Lightmatter unveiled a 204.8 Tbps CPO switch tray built with Flex and a 102.4 Tbps four-XPU tray at the 2026 OCP Global Summit, the first systems to carry Passage co-packaged optics and Guide DR lasers in Open Rack Version 3 and Modular Hardware System designs. Futurum assesses whether open rack standards can pull co-packaged optics into volume production.
What is Covered in this Article
- A 204.8 Tbps CPO switch tray in 1RU developed with Flex as a production validation system
- A 102.4 Tbps XPU tray carrying four accelerator ASICs and two CPUs on DC-MHS chassis infrastructure from a major ODM partner
- The first ORv3 and MHS-based systems with Passage CPO, Guide DR laser modules, and vClick detachable optics
- Fiber connectivity from SENKO and expanded beam PRIZM TMT connector technology from Corning
- The first proof of concept for the Open Silicon Photonics for AI Systems white paper, with the OCP initiative now above 20 companies
- Lightmatter’s roadmap through NPO form factors in 2027-2028 and the Passage M1000 interposer in late 2028
The News: Lightmatter unveiled Passage co-packaged optics (CPO), Guide DR external laser modules, and vClick Optics integrated into a switch tray and an XPU tray built to Open Compute Project (OCP) rack and server specifications at the 2026 OCP Global Summit in San Jose. Flex collaborated with Lightmatter on the 204.8 Tbps CPO switch tray, a 1RU production validation platform that integrates eight Passage CPO bidirectional (BiDi) engines powered by Guide DR laser modules in a liquid-cooled chassis. Separately, Lightmatter developed a 102.4 Tbps CPO XPU tray housing four accelerator ASICs and two CPUs on Data Center Modular Hardware System (DC-MHS) infrastructure provided by another major ODM, with expanded beam PRIZM TMT optical connector technology from Corning and fiber connectivity from SENKO. Both designs follow the Open Rack Version 3 (ORv3) and Modular Hardware System (MHS) specifications, and the XPU tray is the first proof of concept for the “Architecture Vision: Open Silicon Photonics for AI Systems” white paper contributed to OCP.
“In an AI factory, interconnect bandwidth sets the ceiling on compute,” said Nick Harris, Ph.D., Founder and CEO at Lightmatter. “Alignment towards a unified rack architecture for silicon photonics will significantly accelerate adoption.” The companies work together in the Open Silicon Photonics for AI Systems initiative, announced in August, which now includes more than 20 companies.
Lightmatter and Flex Bring 204.8 Tbps Co-Packaged Optics Into Open Racks
Analyst Take: The Lightmatter CPO systems at OCP Global Summit mark the moment co-packaged optics stopped being a technology demonstration and started looking like data center infrastructure. Every CPO rack shown publicly to date, including prototypes from NVIDIA and Broadcom, has been a proprietary design that bears little resemblance to the hardware hyperscalers actually deploy. Lightmatter and Flex built the opposite artifact: a 204.8 Tbps switch tray in a single rack unit that follows ORv3 and MHS specifications, uses DC-MHS chassis infrastructure, and exists specifically so Flex can validate its production design and manufacturing flow for photonic systems. The companion XPU tray carries four accelerators and two CPUs with 102.4 Tbps of scale-up optical bandwidth, with Guide DR laser modules that blind mate as the tray slides into the rack. Lightmatter claims the switch design packs the optical bandwidth of a two-to-four rack unit system into one, with 2-4x better rack faceplate density than conventional ELSFP lasers and unidirectional fiber harnesses. The industry’s CPO conversation is shifting from whether the optics work to whether an ODM can build them in volume, and that shift favors whoever defines the open template first.
Futurum forecasts AI networking spend approaching $350 billion in 2030, and optics can win a significant share of it only on terms cloud operators have already set: chips and switches from different vendors connecting over an open optical interface, with roadmap stability and multi-vendor supply. That is the gap the Open Silicon Photonics for AI Systems initiative is built to close. Launched in August with 19 companies and now above 20, including Celestica, Dell Technologies, Flex, Foxconn Interconnect Technology, Qualcomm, and Quanta Cloud Technology, the initiative published a 300-page architecture vision document that maps a shared MHS-based CPO rack architecture spanning silicon photonics, VCSELs, and micro-LEDs, with first specifications expected in Q4 2026. The white paper incorporates link-layer work from OIF and the OCI MSA, which was first to specify BiDi and DWDM for data center optics, rather than duplicating it. Lightmatter demonstrated BiDi and DWDM in rack validation systems a year before the OCI MSA was announced and reconfigured its 16-wavelength demo to replicate the MSA’s Gen 2 spec at OFC. Standards participation is necessary rather than sufficient. The initiative must now convert a well-regarded document into adopted specifications while NVIDIA ships proprietary CPO switches into the same window.
