Imagine buying a Ferrari capable of 200 MPH, only to drive it exclusively in rush-hour gridlock. That’s the problem facing the semiconductor industry today.
We have built processors—like NVIDIA’s Blackwell—that were unimaginable ten years ago, but we’re choking them with copper wires that haven't fundamentally changed since the telegraph.
The chip isn't the problem anymore; the road is. And we are about to repave the entire highway with light.
Enter Silicon Photonics, the shotgun marriage of two pillars of the 20th century: the silicon microchip and the laser.
Why Silicon Photonics, Why Now?
Over the past fifty years, Silicon Valley has mastered electronics—using electricity to perform logic. We are now geniuses at the "thinking" part.
But we have been surprisingly primitive at the "talking" part. To get data from Chip A to Chip B, we are still pushing electrical current down a copper wire.
This is where the bottleneck is forming. The Copper Wall.
Until recently, the chips were slower than the wires. But with the inflection of AI and the arrival of chips like the GB200, raw computing power has exploded while the wiring has stayed the same.
In modern data centers, up to 50% of the total energy consumed is spent simply moving data around and cooling the system, rather than performing the actual calculations.
Silicon Photonics changes the medium of transport from electricity to light.
Think of it as the technology behind trans-Atlantic fiber-optic cables, but shrunk to the size of a bacterium. It allows data to move between chips as fast as it moves inside the chip, effectively turning a rack of 1,000 separate servers into one giant, unified super-brain.
This report highlights the top silicon photonics stocks to watch, grouped by their role in the value chain. Let’s look at where the investable terrain is forming.
Laser Source
Every silicon photonics system starts with a laser. And here's an inconvenient truth about silicon: it's terrible at producing light. It's a fantastic highway for photons, but it can't generate them. For that, you need exotic materials — specifically, indium phosphide (InP).
This is a structural chokepoint, and in late 2025, it became headline news. NVIDIA has been aggressively pre-allocating capacity at the world's top EML (electro-absorption modulated laser) suppliers, pushing lead times out past 2027 and triggering a supply shortage across the industry. Optical module makers and hyperscalers have been scrambling for secondary suppliers.
The EML market is dominated by a handful of names: Lumentum, Coherent, Mitsubishi, Sumitomo, and Broadcom.
- Of these, Lumentum (NASDAQ: LITE) holds what may be the strongest hand. It is currently the only supplier shipping 200G-per-lane EMLs at volume — the critical component for next-generation 1.6T pluggable transceivers. Its CEO has called 2026 "a breakout year for laser chip sales," and the company recently received the largest single purchase commitment for ultra-high-power CPO lasers in its history. Lumentum is investing heavily in expanding its San Jose InP fab, and analysts expect double-digit price increases on 200G EMLs in 2026 due to the lack of viable second sources.
- Coherent (NYSE: COHR) is the volume leader, with roughly 25% market share in optical transceivers and FY2025 revenue of $5.81 billion, up 23% year-over-year. It is NVIDIA's named collaborator on silicon photonics for the Spectrum-X platform and a major CW (continuous-wave) laser supplier. If Lumentum is the specialized performance leader, Coherent is the industrial-scale factory.
Crucially, this is not a commoditized market. The production thresholds are extreme, capacity is finite, and NVIDIA's strategic lockup of supply has created pricing power that may persist for years. For investors, the laser layer represents the picks-and-shovels play among silicon photonics stocks.
Silicon Photonic Chip (the PIC)
The Photonic Integrated Circuit (PIC) is the core of the system. It's where the waveguides, modulators, and photodetectors all live on a single sliver of silicon. Fab-compatibility magic happens here, and it's where the biggest R&D dollars are flowing.
The companies competing here fall into three camps:
Vertically Integrated Giants
- NVIDIA (NASDAQ: NVDA), which stunned the industry at GTC 2025 by unveiling its own co-packaged silicon photonics switches (more on this in a moment).
- Cisco (NASDAQ: CSCO), through its acquisition of Acacia.
- Broadcom (NASDAQ: AVGO), which has been among the most aggressive advocates for co-packaged optics.
- And Marvell (NASDAQ: MRVL), which in December 2025 signed a blockbuster deal to acquire Celestial AI for up to $5.5 billion — a bet that optical interconnects will become standard plumbing for every AI data center.
