Marvell Technology, known for designing custom chips for cloud infrastructure and enterprise storage, is targeting a potentially larger opportunity in the connectivity linking those processors. KeyBanc analyst John Vinh estimates that scale-up networking could represent a $30 billion total addressable market by 2030, including approximately $20 billion in optical technologies and $10 billion in switching capabilities.

Vinh raised his price target for Marvell by 48% to $385, highlighting the growing importance of optical networking to the company’s business.

Marvell’s networking revenue outlook

Marvell now targets approximately $12 billion in revenue for fiscal year 2027 and $18 billion for fiscal year 2028. Together, those figures represent roughly $30 billion across the two fiscal years, compared with previous guidance of about $20 billion.

The $30 billion market estimate and Marvell’s revenue targets are different measures: the former describes potential industry spending on scale-up networking by 2030, while the latter describes the company’s own targets for two fiscal years. They therefore should not be read as equivalent forecasts.

The company’s data center interconnect segment is expected to post year-over-year growth of more than 70% in FY2027. As AI accelerators become more powerful, data movement between processors is becoming a larger constraint for AI infrastructure. Training a large language model across thousands of GPUs requires substantial communication between chips, while conventional electrical connections face physical limitations at higher performance levels.

Marvell is addressing the market through near-packaged optics and co-packaged optics. Near-packaged optics places optical components close to a processor while keeping them on a separate substrate. Co-packaged optics integrates optical transceivers directly into the chip package. Both approaches are designed to reduce latency and power consumption compared with pluggable optical modules positioned at the edge of a server rack.

Acquisitions and partnerships

Marvell has expanded its capabilities through acquisitions. In December 2025, the company acquired Celestial AI for approximately $3.25 billion, bringing its silicon photonics technology into Marvell. Silicon photonics uses standard semiconductor manufacturing processes to build optical components, potentially making them less expensive and easier to scale than traditional discrete optics.

Marvell also acquired XConn for roughly $540 million. The deal added chiplet interconnect technology to the company’s portfolio. Chiplets are smaller, modular pieces of silicon assembled into a larger package, and efficiently connecting them remains a significant challenge in chip design.

The company has expanded its partnership with NVIDIA to include custom silicon, optics, and NVLink. NVIDIA uses NVLink, its proprietary high-speed interconnect, to link GPUs in AI training clusters. Marvell also secured a multiyear supply agreement with Google, a major buyer of custom networking silicon.

Demand for optical connections

Major hyperscalers are building clusters containing tens of thousands of accelerators that must communicate at terabit-per-second speeds. Copper electrical connections lose performance over distance and generate significant heat at high data rates. Optical connections avoid those limitations over the distances typically involved inside a data center.

Marvell’s shares have risen more than 180% in 2026, following multiple analyst upgrades and the company’s improved financial outlook.