According to LightCounting’s 2026 forecast, shipments of silicon-based pluggable modules are expected to grow roughly tenfold by 2026 compared with the total forecast for 2024. As the industry races to keep up with demand, silicon photonics is further cementing its role as a critical enabling technology—one that not only supports scale-up but also propels the field into its next era of innovation, an era defined by a shift from front-panel pluggable modules to advanced interconnect architectures.
Yet the challenges posed by surging AI workloads and the demands of network infrastructure are evolving rapidly. The concept of scale-up, for instance, spans the entire industry value chain, which means bottlenecks and potential constraints can arise at many different points. Technological progress is therefore gradual and iterative. Deployment plans with firm timelines—particularly for co-packaged optics (CPO)—are essential milestones on the path to productization and, ultimately, delivery to end users. But meeting those deadlines while ensuring process repeatability requires new technical benchmarks in areas such as testing and packaging, both of which present distinctive challenges.
Moreover, the tenfold shipment growth projected between 2024 and 2026 may even understate actual future demand.
To explore this possibility and the important questions it raises, Soitec convened the 2026 Silicon Photonics Workshop, bringing together industry leaders to discuss how to advance a technology that is both an enabler and a catalyst for progress. The seminar focused on several key themes and trends shaping the future of optical networking.
Laser Demand Comes to the Fore
The sharp rise in production of silicon photonics-based optical modules is prompting foundries to expand capacity in tandem. Ed Preisler, Director of Technology Development at Tower Semiconductor, outlined the company’s capacity expansion. Preisler noted that the current growth cycle differs in both nature and scale from previous fab expansion cycles. The driving force, he said, is that multiple module standards—spanning multi-wavelength, single-channel 100 Gb/s, and single-channel 200 Gb/s rates—all require photonic integrated circuits (PICs), rather than being tied to a single product or product category. Tower currently operates three 8-inch fabs and one 12-inch fab; to meet customer demand, the company has taken a range of measures to accelerate process qualification and production ramp-up.
Preisler cited cross-fab matching and the information sharing it enables as key to Tower’s ability to adapt its silicon fabs to current and future demand. This has also opened up new opportunities for the company. Optical circuit switching (OCS) is one major area of focus. Preisler said Tower is actively engaged in OCS and is working closely with partners such as Oriole Networks and Salience Labs.
According to Preisler, the biggest wildcard today is near-packaged optics (NPO). Although silicon photonics already plays an active role in several areas—including OCS systems and the push toward single-channel rates above 200 Gbps—NPO is expected to become the real core driver of the technology’s future growth.
That raises an obvious question: where will the enormous number of high-power lasers needed to support this growth come from?
The Answer Follows
Like Tower Semiconductor, equipment developer and manufacturer Lumentum is expanding capacity at its facilities. As a producer of indium phosphide (InP) lasers, Lumentum occupies a critical position in the optical networking value chain. Supply chain issues took center stage in a presentation by Chief Technology Officer Matt Sysak.
Lumentum’s external laser source modules are used in NVIDIA’s Spectrum-X CPO switches, which entered mass production this summer. Beyond this commercial milestone, 2026 is also a pivotal year for Lumentum’s R&D. The company showcased an 800 mW super-high-power (SHP) InP laser for CPO, along with a 16-channel dense wavelength division multiplexing (DWDM) ultra-high-power (UHP) light source. These demonstrations come as InP lasers advance along two distinct yet interrelated paths: pursuing ever-higher power levels, and transmitting data in different directions. The development of the 800 mW SHP laser, combined with the existing strengths of UHP multi-wavelength sources, illustrates the technical progress now being achieved.
In addition to its 16 DWDM channels, Lumentum’s multi-wavelength UHP laser offers wavelength control accuracy within ±25 GHz and high efficiency (up to 350 mW per laser and/or 23–24.5 dBm per fiber). A pluggable external laser (ELSFP) architecture is also available. Meanwhile, the 800 mW SHP product is expected to further increase ELSFP density.
