Marvell Acquires Celestial AI for $3 Billion to Secure CPO Core Technology, Expands Silicon Photonics Collaboration with NVIDIA

Global Semi Research USA

Overview

Marvell has strategically transformed into a key player in AI data center interconnect and custom chips by acquiring Celestial AI for approximately $3 billion in December 2025, gaining crucial Co-Packaged Optics (CPO) core technology. The company also expanded its collaboration with NVIDIA on DSP, TIA, driver chips, and silicon photonics technology for scale-up networks. This move aims to reduce power consumption by 50-60% through optical-electrical chip integration, as AI infrastructure shifts from compute-centric to networking-driven architectures.

In Depth

Key Findings

Marvell has undergone a strategic transformation, becoming a key player in AI data center interconnects and custom chips, following its acquisition of optical interconnect firm Celestial AI for approximately $3 billion in December 2025, thereby securing crucial Co-Packaged Optics (CPO) core technology. In addition to this acquisition, Marvell announced an expanded collaboration with NVIDIA, extending its supply relationship for DSP, TIA, and driver chips, and deepening cooperation on silicon photonics technology development for AI scale-up networks. These moves aim to reduce power consumption by 50-60% through the integration of optical and electrical chips, as AI infrastructure shifts from a compute-centric to a networking-driven architecture.

Technical / Clinical Details

Marvell’s strategic transformation focuses on the following technical elements:

• Co-Packaged Optics (CPO): The CPO technology acquired from Celestial AI dramatically shortens the transmission distance of electrical signals by integrating optical engines within the same package as ASICs like CPUs or GPUs. This allows for significant reductions in signal loss, latency, and power consumption, which is essential for meeting the high bandwidth and low latency requirements of AI workloads.
• Silicon Photonics: Collaboration with NVIDIA will accelerate the development of silicon photonics technology. Silicon photonics enables the integration of optical components using standard silicon manufacturing processes, reducing the cost of optical interconnects and facilitating mass production. This is a crucial factor for the widespread adoption of optical interconnects in AI data centers.
• DSP, TIA, and Driver Chips: Marvell will expand its supply of these key optical transceiver components to NVIDIA. High-performance DSP (Digital Signal Processing), TIA (Transimpedance Amplifier), and driver chips ensure accurate modulation and demodulation of optical signals, enhancing the overall performance and reliability of optical communication systems.

Through the combination of these technologies, Marvell aims to eliminate the “copper bottleneck” in AI data center connectivity and support next-generation AI computing platforms.

Background & Context

The explosive growth of AI is placing unprecedented demands on data center interconnects, with traditional copper-based connections reaching their limits in terms of power consumption and bandwidth. Major players like NVIDIA are making a strategic shift towards optical interconnects for scaling AI clusters, and Marvell’s acquisition of Celestial AI and expanded collaboration with NVIDIA reflect this industry-wide trend. NVIDIA’s investment of over $6.5 billion in photonics-related initiatives further underscores that optical technology will be a foundational layer for computing in the AI era. Marvell’s move indicates a shift in its business focus from traditional storage and networking solutions to becoming a key enabler of AI infrastructure.

Strategic Significance & Outlook

Marvell’s acquisition of Celestial AI and expanded collaboration with NVIDIA are expected to dramatically accelerate the adoption of co-packaged optics and silicon photonics in AI data centers. A 50-60% reduction in power consumption through integrated optical-electrical chips will not only significantly reduce operational costs and environmental impact for AI data centers but also dramatically improve the efficiency of AI model training and inference. Marvell’s expanded interconnect portfolio and silicon photonics acquisition are key drivers supporting the company’s significantly upward revised performance outlook for the coming years, and are expected to further accelerate the industry-wide transition to optical technology. This marks a critical step towards building sustainable and high-performance computing infrastructure for the AI era.