Publicly Traded Specialists
- Coherent (NYSE: COHR) and Lumentum (NASDAQ: LITE) are the two largest pure-play optical component companies with deep silicon photonics exposure.
- MACOM (NASDAQ: MTSI) supplies high-speed analog and photonic semiconductors across the data center optical chain.
- Applied Optoelectronics (NASDAQ: AAOI) is a vertically integrated transceiver maker with domestic U.S. manufacturing, a strategic asset as supply chain sovereignty gains importance.
Two smaller, higher-risk names deserve mention:
- POET Technologies (NASDAQ: POET), whose patented Optical Interposer platform integrates lasers, modulators, and electronics on a single chip — still early-revenue (about $5.6 million in production orders landed in Q3 2025), but backed by partnerships with Foxconn and with over $300 million in cash after a major raise.
- And Lightwave Logic (NASDAQ: LWLG), which is pursuing a fundamentally different approach: proprietary electro-optic polymers (branded Perkinamine®) that can be "spun on" to standard silicon wafers to create ultra-low-power modulators. LWLG is pre-revenue but crossed a key threshold in late 2025 when two separate Fortune Global 500 companies advanced to Stage 3 of its design-win cycle, targeting products for hyperscale data centers and AI factories. If its polymers achieve volume qualification, the licensing model could generate significant royalty streams; if not, it remains a speculative bet.
Private Bellwethers
These are the venture-backed companies you can't yet buy on a public exchange, but whose trajectories will shape the industry. Watch these as IPO candidates or acquisition targets.
- Ayar Labs (valued above $1 billion after its 2024 Series D, backed by AMD, Intel, and NVIDIA) is building optical I/O chiplets designed to sit directly on the processor substrate.
- Lightmatter ($4.4 billion valuation, $850 million raised) is shipping its Passage M1000 photonic superchip — a 3D photonic interposer claiming 114 Tbps of bandwidth — and its L200 CPO product is expected in 2026.
- And nEye Systems (backed by Alphabet's CapitalG, Microsoft's M12, and NVIDIA) is developing wafer-scale optical circuit switches, with production samples expected this year.
Foundries
Someone has to manufacture all these photonic chips. That someone is an increasingly competitive group of foundries racing to lock up capacity before demand overwhelms supply.
- Tower Semiconductor (NASDAQ: TSEM) is the current standout. Its silicon photonics revenue hit roughly $52 million in Q3 2025 alone — about 70% growth year-over-year — and management is targeting over $220 million in SiPh revenue for full-year 2025, with a Q4 annualized run rate exceeding $320 million. Tower is investing $650 million to triple its SiPh shipments by mid-2026.
- GlobalFoundries (NASDAQ: GFS) made its own power move in November 2025, acquiring Singapore's Advanced Micro Foundry (AMF) — arguably the most important pure-play silicon photonics fab outside the U.S. — giving GlobalFoundries a claim to being the largest dedicated SiPh foundry in the world. Its "GF Fotonix" platform has become the go-to for fabless photonics companies, and the AMF acquisition allows it to offer a "China-free" supply chain for optical chips, a selling point that may become mandatory for U.S. hyperscalers.
- TSMC (NYSE: TSM) is playing the long game, providing the advanced packaging (its COUPE platform) that enables NVIDIA's co-packaged optics. TSMC's chairman publicly called silicon photonics a "More-than-Moore" pillar at the company's 2025 North America Technology Symposium.
- Fabrinet (NYSE: FN) sits in a related but distinct niche: precision optical manufacturing and packaging. It doesn't design photonic chips, but it assembles and tests a huge share of the world's optical transceivers and components for clients including Lumentum and Coherent. Think of Fabrinet as the contract manufacturer to the laser and transceiver makers.
- And the newest entrant: UMC (NYSE: UMC), which licensed imec's iSiPP300 silicon photonics process and plans to begin risk production in 2026–2027.
The finished, shippable product is the transceiver module, a device that plugs into a server and converts electrical signals to optical and back again. The industry has already begun the upgrade from 800G to 1.6 terabits, with volume shipments ramping through 2026.
TrendForce projects that global shipments of 800G-and-above transceivers will leap from 24 million units in 2025 to nearly 63 million in 2026 — a 2.6x jump in a single year.
