Intel在OCP宣称将以12英寸InP硅光供应CPO光源，但仍是2028年后路线图、无量产时间表或订单

Intel is pitching its foundry as the industry's supplier of light sources for AI data center optics, claiming at the 2026 Open Compute Project (OCP) Global Summit that it is the only high-volume fab that can deliver 12-inch indium phosphide (InP) silicon photonics (SiPh) at scale to the industry.

The claim comes as co-packaged optics (CPO) moves from roadmaps toward deployment, with Nvidia's CPO switches in production and switch makers unveiling systems on Broadcom's CPO silicon, raising the stakes over who can supply lasers and photonic wafers in volume.
Nic Dubé, corporate vice president of Intel's Data Center Systems and Solutions Group, called silicon photonics the "golden hammer" of next-generation systems, according to Wccftech . Dubé, who delivered a keynote at the summit in San Jose on October 13, said the industry is hitting limits in copper links' reach, shoreline bandwidth, and energy per bit, and that Intel's teams are working around the clock to make silicon photonics a reality in the next couple of years.
Moving lasers onto 12-inch silicon
Silicon guides and modulates light well but cannot emit it efficiently, so photonic systems rely on compound semiconductors such as InP for their lasers. Conventional InP substrates are still limited to 6-inch (6-inch) wafers, with 3- and 4-inch formats widely used, compared with the 12-inch standard in silicon CMOS manufacturing, Sohu reported. Because InP crystals are brittle, grow slowly, and are hard to keep defect-free, each wafer yields far fewer laser chips, slowing capacity expansion. Intel plans to integrate InP light sources into its 12-inch silicon photonics process.
Intel said the 12-inch wafers will carry the silicon needed for future near-package optics (NPO) and CPO in HPC and AI, Wccftech reported. The company is offering integrated lasers designed not to need servicing, while also supporting community standards for external light sources, which it said would give the industry a US fab supply of multi-wavelength light sources. Intel estimates its photonic systems can bring energy consumption below 5 picojoules per bit.
On reliability, Sohu reported that Intel has shipped more than 8 million optical transceivers and 32 million integrated lasers since its silicon photonics business went commercial in 2016, and that its hybrid lasers showed a failure rate below 0.1 FIT in accelerated life testing.
Targets rather than timelines
The 5-picojoule figure is a technical goal, and Intel has not announced an actual mass-production schedule. Brendan Burke, an analyst at Futurum Research, called Dubé's photonics roadmap "the first coherent systems narrative Intel has offered since exiting the supercomputing-prime business," but added: "That is a 2028-and-beyond story, and Intel's photonics commercialization history warrants patience." Among the items Burke flagged to watch is whether silicon photonics moves from keynote commitment to a named product milestone.
Sohu also questioned how much trust customers will place in Intel's pledge to serve the whole industry when the company is at once a chip designer, a system architecture promoter, and a foundry.
CPO enters production elsewhere
Intel's push lands in a crowded field. Optics sessions at OCP nearly doubled from last year, wrote Sameh Boujelbene of Dell'Oro Group, who also noted that Nvidia has moved its Spectrum-X Ethernet Photonics switches into production, with early customer deployments beginning. Celestica introduced its DS6005, a 102.4 Tbps switch built on Broadcom's Tomahawk 6-Davisson CPO platform that uses field-replaceable external lasers.
Pluggable modules are not standing still, Boujelbene added, pointing to an Nvidia and Cisco plan for a 400 Gb/s-per-lane pluggable form factor. "Early commercialization should not be confused with broad adoption," the analyst wrote.
TSMC is also positioning its Compact Universal Photonic Engine (COUPE) platform for CPO manufacturing, while Chinese suppliers are moving in parallel, according to Sohu . Zhongji Innolight has built a dedicated CPO production line and begun small-volume production of 800G CPO modules, and Huawei launched its Ascend 960 supernode in September, which uses NPO technology.