As artificial intelligence (AI) data centers rapidly multiply, the importance of optical communication is growing. The spread of AI inference is sharply increasing the volume of data moving between graphics processing units (GPUs). As data center scale expands, conventional copper wire connections reveal limitations in speed and power efficiency. The burden of heat generation and signal loss also grows. Optical connections, by contrast, enable faster and more efficient data transmission, expanding their range of application.
The company drawing attention in this process is Marvell Technology. Marvell supplies optical connectivity chips that transmit data at ultra-high speeds inside AI data centers. If GPUs handle computation, Marvell is the company responsible for the network that ensures data moves between GPUs without bottlenecks. It holds particular strength in the scale-out segment, which refers to connections between servers and server racks. In its latest earnings release, the company raised its growth outlook for the interconnect business from the previous 50% to more than 70%. This is a signal that demand for AI-related data movement is growing faster than expected.
The new growth driver is entry into new markets. Marvell has recently expanded its business scope beyond server-to-server connections into the scale-up segment, which refers to the network inside GPUs. The company expanded its business areas by acquiring Celestial AI, which increases connection speeds between GPUs, and Xconn, a switch technology company that controls data flow between GPUs. In addition, by expanding the scope of its collaboration with Nvidia, it is entering the overall data movement market inside AI infrastructure.
Product adoption is likely to lead to improved profitability. Currently, optical communication equipment centered on 800G (gigabit-class) is rapidly spreading in AI data centers. Recently, the adoption of the next-generation 1.6T (terabit-class) standard has also begun in earnest. The more demand grows for high-value-added products supporting higher bandwidth and speed, the greater the expected effect of product mix improvement. Marvell is expected to widen its profitability improvement by expanding supply of next-generation optical components and network solutions.
Earnings were also solid. First-quarter revenue for fiscal year 2027 (February–May 2026) was $2.42 billion, up 28% from the same period a year earlier. Data center revenue reached a record high of $1.83 billion. The company presented annual revenue guidance of $11.5 billion for fiscal year 2027, a level 40% higher than the previous year. It projected fiscal year 2028 revenue of $16.5 billion, a 45% increase. As figures significantly exceeding market expectations from just a few quarters earlier were presented, expectations for earnings growth are also rising.
However, the fact that AI infrastructure investment can be affected by specific large customers and the GPU investment cycle is a variable. Although demand for high-performance optical communication is growing rapidly, short-term earnings volatility could increase if the pace of data center investment slows. Nevertheless, considering that the bottleneck in AI data centers is shifting from computing performance to data movement, there is room for Marvell to be evaluated as a key beneficiary of optical communication infrastructure expansion.
