We attended the Marvell Industry Analyst Day in Santa Clara, California today. The company focused almost exclusively on the Data Center market opportunities the company is pursuing. Numerous spokespersons said that it thinks it has enough manufacturing capacity to hit its 2026 plans and it is working with customers and its suppliers to set up for a successful 2027 and beyond. Marvell shared many roadmaps for enabling hyperscalers that are building AI factories, most notable about Scale-Up, Scale-Out and Scale-Across (DCI) networks. While Marvell made numerous announcements, the main one we focus on here is the company’s co-packaged optics (CPO) and related topics.
Scale-Up.
The company set expectations for ESUN and UALink sampling from the Marvell portfolio, which generally will become available in late 2026 and 2027. Marvell’s portfolio of ESUN and UALink is an alternative to NVLink, NVIDIA’s scale-up technology. In one presentation, the company said its UA Link switch will be sampling in 2H26, while at other opportunities 2027 was mentioned. Marvell also emphasized it is partnering with NVIDIA to make custom NVLink “solutions.” 650 Group has been projecting scale-up in our AI forecast for over a year including the split of Ethernet (ESUN) and UALink.
Scale-Out.
Marvell made the point that the switch market is headed to CPO and emphasized numerous times that it will have a very complete portfolio that enables it to be a leader. The previously announced Celestial AI acquisition includes technology that will move to volume when its hyperscaler customer releases its next-gen architecture, according to statements from its CEO. Based on its roadmap slide, the company plans to make CPO sampling in 2027, which will represent a fundamental shift in the Ethernet Switch industry.
Scale-Across (DCI) 1.6 Tbps ZR/ZR+.
Marvell expects to be first to ship with a full feature set but declined to set a date for shipments. Generally, the company plans to offer semiconductors that enable 1.6Tbps 2nm ZR modules capable of transmitting at 1000km+ using QAM 16. In summary, the company says 1.6 Tbps is “coming soon.
Additionally, Marvell shared its views on numerous other initiatives such as custom XPUs, CXL (including its DDR4-reuse capabilities), and its advanced packaging initiatives, which we cannot address in a single blog post
Conclusion
Overall, Marvell plans to capture the largest networking blocks in what we expect to be a $200B/year DC Networking TAM by the end of the year. Networking remains the only market that is keeping pace with the GPU/XPU in the CAPEX expansion from Hyperscalers and Neo Clouds.