Lightmatter is accelerating the shift toward optical interconnects in artificial intelligence infrastructure through a coordinated set of collaborations with Cadence, Global Unichip Corp. (GUC), and Synopsys. Together, these partnerships point to a decisive industry move away from traditional electrical connectivity and toward co-packaged optics (CPO) as AI systems scale in size, power demand, and architectural complexity.
As AI models expand into the trillions of parameters, data movement has emerged as a primary bottleneck. Electrical interconnects struggle to keep pace with rising bandwidth requirements while staying within practical power and thermal limits. Pluggable optics and near-packaged solutions offer incremental gains, but they remain constrained by physical input-output boundaries, often referred to as the “shoreline” of the chip. CPO addresses this challenge by integrating optical engines directly alongside compute and switching silicon, shortening electrical paths and improving efficiency.
Lightmatter’s approach centers on its Passage platform, a 3D co-packaged optics architecture built on silicon photonics. Passage is designed to deliver high bandwidth density with significantly lower energy per bit, supporting both scale-up and scale-out AI architectures. Through its collaborations, Lightmatter is aligning its photonic technology with silicon-proven electrical interfaces, advanced packaging workflows, and production-ready design ecosystems.
In its collaboration with Cadence, Lightmatter is working to integrate Cadence’s high-speed SerDes and Universal Chiplet Interconnect Express (UCIe) IP with the Passage optical engine. The effort focuses on advanced-node CMOS integration and industry-standard packaging flows. By combining optics-optimized SerDes, chiplet interconnect IP, and electronic design automation expertise, the two companies aim to reduce barriers to CPO adoption. The goal is to make optical interconnect a practical option for hyperscalers designing custom AI accelerators and switches.
Executives from both companies framed the collaboration as a response to fundamental limits in current AI infrastructure. Lightmatter said the next leap in AI performance depends on rethinking how data moves inside clusters. Cadence emphasized that scale-up and scale-out architectures are reshaping data centers and require new interconnect strategies that balance bandwidth, power, and manufacturability.
Analysts see this work as a critical step in moving CPO from concept to deployment. Roy Chua, founder of AvidThink, said the industry has debated CPO timing for years, but integrated photonic designs are now unavoidable. He noted that aligning advanced CMOS IP with future-ready optical architectures gives the ecosystem a practical path beyond conventional electrical limits.
Lightmatter has also taken steps to bring CPO into commercial production through its partnership with Global Unichip Corp., a major provider of advanced ASIC design services for hyperscale customers. The collaboration combines Lightmatter’s Passage photonic interconnect with GUC’s expertise in advanced-node silicon, chiplet architectures, and complex packaging. The joint solution targets AI and high-performance computing workloads that are increasingly constrained by connectivity rather than raw compute.
The Passage platform aims to redefine chip-to-chip communication by breaking free from shoreline limits. It increases optical bandwidth per package and improves power efficiency across large AI clusters. By extending optical scale-up domains across racks, the platform can reduce training time and improve throughput for large foundation models. Lightmatter describes this shift as necessary as networks increasingly function as the computer itself.
GUC’s role focuses on execution and integration. The company brings experience in addressing the thermal, mechanical, and signal integrity challenges that arise when combining dense compute with optical engines. Its involvement strengthens Lightmatter’s ability to deliver a manufacturable and reliable CPO solution to hyperscale customers.
Industry observers see the Lightmatter-GUC partnership as a signal that the CPO supply chain is maturing. Dr. Wei-Chung Lo of Taiwan’s Industrial Technology Research Institute said optical interconnect is no longer optional for hyperscale AI. He said the collaboration demonstrates that both technology and ecosystem readiness are advancing, offering hyperscalers a credible blueprint to overcome bandwidth and power constraints.
In parallel, Lightmatter is working with Synopsys to integrate advanced interface IP into the Passage platform. The collaboration brings Synopsys’ silicon-proven 224G SerDes and UCIe IP, targeting 3nm processes, into Lightmatter’s 3D photonic architecture. The focus is on optimizing the electrical-to-optical interface, which has become a critical design factor as AI systems scale.
By using pre-verified IP, the collaboration reduces design risk and complexity. It also supports high-volume manufacturing by relying on proven design flows. Synopsys’ electronic and photonic design tools play a central role in this effort. These tools enable co-design of electrical and photonic components and help shorten development cycles.
Analysts say such integration is essential for CPO adoption. Alan Weckel, founder of 650 Group, said validated high-speed interface IP is a key requirement for bringing CPO-enabled AI silicon to market. He noted that the collaboration addresses one of the main barriers to deployment by providing a clear and reliable path from design to production.
Taken together, the three collaborations illustrate how AI infrastructure design is shifting. Performance gains increasingly depend on how efficiently systems move data rather than how many transistors they pack onto a chip. Optical interconnect now sits at the center of that equation. Lightmatter is positioning its Passage platform as a foundational technology for this transition.
