Analyst(s): Brendan Burke
Publication Date: September 24, 2026
Marvell demonstrated 2nm optical technologies at ECOC 2026 to increase bandwidth and improve power efficiency within and between AI data centers. The demonstrations follow Dave Lazovsky presenting Marvell’s case for multi-rack optical scale-up at the AI Infra Summit, and they coincide with Coherent’s PhotonLink launch and Broadcom’s Taurus sampling, setting rival production clocks.
What is Covered in this Article
- Marvell’s 2nm demonstrations at ECOC 2026 and the move toward 3.2T connectivity
- Scale-up optics requirements from Dave Lazovsky’s AI Infra Summit talk
- Competitive positioning against Coherent PhotonLink and Broadcom’s Taurus and Davisson
- Manufacturing milestones, Photonic Fabric revenue targets, and the 2028 CPO ramp
The News: Marvell announced optical interconnect demonstrations for ECOC 2026, September 20–24 in Málaga, Spain. They include 2nm 400G-per-lane PAM4, an 800G ZR/ZR+ pluggable with MACsec, and 1.6T ZR and O-band coherent-lite technologies, all described by Marvell as industry firsts.
Its 38 demonstrations also include a 102.4 Tbps co-packaged optics platform, 224G SerDes, and optical memory fabrics. The ECOC announcement outlines Marvell’s connectivity portfolio as AI data center architectures move from 1.6T toward 3.2T and beyond. The demonstrations follow AI Infra Summit 2026 in Santa Clara, the prior week, where Dave Lazovsky, Executive Vice President and General Manager of Data Center Networking at Marvell, presented the scale-up networking requirements this portfolio targets.
Marvell’s 2nm Optics Target AI Scale-Up Beyond the Rack
Analyst Take: Marvell’s ECOC demonstrations supply the product answer to a demand case the company argued the week before. At AI Infra Summit 2026, Lazovsky presented multi-rack scale-up as a requirement of current AI workloads and optics as the only medium that can deliver it once copper runs out of reach. The 2nm portfolio shown at ECOC maps directly onto that case with faster PAM4 links, secure coherent connections, and co-packaged optics. What the announcements still lack is measured efficiency data and confirmed production timing. Coherent disclosed both at its PhotonLink launch during the same show, and Broadcom is already sampling its competing DSP, so Marvell’s portfolio breadth now needs the numbers and dates that let buyers act on it.
AI Infra Summit Defined the Demand Marvell’s 2nm Optics Serve
Lazovsky’s summit talk quantified why scale-up domains must leave the rack. Multi-trillion-parameter models and KV caches that grow with context length and shared user bases already push memory demand past a 72 XPU rack. Signal integrity confines copper scale-up links to about 2 meters as lane rates move from 224G to 448G, which he identified as the reason scale-up domains reside inside the rack today, and he framed optics for multi-rack domains as a must-have driven by AI workloads.
The engineering priority ties to economics, since latency in reasoning models translates directly to revenue. The commercial logic matches Futurum’s survey data, where 35.2% of respondents in Futurum’s 1H 2026 Data Center Semiconductor Decision Maker Survey identify tokens per watt as their primary AI infrastructure productivity benchmark. Marvell’s 400G-per-lane PAM4 demonstration and its 102.4 Tbps CPO platform are built for that transition, and Lazovsky said Marvell customizes end-to-end scale-up links from XPU I/O through the switch, offers both NPO and CPO form factors, and supports UALink, NVLink through its NVIDIA partnership, and low-latency Ethernet variants.
