Marvell Technology

Marvell Senior Vice President of Foundry and Advanced Packaging Dr. Hamid Azimi has received the 2026 Hall of Fame award from the International Semiconductor Industry Group, recognizing lifetime contributions that have significantly advanced the semiconductor industry. Over a career spanning more than 30 years, Dr. Azimi has led teams responsible for some of the industry's most consequential advances in packaging technology, including the first flip-chip packaging with ABF substrates, EMIB technology from development to high-volume manufacturing, and pioneering work in glass substrates. He holds over 40 patents. The article traces his career and makes the case for why advanced packaging has become a critical performance driver in the AI era, determining whether next-generation AI systems can operate at full capability. Read the full article here: mrvl.co/4u2gR73

Senior Vice President and General Manager of Data Center Interconnect Russ Esmacher will join the "AI Networking and Chips" panel at the Bloomberg Generative AI Forum. Russ will speak to the expanding role of optics in AI networking. While copper remains the primary driver in scale-up applications, IMDD/PAM4, CPO and Coherent optics are seeing significant adoption in scale-out and scale-across architectures, driven by two fundamentals: bandwidth and power. Learn more: mrvl.co/4wmXmYD

Interconnect health is increasingly central to AI infrastructure performance. As data center fabrics scale in speed and reach, live visibility into those networks becomes operationally essential. Marvell RELIANT is a new software platform that delivers real-time intelligence across scale-up, scale-out, and scale-across networks. Built on over a decade of interconnect software development, it enables data center operators to monitor link health, run diagnostics, and address issues before they cause downtime. Vice President of Software Engineering Todd Rope and UI Architect Wesley Yeh detail how RELIANT works and what it makes possible at scale: mrvl.co/4uCzZJ0

AI infrastructure is only as strong as the network connecting it. As clusters scale to tens of thousands of accelerators, XPU-to-XPU communication becomes a defining constraint. In a new article, Senior Vice President and General Manager of the Photonic Fabric Business Unit Preet Virk outlines how Marvell delivers end-to-end connectivity, from co-packaged copper and optical chiplets to a switch portfolio spanning PCIe, UALink, ESUN, and NVLink-compatible solutions. Marvell connects compute end to end—from inside the package to across data centers. That’s how AI scales. Read more: mrvl.co/4cUbHE8

As AI infrastructure scales across greater distances, securing data in motion is becoming as critical as the bandwidth carrying it. In a new blog, Joseph Chon explains why integrating MACsec directly into coherent DSPs inside ZR/ZR+ pluggable modules represents a meaningful shift in how scale-across networks are secured. Compared to embedding MACsec in switch ASICs or placing dedicated security devices on circuit boards, module-based MACsec reduces cost, complexity and deployment risk while making upgrades significantly easier. Marvell has already introduced several industry firsts enabling this approach, including the Electra 2nm 1.6T coherent DSP, the Libra 800G coherent DSP and the Aquila M coherent-lite DSP, all with integrated MACsec. By 2030, pluggables are projected to account for approximately 85% of scale-across ports shipped. Read the full article: mrvl.co/4umGGOU

AI infrastructure is not built on a single architectural choice. It is built on flexibility. In this conversation with @TheSixFiveMedia, Marvell President and COO Chris Koopmans explains why the future of data center design will be driven by “and,” not “or.” From co‑packaged optics and pluggables to multiple modulation technologies and diverse compute architectures, different approaches will coexist to meet rapidly evolving AI workloads. Koopmans shares how Marvell is investing across the full spectrum of optical and high-speed connectivity technologies to help customers scale AI infrastructure across a range of architectures and deployment models. Watch the video for perspectives on why multiple paths to scale will define the next era of AI infrastructure: mrvl.co/4txZvOF

