Marvell Technology, Inc. (NASDAQ: MRVL) has expanded its 1.6T optical digital signal processor (DSP) platform portfolio, strengthening its position in next-generation connectivity solutions for AI data centers. The expanded lineup introduces several new DSP platforms designed to improve power efficiency, reliability, security, and design flexibility as hyperscale infrastructure moves beyond 800G connectivity toward 1.6T networks.
The move builds on Marvell’s multi-generational DSP development roadmap. In 2023, the company introduced the 5nm Marvell Nova platform, which delivered 200G-per-lane 1.6T DSP capabilities. This was followed by the 3nm Ara platform in 2024, designed to improve performance while lowering the overall power envelope for high-speed optical modules. Ara is currently shipping in volume and is being deployed by hyperscale and cloud providers to support 1.6T pluggable connectivity in AI data centers.
The latest expansion of the 3nm 1.6T DSP portfolio focuses on optimizing performance per watt across several high-volume deployment scenarios. The portfolio additions include Ara T, an 8x200G transmit-retimed optics (TRO) DSP designed to improve power efficiency while lowering the total cost of ownership for optical network deployments. Another addition, Ara X, incorporates enhanced link reliability capabilities to improve resilience across high-speed optical connections.
Marvell has also introduced Petra, a 3nm gearbox device supporting 8x100G to 4x200G configurations. The component enables more flexible infrastructure designs while maintaining power efficiency across high-speed links. The fourth addition, Aquila M, is an O-band–optimized coherent-lite optical DSP that integrates media access control security (MACsec), providing built-in security capabilities for next-generation optical connectivity.
As AI infrastructure expands, connectivity has increasingly become a limiting factor in data center performance. Modern AI clusters often contain hundreds of thousands of GPUs, XPUs, and other advanced compute engines, all requiring high-speed communication across optical networks. In this environment, efficient interconnect technologies are essential to ensure that compute resources can operate at peak performance.
“Marvell pioneered PAM DSP technology, and we continue to lead with advanced SerDes and production-proven 800G platforms,” said Xi Wang, senior vice president and general manager of Marvell’s Connectivity Business Unit. “We are extending that multi-generational product leadership into the 1.6T era to support the rapid growth of next-generation AI data centers.”
Industry analysts point to the growing importance of high-performance interconnect technologies in enabling large-scale AI systems. Vladimir Kozlov, founder and CEO of LightCounting, noted that the performance of modern data centers depends heavily on the optical links connecting compute resources. According to Kozlov, DSP technologies play a key role in enabling high-speed data transmission while maintaining power efficiency and reliability.
Marvell’s broader connectivity portfolio spans the full AI data center network stack, including DSPs, SerDes, switching technologies, interconnect components, drivers, and transimpedance amplifiers (TIAs). The company also offers system-level monitoring through the Marvell RELIANT interconnect telemetry platform, designed to help operators reduce operational complexity while improving network performance and reliability.
The company’s 1.6T connectivity roadmap also includes additional technologies such as Alaska® and Nova DSP platforms, Ethernet PHY solutions, Silicon Photonics Light Engines, and linear pluggable optics (LPO) chipsets. Together, these technologies support scale-up, scale-out, and scale-across AI infrastructure architectures.
The newly introduced Ara X, Ara T, Petra, and Aquila M DSP platforms are currently sampling with customers, with broader deployments expected as data center operators transition to 1.6T optical connectivity to support the next generation of AI workloads.
