 terabit switch with co-packaged optics has been shipping since October 2025, its third generation of the technology, claiming 70% lower interconnect power. Meta, Google and Microsoft custom chips run through Broadcom, so as they go optical, Broadcom collects.

Marvell NASDAQ: MRVL

Marvell wrote a $3.25 billion check (plus up to 27 million more shares on milestones) for Celestial AI, closing in February. Celestial's "Photonic Fabric" puts optical links directly under the chip, not at the edge, claiming 30x the reach of copper at 4x the bandwidth and a quarter of the power.

Marvell's own timeline: Celestial revenue starts late fiscal 2028, hits a $500 million run-rate by that year's fourth quarter, and $1 billion a year later. So you're paying now for a 2028 product. That's either patience or a warning, depending on your temperament.

Credo NASDAQ: CRDO

Credo is the hedge. It makes active electrical copper cables for inside the rack, which is exactly what Jensen meant when he said copper isn't dead. Revenue tripled year-over-year to $407 million in its fiscal Q3 at a 68% gross margin. Now it's crossing over with 1.6T optical DSPs and "ZeroFlap" 800G transceivers.

If optics wins fast, Credo's DSP business grows. If copper hangs on, Credo's cables grow. Either way you're covered. The stock's been the laggard of the group this year, which is sort of the point.

Someone has to build the modules, pull the fiber and make the glass. These companies are less exciting and, right now, that's exactly why you should look at them.

Fabrinet NYSE: FN

Fabrinet is the Thai contract manufacturer that assembles a huge share of the world's 800G optics for Nvidia, Lumentum and friends. Fiscal Q4 revenue was a record $1.32 billion, up 45%, with record EPS of $4.10. Then it guided $1.375-1.425 billion for the next quarter and the stock fell 20% in a day to $477.

The business didn't change. The multiple did. Contract manufacturers earn thin margins, so any hint that customers are pausing hits them first. That also makes Fabrinet the canary. When FN guides up again, the whole group tends to follow.

Corning NYSE: GLW

Corning makes the fiber, and it's building the glass substrates Broadcom uses to package CPO switches. It's up roughly 90% over the past year, which is a lot for a 175-year-old glass company. Hyperscalers pre-buying fiber is the near-term story; CPO packaging is the 2027 one. This is the name to own if you want the theme without a stock that moves 15% on a rumor.

POET Technologies NASDAQ: POET

POET has a clever idea: an "optical interposer" that lets you assemble optical engines without the expensive, slow active-alignment step everyone else uses. It just landed a $50 million initial purchase order from Lumilens, part of a framework that could exceed $500 million over five years, and raised $400 million at $21 a unit, priced above market.

Now the other side. Q2 revenue was $569,925. Not million. Trailing revenue is under $2 million against a market value well above a billion dollars. Net loss was $11.3 million for the quarter. Everything here depends on the Lumilens order actually shipping and the 30,000-plus engines management guided for 2026 actually leaving the building.

Lottery ticket. Treat it like one.

The most interesting CPO technology is still private, and two of the three below have IPO chatter attached.

Ayar Labs

Ayar Labs makes optical I/O chiplets that sit right next to the GPU. Nvidia, Intel and AMD have all invested. It raised $500 million in March at a $3.75 billion valuation and described the round as charting a path to IPO. Nasdaq Private Market's August estimate puts secondary shares at about $74. If you see an S-1 in the next 12 months, this is the most likely name on it.

Lightmatter

Lightmatter's "Passage" is a 3D photonic interposer: instead of optical connections at the edge of a chip, the whole surface becomes I/O, claiming 114 terabits per second from one package. Last known valuation is $4.4 billion. It's the direct technical rival to what Marvell just paid $3.25 billion for, which tells you what an acquirer might pay.

Lightelligence HKEX: 1879

Shanghai-based Lightelligence listed in Hong Kong on April 28, 2026, the first optical interconnect company in the wave of Chinese AI listings, and closed its first day up roughly 380%. Worth watching as a read on how the theme prices outside the US, especially with Washington now threatening Chinese optics imports.

Copper refuses to die. Every time Nvidia extends copper another generation (and it did, at GTC in March), the optical timeline slips a year. Barclays argued in July that broad CPO adoption might not happen until 2029 or 2030. If that's right, the stocks that ran on CPO hopes are two years early. Watch: Nvidia's Rubin Ultra networking specs, and whether Lumentum's CPO laser ramp stays on its "second half of 2027" schedule.

AI capex catches its breath. The August 18 selloff was about a $65 billion Anthropic run-rate that came in below whispers and a report of roughly $3 trillion in off-balance-sheet AI commitments across big tech. Photonics is a leveraged bet on hyperscaler spending. If Microsoft or Meta trims capex 10%, expect these stocks to fall 30%. Watch: hyperscaler capex guides in late October, and Fabrinet's next guide as the early-warning signal.

Then there's the Chinese import question. Reuters reported August 4 that the Trump administration is weighing a ban on new Chinese-made optical modules, and the FCC is preparing a rule. Great for Coherent and AOI, who build in Texas. Bad for anyone whose supply chain runs through Ningbo. Nobody's fully sure which side they're on yet.

The next 12-18 months turn CPO from PowerPoint into invoices. Here's the sequence to watch:

• September 21 – Coherent unveils PhotonLink. Watch for named customers.
• This fall – Nvidia ships Spectrum-X Photonics and Vera Rubin. First real CPO volume.
• November – Lumentum's first $100M+ OCS quarter; AOI reports November 5.
• December quarter – Coherent's first CPO revenue lands (or doesn't).
• Late 2026 – FCC decision on Chinese transceivers. Binary for the supply chain.
• March 2027 – OFC and GTC back to back. Expect 3.2T transceivers and Rubin Ultra optical details.
• H2 2027 – Lumentum's CPO laser ramp.
• Any time – Ayar Labs S-1, Lightelligence's first earnings as a public company.

The pluggable transceiver boom is real, priced, and probably has one more leg as 1.6T ramps. The bigger prize is inside the package, and the companies with a credible path there (Lumentum, Broadcom, Coherent, and whoever buys Ayar or Lightmatter) are where the 2027 money gets made.

Copper had a good 50-year run. Light is up next.