Memory is the second largest semiconductor cost in the data center, and the gap between what AI infrastructure needs and what conventional architectures can deliver is widening fast. Jianping Jiang explains how the Structera S family of CXL switches addresses the AI memory wall by enabling terabytes of shareable, dynamically allocated memory across CPUs, GPUs and XPUs. The article covers real-world performance results, including a 4.8x improvement in inference throughput and an 82.7% reduction in time to first token, and introduces Structera S 30260, which supports up to 48TB of shared memory and 4TB per second of cumulative bandwidth. It also outlines how Structera S fits within a broader CXL portfolio spanning memory expansion, acceleration and pooling, the only such complete portfolio in the industry. Read the full article: mrvl.co/4t4CYsk

As AI reshapes the data center market, scale and focus have become critical. In this conversation with @TheSixFiveMedia, Marvell President and COO Chris Koopmans discusses how the company is evolving its strategy to support rapid growth while staying disciplined about where to invest. He explains how Marvell is prioritizing data center connectivity opportunities and aligning its engineering talent, customer relationships, and supply chain capacity to address the expanding AI infrastructure market. See the conversation here: mrvl.co/4cYEhmq

As AI data center infrastructure evolves from isolated facilities to multi‑site campuses and regional AI clusters, the need for high‑capacity, energy‑efficient data center interconnect (DCI) to move terabits of data between locations continues to grow. The industry is shifting toward a more scalable model: coherent pluggable optics integrated directly into IP switches and routers. This approach reduces system complexity, improves power efficiency, and enables hyperscalers to scale their networks faster. Marvell is leading this transition with the industry’s first 1.6T and 800G coherent DSPs, built on 2 nm technology to deliver the highest‑density pluggable modules. These solutions integrate on‑chip MACsec, providing secure, high‑capacity connectivity optimized for regional AI data centers. In this demonstration, Joseph Chon discusses how the COLORZ 800 module is operating in both 800G ZR and ZR+ modes. As AI clusters expand across campuses and regions, network architecture must evolve. Marvell develops coherent pluggable solutions enabling terabit‑class connectivity directly from standard IP infrastructure—delivering the scalable, energy‑efficient DCI required for the next generation of AI data center networks. See the video here: mrvl.co/4tCH4ch

Next‑generation AI networking fabrics and high‑performance data center interconnects demand extreme bandwidth with lower power and smaller footprints. The Marvell® Ara T™ 1.6T PAM4 DSP with Transmit Retimed Optics (TRO) is designed to deliver significant power and area savings for optical module interconnects. This demonstration by Rittik Shah shows Ara T operating in TRO mode, optimizing optical connectivity while simplifying module design. Built on 3 nm process technology, Ara‑T is the industry’s first 1.6T PAM4 DSP, enabling scalable, energy‑efficient optical links for next‑generation AI infrastructure.

. @MarvellTech Unlocks the Physics Layer: Mastering Plasmonic Integration to Pace Set the 6.4T AI Connectivity Era - Acquisition of @PolaritonTech establishes pivot toward plasmonics to scale bandwidth toward 6.4T shorturl.at/4sGsV @StevenDickens3

AI is rewriting the rules of data center networking, and the switch architectures powering it must evolve to keep pace. Rishi Chugh of Marvell sat down with Roy Chua of @AvidThink to discuss how AI workloads are driving a fundamental shift in network design, from scale out switching and the Ultra Ethernet Consortium to scale up fabrics and protocols like UAlink and ESUN. Rishi also shares a preview of what's ahead on the Teralynx roadmap, including 100T and 200T fabric devices targeted for the first half of 2027. Watch the full interview: mrvl.co/3QOiNRT

The submission deadline for MemNetAI workshop papers has been extended to May 11, giving researchers more time to contribute to an important industry conversation. Marvell is part of the organizing team for MemNetAI, the first workshop on Memory-Semantic Networking for AI-Scale Systems, co-located with ACM SIGCOMM 2026. The workshop brings together researchers and practitioners to explore networking principles for memory-semantic fabrics, spanning scale-up fabric interaction, disaggregated and pooled memory architectures, cross-tier congestion and resource management, simulation and benchmarking for AI-scale systems. Original research papers of up to six pages and extended abstracts of up to two pages are welcome. Submit a paper and help shape this emerging field: mrvl.co/4mlkaDs

