Intel Claims Monopoly on 12-Inch Indium Phosphide Silicon Photonics Wafer Capacity as Co-Packaged Optics Enters Mass Production Race
nashnova research
Intel declared at the OCP Summit that its foundry is the only one capable of mass-producing 12-inch indium phosphide silicon photonics wafers, but with Nvidia, Broadcom, TSMC and others already shipping or ramping CPO products, the real test is whether the claim translates into named, deliverable products.
Why does Intel say it stands alone?
Silicon can guide and modulate light but cannot emit it — photonic chips need indium phosphide (InP), a compound semiconductor that produces laser light, as their light source.
Conventional InP substrates top out at 6 inches or smaller, while the silicon chip industry standard is already 12 inches. This means → whoever puts InP onto a 12-inch line first can ride mature silicon-fab economics to drive down cost.
Intel claims its foundry is the only facility capable of volume-producing 12-inch InP silicon photonics wafers, and plans to integrate InP light sources directly into its 12-inch silicon photonics process.
What problem is this technology trying to solve?
Intel executive Nic Dubé called silicon photonics the "golden hammer" for next-generation systems, pointing at a clear bottleneck: copper interconnects — today's main way of moving data between chips — have hit limits on reach, bandwidth and power.
In plain terms = copper is too short, too slow and too power-hungry. Replacing copper with light is the obligatory path for scaling AI data centers.
Intel estimates its photonic system can push energy consumption below 5 picojoules per bit, while offering both integrated lasers and support for external light sources — what it calls a US-based, multi-wavelength light-source supply for the industry.
How much production history does Intel actually have?
Since commercializing its silicon photonics business in 2016, Intel has shipped more than 8 million optical transceivers and 32 million integrated lasers.
Accelerated life testing shows hybrid-laser failure rates below 0.1 FIT (FIT = failures in a billion hours — below 0.1 means extremely high reliability).
This reflects genuine long-term manufacturing experience in silicon photonics, but the shipments are concentrated in conventional transceivers, not CPO products.
Why can't the market take this at face value right now?
5 picojoules per bit remains a technical target — Intel has not disclosed an actual volume-production timeline.
Analyst Brendan Burke put it bluntly: this is a "2028-and-beyond story," and Intel's photonics commercialization history demands patience. His focus is on whether silicon photonics can move from keynote promises to named product milestones.
Intel simultaneously plays chip designer, system-architecture advocate and foundry — three roles at once. Customers question how credible the "serving the whole industry" pledge really is. In plain terms = when you are both referee and player, customers worry you will serve yourself first.
How far ahead are the competitors?
Nvidia's Spectrum-X Ethernet photonic switch is already in volume production with early customer deployments under way.
Celestica launched a switch on Broadcom's Tomahawk 6-Davisson CPO platform at 102.4 Tbps, using field-replaceable external lasers.
TSMC is building out CPO manufacturing with its COUPE (Compact Universal Photonic Engine) platform; Zhongji Innolight has a dedicated CPO line and has begun small-batch production of 800G CPO modules; Huawei unveiled an Ascend 960 super-node in September using NPO — near-package optics, a design that places optical modules close to the chip.
Analyst Sameh Boujelbene cautioned: "Early commercialization should not be confused with broad adoption."
What is the real verification point for this race?
Intel's 12-inch InP capacity claim is essentially a capacity-monopoly card — but its face value depends on converting it into named, deliverable products.
This means → the market will judge Intel's AI optical-interconnect standing not by "what wafer can you make" but by "what product, at what price, on what date."
Optical topics at this year's OCP nearly doubled versus last year. This reflects an industry shifting from roadmap phase to live deployment — the window for promises without products is closing.
市场有风险，内容仅供研究参考，不构成投资建议。