These solutions carry significant market implications and do not exist in isolation. According to Sysak, Lumentum drew heavily on its existing capabilities in the pump laser business when developing its 400 mW lasers. He also noted that process elements from single-frequency laser manufacturing can be applied to the company’s current multi-wavelength light source products.
On scaling—the question Preisler raised—Sysak said Lumentum’s newly announced facility in Greensboro, North Carolina, will help the company meet current and future demand for continuous-wave (CW) lasers, which are essential to the emerging NPO and CPO architectures. The plant is expected to come online in mid-2028.
A Materially Rich Technology Roadmap
While Lumentum’s North Carolina fab is the physical embodiment of its capacity expansion, Sysak noted that the surge in InP investment has made the entire supply chain more noteworthy—particularly as standardization for network scale-up accelerates.
Materials were another major focus of discussion.
Philippe Absil, a speaker from imec’s silicon solutions department, said imec’s materials capabilities are at the industry—and indeed global—leading edge. In the context of current AI data center architectures, these capabilities are essential to meeting the needs of scale-out, scale-up, and scale-in. Each scaling framework—scale-in, scale-out, and scale-up—maps to different goals within a comprehensive scaling roadmap. These goals span the evolution from pluggable optical modules (for larger-scale horizontal connectivity) to co-packaged optics (CPO, for vertical interconnection), to 3D integrated optics (for scale-in interconnects).
Absil argued that silicon provides a solid foundation for scaling. It is compatible with advanced packaging technologies and offers predictable performance, making it well suited to the fabless-plus-foundry business model.
The industry’s familiarity with silicon also makes it an ideal base material. Absil said the goal now is to transcend the intrinsic performance limits of silicon through silicon photonics.
Absil highlighted a range of material platforms imec is pursuing at varying levels of maturity, all aimed at achieving that goal. At the component level, these include silicon germanium (SiGe) for modulators, germanium (Ge) for photodetectors, and silicon nitride (SiN) for edge couplers. Imec is also leveraging its technical capabilities and expertise—particularly its 300 mm wafer platform—to work with organic electro-optic polymers for modulators. Absil also shared information on emerging materials such as barium titanate, which are expected to play an important role in advancing active devices, enabling 3D integrated optics, and achieving interconnect energy efficiency below 1 pJ/bit. Thin-film lithium niobate (TFLN) is attracting growing attention. Mian Zhang, CEO of HyperLight, said the TFLN platform is maturing from a “new” material into a full-fledged ecosystem. LightCounting has doubled its shipment forecast for 1.6T-port TFLN devices over the 2025–2026 period (i.e., within the next five years), and has raised its 3.2T-port forecast to three times the previous level.
Over the past eight years, HyperLight has introduced multiple TFLN integration solutions and continually refined its manufacturing processes, fully exploiting TFLN’s material advantages—not only relative to bulk lithium niobate (bulk LN) but also in competition with other material platforms. This winter, the company officially launched its single-channel 400G TFLN photonic integrated circuit (PIC).
HyperLight’s progress also highlights a challenge common across the industry ecosystem. Zhang noted that material property concerns are no longer the primary constraint; the bottleneck now is capacity.
Zhang also pointed to another shift over the past 12 months. Previously, many in the industry asked why the company chose to build its platform on TFLN, he said; over the past year, however, such questions have disappeared.
Silicon Photonics Comes into Its Own
During the conference, imec’s Philippe Absil discussed silicon photonics from a materials perspective. He noted that silicon photonics can transcend the limits of pure silicon, with performance that extends beyond silicon itself—thanks in part to its ability to leverage the strengths of other materials in the ecosystem.
Guilhem de Valicourt, Senior Director at Ciena, echoed this view in his presentation, focusing on hybrid-media deployment and the system architectures needed to move beyond the limits of copper interconnect.