Marvell has been named to the LinkedIn Top Companies list for 2026 in India — recognition rooted in data, not self-reporting. LinkedIn's methodology evaluates companies across eight pillars that have been shown to drive career progression, including ability to advance, skills growth, company stability, gender diversity and educational background. The ranking is built entirely from LinkedIn member data, reflecting real career outcomes across more than a year of activity. It is a meaningful signal that we are building an environment where careers genuinely move forward. Read more about the companies honored: mrvl.co/4cQbzEf #LinkedInTopCompanies

As AI models evolve, connectivity demands are accelerating faster than compute. In this conversation with @TheSixFiveMedia, Marvell COO Chris Koopmans discusses how rising hyperscaler capital investment, expanding cluster sizes, and new AI workloads are driving sustained growth in high speed connectivity. He also outlines why optics are becoming a critical enabler as bandwidth requirements continue to scale across and within data centers. Watch the full discussion to learn about the market forces shaping next generation AI infrastructure: mrvl.co/4cJhpqX

Last call for MOSAIC ’26 submissions. The future of AI infrastructure is being re-architected, and the biggest challenges are no longer just about compute. Memory, interconnects, and system design are defining what comes next. If you’re working on optical interconnects, CXL-based memory, rack-scale systems, or hardware-software co-design, Marvell wants to hear from you. Submissions are due May 8. Submit your paper and help shape the future of AI infrastructure: mrvl.co/4doecze

This Earth Day, Marvell is being recognized in the America’s Climate Leaders 2026 award list. This marks the second consecutive year Marvell has have received this distinction, reflecting continued progress in reducing the Marvell corporate carbon footprint, procuring renewable energy, and making progress towards sustainability targets. The recognition builds on other recent honors for ethical leadership and responsible business practices, as Marvell continues to embed sustainability into its operations, supply chain, product design, and power efficiency improvements across the portfolio. Learn more about the approach Marvell takes for sustainability and environmental stewardship: mrvl.co/4u4MAUR

An AI data center simulation at OFC recently validated 800G/1.6T Ethernet fabrics using realistic AI traffic, congestion patterns, and UEC features—supporting pre‑deployment testing for vendors and hyperscalers. Ashwin Gopalan from Marvell and Avik Bhattacharya from @Keysight discussed how AresONE‑800GE emulates switches, NICs, and AI workloads to verify high‑speed Ethernet performance. Marvell’s UEC‑Ready data center switch (Teralynx 10 scale out fabric) enables low‑latency, GPU‑to‑GPU communication with Ethernet optimized for AI. Keysight ensures Marvell platforms (switch silicon, NICs, DPUs, PHYs) meet UEC specifications, while Marvell delivers the high‑speed, AI‑optimized Ethernet infrastructure. Together, they validate that AI workloads run efficiently and scale reliably. See the video: mrvl.co/48a7VDX

The infrastructure decisions being made today will determine how well enterprises scale AI tomorrow. Marvell will be at AI Networking Summit 2026 with two sessions examining the networking and infrastructure challenges at the center of enterprise AI deployment. On May 13, Jeremy Moschner presents on building AI-aware networks, and on May 14, Prathyaya Bhandarkar joins a panel on breaking through the barriers keeping AI initiatives from reaching production. Learn more about the event: mrvl.co/42c9EoH

For AI/ML training clusters, the Lumentum R300 OCS enables optical connectivity across the rack, allowing hyperscale operators to replace or bypass traditional spine switches. This creates dynamic, high‑bandwidth paths for GPU‑to‑GPU communication in large‑scale AI systems while significantly reducing network power consumption. The solution integrates seamlessly with the Marvell optical connectivity portfolio and the RELIANT™ interconnect telemetry platform, providing scalable, efficient, and intelligent network fabric control for next‑generation GPU clusters.