Lightmatter Achieves Record 1.6 Tbps Per Fibre to Accelerate AI Optical Interconnect — Inside the Qualcomm Collaboration, the 8X Bandwidth-Density Breakthrough, and What It Means for the Future of Hyperscale AI Clusters
Combining Lightmatter's Passage™ Co-Packaged Optics chiplet with Qualcomm Technologies' 112G PAM4 SerDes in a 16-wavelength DWDM architecture, the collaboration has produced the first photonic interconnect platform to demonstrate 1.6 Tbps per fibre — silicon-proven, high-volume-ready, and on a roadmap to 100 Tbps per package for next-generation XPUs and switches.
5 min read
Lightmatter, the leader in photonic computing, has announced a historic milestone in AI interconnect performance: the successful sampling of its Passage™ Co-Packaged Optics (CPO) chiplet, demonstrating a record-breaking 1.6 Tbps throughput per fibre. Achieved through a strategic collaboration with Qualcomm Technologies — integrating Qualcomm's advanced 112G PAM4 optical SerDes chiplet with Lightmatter's industry-leading Passage photonic engine — the platform delivers up to 8X more bandwidth per fibre than existing Near-Packaged Optics (NPO) and CPO solutions. It is the first photonic interconnect architecture in the industry to reach this bandwidth density, and it is silicon-proven and high-volume-ready for hyperscaler deployment.
The announcement — made at the 2026 Optical Fiber Communication Conference (OFC) in Los Angeles — is the centrepiece of an unusually concentrated series of photonic interconnect advances from Lightmatter this week, which also includes the unveiling of vClick™ Optics (the industry's first detachable Fibre Array Unit for CPO advanced packaging) and the Passage L20 unified optical engine for NPO and On-Board Optics applications. Together, they represent a coordinated technology platform push designed to establish Lightmatter's photonic engines as the foundational interconnect layer for next-generation AI infrastructure.
The Interconnect Bottleneck — Why 1.6 Tbps Per Fibre Matters
The fundamental challenge facing frontier AI infrastructure today is not compute power in isolation — it is the speed at which data can move between processors. As generative AI clusters scale from thousands to tens of thousands of accelerators, and as individual GPU and XPU chip designs push bandwidth requirements ever higher, the interconnect fabric that ties them together has become the primary constraint on system performance. Traditional copper-based interconnects cannot cover the distances or carry the bandwidth that next-generation AI clusters require. Legacy optical solutions improve on copper, but incremental bandwidth density gains are no longer sufficient to keep pace with the rate at which AI cluster architectures are evolving.
The Lightmatter-Qualcomm achievement directly confronts this bottleneck. By demonstrating 1.6 Tbps per fibre using a 16-wavelength Dense Wavelength Division Multiplexing (DWDM) architecture at 112G per SerDes lane, the Passage L-Series CPO platform maximises the amount of data carried per physical fibre — simultaneously reducing the total fibre count required in a data centre, simplifying cable management, and lowering optical infrastructure costs, while unlocking bandwidth levels that legacy interconnects cannot approach.
"The industry has reached a point where incremental bandwidth density improvements are insufficient to keep pace with the growth in AI cluster size. By delivering 1.6 Tbps per fibre through a 16-wavelength DWDM architecture, this collaboration represents a significant advancement in addressing fibre cabling and physical space constraints of data centre infrastructure. Moving toward a 100 Tbps per-package I/O threshold is a key milestone for the ecosystem in the development of next-generation XPUs and switches."
— Vlad Kozlov, Founder and CEO, LightCounting
How It Works: Passage L-Series CPO, DWDM, and the Qualcomm SerDes Integration
Co-Packaged Optics (CPO) places the optical engine in the same package as the host chip — eliminating the electrical signalling loss and latency that accumulate when optical transceivers are located on a pluggable module at the edge of a switch or server. Lightmatter's Passage™ platform is the world's first 3D-stacked silicon photonics engine, designed from the ground up for co-packaging with XPUs, GPUs, and switching silicon. Its architecture stacks photonic and electronic layers in a single 3D package, delivering the bandwidth density and power efficiency that co-packaged deployments require.
The record 1.6 Tbps per fibre result was achieved by integrating the Passage L-Series photonic engine with Qualcomm Technologies' 112G PAM4 optical SerDes chiplet — itself derived from technology developed by Alphawave Semi, now part of Qualcomm Technologies. The 16-wavelength DWDM architecture multiplexes sixteen independent data streams onto a single fibre simultaneously, each carried on a distinct wavelength of light. At 112G per SerDes lane, the combined throughput reaches 1.6 Tbps per fibre — 8X the bandwidth density of existing NPO and CPO solutions. The Passage L-Series is on track to deliver 100 Tbps and beyond per package for next-generation XPUs and switches, at industry-leading power efficiency.