Coherent and Broadcom Set the Competitive Clock
Coherent and Broadcom disclosed the production timing that Marvell’s ECOC release omits. Coherent launched PhotonLink at the same show, an integrated optics platform spanning CPO, NPO, and chip-to-chip connectivity, with more than 25 customer engagements, anchor hyperscale agreements, scale-out CPO revenue guided to begin in the December quarter, and a 2030 addressable market claim raised to $90 billion. Broadcom has begun sampling its 3nm Taurus DSP with 400G-per-lane optical interfaces and already ships Tomahawk 6 Davisson, the first 102.4 Tbps Ethernet switch with co-packaged optics. Marvell’s counterpositions are node leadership and link ownership. Its 2nm demonstrations put it a process node ahead of Taurus, and Lazovsky argued that owning the full load-store path from XPU I/O through electrical-to-optical conversion to the switch lets customers bring up systems with a single supplier rather than three or four. The Photonic Fabric adds an analog SerDes designed for the optical link rather than adapted from copper, which Lazovsky said cuts system power significantly. Marvell has progressed from 5nm Nova through 3nm Ara to these 2nm demonstrations with limited disclosures of efficiency gains against earlier generations, and buyers weighing Coherent’s stated content of up to $15,000 per 100 Tbps switch or compute chip will ask for those numbers.
Secure Coherent Pluggables Serve the Scale-Across Domain
Integrated security gives Marvell’s coherent pluggables a practical role as AI workloads extend across data centers, the scale-across domain that sits alongside scale-up and scale-out in Marvell’s framing. The Electra-powered COLORZ 1600 combines MACsec with 1.6T connections across campus, metro, and regional distances, reaching up to 1,000 km. The Libra-based COLORZ 800 supports 800G at up to 1,000 km, with longer reaches of 2,000 km at 600G and 3,000 km at 400G. Both support OIF, OpenZR+, and OpenROADM interoperability, and coherent pluggables connect through standard Ethernet switches and routers without separate external transport equipment. Coherent is demonstrating a 3.2T pluggable transceiver in an OSFP-size package at the same show, so generation gaps in the pluggable race are compressing. Operators should assess these offerings against their required reach and security, since the disclosed configurations serve different network needs.
Manufacturing Milestones Decide the 2028 Ramp
The 2028 volume window disclosed at AI Infra Summit sets the schedule Marvell’s manufacturing must meet. Summit panelists put the yield threshold for co-packaged optics at 99.9% for components and attach, since a failed attach scraps a package carrying tens of thousands of dollars of HBM and compute. Marvell identifies resource planning, test capacity, and manufacturing yields as requirements for supplying these networks at volume, and its March announcement targeted second-half 2026 sampling for Electra, Libra, COLORZ 1600, and the Libra-enabled COLORZ 800. The ECOC release adds live demonstrations without confirming whether sampling has begun. The revenue commitments attached to the Celestial AI acquisition give the market checkable milestones, with Marvell targeting a $500 million annualized Photonic Fabric run rate in Q4 fiscal 2028 and $1 billion by Q4 fiscal 2029. Lazovsky said 4 customers and the ecosystem around them represent roughly 80% of data center capex, he sized above $1 trillion today and above $3 trillion ahead, so a small number of qualification decisions will determine whether those targets land. Marvell is also working with hyperscale customers on micro-LED and micro-VCSEL integration, and the Lumentum, Qualcomm, and Corning VCSEL die-to-die demonstration at ECOC shows the laser field remains open.
What to Watch
- Whether Marvell quantifies power-per-bit gains from its 2nm optics against Nova and Ara
- Whether Electra, Libra, COLORZ 1600, and COLORZ 800 meet the second-half 2026 sampling target
- Whether Photonic Fabric revenue reaches the $500 million annualized run rate targeted for Q4 fiscal 2028
- Whether Broadcom’s Taurus sampling converts to module design wins in the 3.2T transition
Read the complete announcement on Marvell’s optical technology demonstrations at ECOC 2026 on the company website.
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.
Other Insights from Futurum
Marvell Q2 FY 2027: Connectivity and Custom Silicon Fuel AI Growth
Broadcom’s DSP Launch Intensifies the AI Optics Race with Marvell
Will NVIDIA Investment Accelerate Marvell’s XPU Growth?
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.