De Valicourt’s roadmap traced the industry’s evolution: from re-timed optical modules, to linear pluggable optics (LPO), to near-packaged optics (NPO) and co-packaged optics (CPO). He said this evolution is intensely focused on high-density integration (and on raising edge bandwidth density); in the three years since 2023, edge bandwidth density has risen from 50 Gb/s/mm to 500 Gb/s/mm, while per-channel rates have moved from 100G toward 200G. De Valicourt sees high-performance optical engines as the key to this leap. Silicon photonics is precisely what ensures optical interconnect can keep pace with the growth in SerDes edge bandwidth density.
Against this backdrop, the value proposition of silicon photonics is becoming ever clearer. Driven by multiple factors across the supply chain, the field is experiencing significant growth, and the technology itself continues to evolve.
Rachid Taibi, General Manager of Soitec’s business unit, explained that silicon-on-insulator (SOI) technology is not only scaling in volume but also improving in performance. With the rise of AI, the challenge for silicon photonics has shifted from a purely technical scaling problem to an industrial scaling challenge.
Soitec is building out and refining its portfolio around this theme. Taibi noted that wafer customization directly affects the consistency and repeatability of manufacturing specifications. The company’s SOI customization work for next-generation devices encompasses improved silicon layer uniformity, control of resistivity and bulk microdefects (to support advanced packaging architectures and through-silicon vias), and a transition to thicker buried oxide (BOX) layers.
Soitec’s ambitions also extend beyond SOI, with the company exploring freestanding and heterogeneous photonic device platforms. Taibi said heterogeneous integration (as referenced by Absil and de Valicourt) expands the capacity and functionality of silicon photonics. Soitec is applying its proprietary Smart Cut technology to its scalable platform to enable integration of lithium niobate (LN) and indium phosphide (InP). The company is also stepping up R&D investment in its lithium niobate-on-insulator photonic platform (Photon-LNOI). According to Taibi, this platform builds on a proven 150 mm (6-inch) manufacturing process, and the company has already begun developing 8-inch solutions.
Taibi also described Soitec’s progress on InPOSi (InP-on-silicon), a platform for large-scale integration of III-V materials. This substrate technology is designed to enable cost-effective electro-absorption modulated lasers, tunable lasers, and/or semiconductor optical amplifiers. The solution employs a relatively thick buried oxide (BOX) layer between the InP seed layer and the bulk silicon substrate. Taibi said Soitec is working this year to scale the platform to 150 mm (6-inch).
High Barriers to Entry
At Soitec’s 2026 Silicon Photonics Workshop, speakers and attendees alike understood without saying it: scaling is a fundamental challenge, and the rapid pace of technological evolution only complicates matters. Take the adoption of NPO (near-packaged optics)—even counting only from this winter’s OFC—this trend has already altered the industry’s evolutionary trajectory. With a slate of industry events approaching in 2027, forecasters will no doubt refine their roadmaps accordingly.
The value of such seminars goes well beyond exchanging the latest statistics and ideas. They also push participants to examine familiar technologies, challenges, and concepts from fresh perspectives. 2027 will bring new challenges, but it will also undoubtedly yield new ways of seeing.
Looking ahead, the conversation will continue at Asia Photonics Expo 2027, Asia’s premier photonics gathering, scheduled for 3–5 March 2027 at the Sands Expo and Convention Centre in Singapore. The 2027 edition marks a strategic expansion: for the first time, the Asia Photonics Expo will be held alongside the Asia Semiconductor Expo and the Asia Electronics Expo, creating a tri-expo platform that spans the full chain from core technologies and key components to end applications. For the silicon photonics community, this convergence is particularly relevant—heterogeneous integration, advanced packaging, and co-packaged optics all sit precisely at the intersection of photonics, semiconductor manufacturing, and electronics that APE 2027 aims to bridge.
With more than 500 exhibitors and 6,000 expected visitors, the show will offer a timely venue for the industry to take stock of the scaling challenges and material innovations discussed here, and to chart the next steps on the roadmap toward 2027 and beyond.
*Source: Translated from Semi-insigts.com WeChat Account