"The successful validation of our 112G/106G PAM4 connectivity technology within the Passage L-Series CPO architecture is a critical step forward for the industry. As generative AI models grow in size and complexity, they demand a fundamental shift in data movement. Our strategic work with Lightmatter meets this challenge head-on by providing a clear path to 100 Tbps of total I/O per package. We are focused on delivering the best-in-class power efficiency and throughput required to build the world's most advanced AI superclusters."
— Tony Pialis, Executive Vice President and General Manager, Qualcomm Technologies, Inc.
Scale-Up and Scale-Out: A Single Platform for Both Network Architectures
A critical dimension of the Passage L-Series achievement is that it addresses both of the distinct network architectures that define a modern AI data centre. In the scale-up network — connecting accelerators within a single node or multi-rack domain — power efficiency and ultra-low latency are paramount, and traditional copper NVLink or InfiniBand connections are rapidly hitting their limits. In the scale-out network — connecting hundreds of thousands of accelerators across an AI factory — fibre count, cable management complexity, and the total cost of the optical infrastructure are the dominant constraints. The Passage L-Series CPO platform, with its combination of massive bandwidth density per fibre and leading power efficiency, directly resolves the primary limiting factor in both architectures simultaneously.
The companion Passage L20 announcement — a 6.4 Tbps (each direction) unified optical engine supporting both NPO and OBO applications — extends Lightmatter's reach into the multi-rack scale-up and high-bandwidth scale-out transition, achieving double the fibre bandwidth density and making a unified photonic engine viable across a wider range of deployment configurations. And vClick™ Optics addresses the manufacturing and deployment challenge that has been a persistent obstacle for CPO at scale: by enabling a detachable Fibre Array Unit with insertion loss below 1.5 dB, it decouples the photonic packaging process from fibre attachment, enabling high-volume manufacturing at the throughput that hyperscaler procurement requires.
"By combining Qualcomm Technologies' industry-leading 112Gbps SerDes with our Passage L-Series technology, we have achieved a record-breaking 1.6 Tbps per fibre. This milestone proves that our photonic engines are delivering the massive throughput and power efficiency that hyperscale data centres demand today to overcome the physical limitations of traditional copper and legacy interconnects. We are providing a validated interconnect foundation that moves the industry beyond the levels of existing performance and into high-volume, high-performance CPO for the AI era."
— Nick Harris, Ph.D., Founder and CEO, Lightmatter
Availability, Evaluation Kits, and the Road to Hyperscaler Deployment
The record-breaking Passage L-Series photonic interconnects are currently available through Passage Evaluation Kits for lead customer testing — the standard pathway for hyperscaler qualification of new silicon at this stage of the commercialisation cycle. Lightmatter is showcasing its full suite of OFC announcements at booth #4817 in Los Angeles (March 15–19, 2026), with Qualcomm Technologies' Retimer DSP featured at booth #2100 at the same event.
Lightmatter's broader ecosystem of partnerships reinforces the platform's trajectory toward production deployment. In January 2026, the company announced a collaboration with Synopsys to integrate Synopsys 224G SerDes and UCIe IP for 3nm process into the Passage CPO platform — adding the next-generation SerDes integration that will be required for the post-112G era. Together, these partnerships and platform announcements establish Lightmatter as the company most directly addressing the interconnect bottleneck that will determine the pace at which AI infrastructure can continue to scale.
Key Takeaways
- Lightmatter and Qualcomm Technologies have achieved a record 1.6 Tbps throughput per fibre — 8X the bandwidth density of existing NPO and CPO solutions — by integrating Lightmatter's Passage L-Series photonic engine with Qualcomm's 112G PAM4 SerDes chiplet in a 16-wavelength DWDM architecture that is silicon-proven and high-volume-ready for hyperscaler deployment
- The Passage L-Series CPO platform is on a validated roadmap to 100 Tbps and beyond per package for next-generation XPUs and switches — addressing both scale-up connectivity (within nodes and multi-rack domains) and scale-out connectivity (across AI factory-scale clusters) simultaneously, while reducing total fibre count, cable management complexity, and optical infrastructure costs
- The 1.6 Tbps milestone is one of three major Lightmatter announcements at OFC 2026 this week: the Passage L20 (6.4 Tbps each-direction unified optical engine for NPO and OBO, doubling fibre bandwidth density) and vClick™ Optics (industry-first detachable Fibre Array Unit with <1.5 dB insertion loss for high-volume CPO manufacturing) together form a coherent platform spanning the full AI interconnect deployment lifecycle
- Passage L-Series interconnects are available now through Passage Evaluation Kits for lead customer testing — the standard hyperscaler qualification pathway — with Lightmatter at OFC booth #4817 and Qualcomm Technologies at booth #2100 in Los Angeles, March 15–19, 2026
- The collaboration builds on the Qualcomm-Alphawave Semi foundation and is complemented by a separate January 2026 Synopsys partnership to integrate 224G SerDes and UCIe IP for 3nm process — positioning Lightmatter's photonic platform as the interconnect layer of record for both current and next-generation AI chip architectures