Marvell Expands Silicon Photonics Solutions for High-Performance AI Infrastructure, Supporting Bandwidth Up to 3.2T for GPU Interconnects

July 5, 2026

Marvell Newsroom USA

Overview

Marvell announced a significant expansion of its silicon photonics solutions portfolio to address the surging demands of AI and HPC data centers. The company’s new optical engines and transceivers will support bandwidths up to 3.2 terabits per second (Tbps) for GPU-to-GPU and intra-rack connectivity, while improving power efficiency by 25% compared to existing solutions. This enhancement is expected to boost overall system performance and efficiency for AI clusters, helping to resolve bottlenecks in next-generation AI workloads.

In Depth

Key Findings

Marvell has announced a substantial expansion of its silicon photonics solutions portfolio, designed to meet the explosive bandwidth requirements for GPU-to-GPU and intra-rack connectivity within AI and HPC data centers. The company’s latest product line supports data transmission speeds of up to 3.2 terabits per second (Tbps) while simultaneously improving power efficiency by 25% compared to existing solutions, thereby elevating AI infrastructure performance and sustainability to new levels.

Technical / Clinical Details

Marvell’s new silicon photonics platform features highly integrated optical engines and Digital Signal Processing (DSP) chips. This maximal integration of optical and electrical components enables high-speed data transmission at low power consumption.
The product portfolio includes optical transceivers supporting 800G and 1.6T, with a future roadmap outlined for 3.2T-capable solutions. These products utilize PAM4 modulation technology in conjunction with advanced Forward Error Correction (FEC) algorithms to ensure reliable data transmission.
Notably, these solutions are also compatible with in-package optics (such as Co-Packaged Optics or Near-Packaged Optics), designed for close proximity to GPUs and AI accelerators. This enables ultra-high-speed optical connections over short chip-to-chip distances, effectively resolving data movement bottlenecks.

Background & Context

The scaling of AI models and the increasing parallelism of processing have placed unprecedented demands on data center internal networks. Particularly in large-scale AI clusters, where thousands or even tens of thousands of GPUs operate cooperatively, fast and low-latency data transfer between GPUs dictates overall computational performance. Traditional copper cables and early optical modules have struggled to meet these demands due to issues of power consumption, thermal density, and signal attenuation. Silicon photonics, due to its inherent scalability and cost advantages, has emerged as the most promising technology to address these challenges.

Strategic Significance & Outlook

Marvell’s reinforced silicon photonics solutions are poised to play a critical role in the design and deployment of next-generation AI infrastructure. The company’s technology is expected to be adopted in high-performance AI accelerator systems and HPC clusters through collaborations with leading AI chip manufacturers and cloud service providers. This will likely push the boundaries of AI computational capabilities further, accelerating innovation across diverse fields such as autonomous driving, drug discovery, and scientific research.

Source: https://www.marvell.com/newsroom/XXXXXX/silicon-photonics-ai-infrastructure.html

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.