The Endgame: Co-Packaged Optics
Here’s the key tension. The pluggable transceiver, the thing that currently is the silicon photonics market, is a transitional technology. The industry knows it.
The true endgame is called Co-Packaged Optics (CPO).
Instead of plugging a separate optical module into the front of a switch, CPO puts the photonics directly onto the switch chip's package itself — millimeters away from the processor, rather than inches.
Why does this matter? Four reasons:
- Energy. NVIDIA's own data shows that its CPO switches deliver 3.5x better power efficiency than traditional pluggable transceivers. At hyperscale, Jensen Huang put it bluntly at GTC 2025: a million-GPU data center using conventional transceivers would need six million individual modules consuming 180 megawatts just for networking. CPO collapses that equation.
- Latency. By eliminating the DSP retimers that pluggable transceivers require, CPO cuts network latency. This is critical for AI inference workloads where models need to answer in milliseconds.
- Resiliency. Fewer components, fewer failure points. NVIDIA claims 10x better network resiliency for its CPO switches, translating to 5x longer uninterrupted AI training runs. When a single hour of downtime on a large GPU cluster costs hundreds of thousands of dollars, that math is irresistible.
- Density. By eliminating bulky front-panel pluggable modules, you free up enormous physical space. NVIDIA's Spectrum-X Photonics switch packs 512 ports of 800 Gb/s — a staggering 409.6 Tb/s of total bandwidth — into a single system.
A couple years ago, CPO was still a "maybe." That changed decisively in March 2025, when NVIDIA unveiled its Quantum-X and Spectrum-X Photonics platforms — the world's first commercial-grade CPO networking switches — built with TSMC, Coherent, Lumentum, Corning, and Foxconn.
The Quantum-X InfiniBand switches began shipping in late 2025. The Spectrum-X Ethernet switches, integrated into NVIDIA's next-generation Rubin AI platform, are expected in the second half of 2026.
Meanwhile, private disruptors like Lightmatter are attacking the problem differently, with a 3D photonic interposer (its Passage platform) that allows I/O connections across the entire surface of a chip, not just the edges, claiming 114 Tbps of bandwidth from a single package. Ayar Labs and its partner Alchip demonstrated a working prototype at TSMC's OIP 2025 event, featuring eight optical engines on a single substrate with over 100 Tbps of scale-up bandwidth per accelerator.
The CPO horse race is now fully underway, and it has real money behind it.
Signals to Watch
For those tracking silicon photonics stocks, here are the near-term signals that matter:
- NVIDIA's Rubin platform ramp: NVIDIA's next-generation Rubin Ultra AI platform is designed from the ground up to integrate silicon photonics networking. The shift to 800G and 1.6T interconnects with CPO creates immediate demand across the entire optical supply chain.
- The foundry capacity race: Tower Semiconductor is investing $650 million to triple SiPh capacity. GlobalFoundries just bought AMF. UMC is entering the market. TSMC is providing the packaging backbone. And yet the industry consensus is that demand will still outstrip supply through at least 2027. Capacity constraints mean pricing power, which means margin expansion.
- The M&A pace: Marvell's $5.5 billion Celestial AI acquisition. GlobalFoundries' AMF deal. Nokia's Infinera acquisition. AMD quietly acquiring Enosemi for optical I/O. These are multi-billion-dollar bets by companies that can see their customers' roadmaps.
- The laser shortage. This is the supply chain story. NVIDIA's strategic lockup of EML capacity at Lumentum and other suppliers has pushed lead times past 2027. CW laser expansion is constrained by long equipment delivery cycles. Both types of laser — EML and CW — use indium phosphide substrates, and InP epitaxial capacity is being stretched to its limits.
- 3.2T development. The industry hasn't even fully ramped 1.6T, and 3.2 Tbps transceivers are already in development, with POET and Quantum Computing Inc. co-developing 3.2 Tbps optical engines targeted for the second half of 2026. Each generational leap requires denser, more sophisticated silicon photonics, which means the demand curve steepens.
The semiconductor industry spent fifty years making transistors smaller. The next chapter is about making connections faster. Silicon photonics is the foundation of that future.
The Ferrari finally gets its highway.