Flex Turns a Demo Into a Production Validation Exercise
The Flex collaboration is the most commercially meaningful element of the announcement because of what Flex is doing with it. Flex uses the switch tray design to validate its entire production process and production flow for photonics, and that the tray is engineered to look like hardware a hyperscaler build chain could produce in volume. That positions Flex ahead of an ODM field that has mostly waited for photonics to arrive. Celestica dominates the switch ODM arena today on copper infrastructure, and the transition to optics opens share-shift windows that forward-leaning manufacturers want to capture. Dell Technologies, a larger player still by hyperscale data center revenue, is a major contributor to the standard and is presenting jointly with Lightmatter at the summit.
The XPU tray extends the story to compute: it is the industry’s first compute tray to connect four XPUs with 102.4 Tbps of scale-up optical bandwidth in a single rack unit, replacing the copper backplanes that limit scale-up domains to one or two racks today. One caution applies. The switch tray is a validation platform rather than a product; no hyperscaler has publicly committed to deploying either design, and Lightmatter’s first production sampling year is 2027, so an execution gap remains between an OCP booth and a purchase order.
Serviceable Connectors and Laser Supply Address CPO’s Two Hardest Objections
The component debuts answer the objections that have kept operators cautious on CPO. vClick Optics, developed with SENKO using its SEAT detachable optical interface and Metallic PIC Coupler technologies, gives Passage a low-loss, field-detachable fiber connection compatible with mold-and-grind advanced packaging. Lightmatter told Futurum the connector demonstrates insertion loss below 1 dB, which compares to an industry benchmark near 1.5 dB for permanently attached fiber requiring active alignment that can take hours, and the company is running a live loopback test at the summit to prove pluggability. No operator will deploy CPO that cannot be serviced in the field, and a detachable connector that beats glued fiber on loss removes the maintenance argument against it.
Guide DR attacks the second objection, laser supply. It is the first external laser module designed to sit inside the tray with liquid cooling integrated into the module architecture rather than wrapped around a pluggable. The underlying Guide laser platform is fabricated in high-volume CMOS fabs rather than the compound semiconductor lines that constrain the industry today, and Lightmatter estimates its foundry partner capacity could add roughly 30% to industry laser capacity over the next couple of years. Lightmatter cites approximately 2 million failure-free device hours on Guide lasers and approaching 1 million on Passage in its validation systems, with both product lines under evaluation at the largest hyperscalers, foundries, and ASIC vendors. Evaluations and device hours build credibility ahead of production.
The Open Template Races NVIDIA’s Vertical Stack to the 2028 Volume Window
The competitive field clarifies what Lightmatter is attempting. NVIDIA’s Quantum-X Photonics InfiniBand switches began shipping in 2026, with Spectrum-X Photonics Ethernet following, and Broadcom’s Tomahawk 6 Davisson delivers 102.4 Tbps with co-packaged optics, so the incumbents are already selling CPO inside vertically defined systems. Coherent launched its PhotonLink integrated optics platform in September, claiming a $90 billion 2030 TAM. Ayar Labs closed a $500 million Series E in 2026 to scale optical I/O chiplets, Celestial AI continues to pursue its photonic fabric, and Marvell is building CPO into custom XPU programs.
Lightmatter’s differentiation is the open systems path. Rather than winning the optical engine socket inside someone else’s switch, it is defining the rack architecture that lets any XPU or switch vendor adopt photonics on OCP infrastructure. The company frames its opportunity as roughly $50 billion for CPO optical engines with a laser opportunity of a similar size, figures consistent with Futurum’s view of AI networking spend approaching $350 billion in 2030, but still vendor-sized estimates. Timing is the pivot. Lightmatter expects 2027-2028 to be the years of near-packaged optics in CPX, OpenCPX, and XPO form factors because NPO requires no silicon redesign, with OpenCPX taking volume in 2028 as front-plate density runs out at 200T-plus switch generations. Late 2028 is when the company plans commercial mass production readiness for its Passage M1000 interposer, arriving alongside the industry’s projected 3D CPO volume ramp into 2029, and it reports two major players engaged in M1000 designs. Hyperscalers have told every supplier they will not accept single-vendor optical lock-in, and the only rack architecture answering that demand now has 20-plus companies and working hardware behind it.
What to Watch
- Whether a named hyperscaler commits to deploying ORv3 photonic trays after the first OCP specifications land in Q4 2026
- Whether Flex’s production validation yields a manufacturing-ready design ahead of Lightmatter’s 2027 production sampling year
- Whether vClick’s sub-1 dB detachable loss holds in third-party field testing
- Whether Guide laser output materializes at CMOS-fab scale and eases the industry laser shortage
- Whether OpenCPX volume in 2028 arrives on schedule and positions the Passage M1000 interposer for the 2029 3D CPO ramp
Sources
1. Lightmatter Delivers Open Rack CPO Switch and XPU Scale-Up Server Through Independent Collaborations with Flex and Other Industry Partners
Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.
Author Information
Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers.
Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research.
Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.
