Every FLOP left idle and every bit of memory left stranded is CapEx that isn't generating return. In this clip from AI Infra Summit 2026, Marvell Senior Vice President of the Photonic Fabric Business Unit Preet Virk makes the case for why connectivity has become the key to unlocking AI data center efficiency. Watch the full conversation with Preet and Senior Director of Product Management Ravi Mahatme at NextGenInfra: mrvl.co/4AsnV0p

Structera X CXL memory expansion is now jointly validated with the NVIDIA Vera CPU platform, confirming consistent CXL 2.0 compliance and full feature support across device enumeration, resets, data exchange, and firmware update. With validated interoperability now spanning several partners, Structera X is the most broadly validated CXL memory expansion solution in the industry. For cloud operators and OEMs managing heterogeneous server fleets, that breadth means deploying CXL memory expansion without re-qualifying for each new platform. Marvell's Arifur Rahman details what was tested, what was confirmed, and what it means for customers building CXL-enabled AI infrastructure at scale: mrvl.co/4yeFpfm

Converge Network Digest covers Marvell's ECOC 2026 demonstrations in depth, with particular attention to what the move to 400G per lane means for the path to 3.2T optical connectivity. The article walks through all five demonstration categories, from 2nm PAM4 and coherent technologies to the 102.4 Tbps CPO platform, and frames why Marvell is applying 2nm across PAM4, ZR/ZR+, and coherent-lite simultaneously rather than treating it as a single-product node transition. It also notes the breadth of AI connectivity domains covered: scale-out, scale-up, scale-across, and shorter-reach campus interconnects. Read the full article: mrvl.co/4iKHTNG

At @ECOC_Exhibition, Marvell is running a live demonstration of the industry's first 1.6T ZR/ZR+ OSFP pluggable module with integrated MACsec encryption, powered by a 2nm coherent DSP. Live 4K video traffic travels across an encrypted optical link between two modules, with MACsec toggled on and off in real time to visually confirm encryption is active. The demonstration at Booth 2002 shows what wire-speed security looks like when it is built directly into the module. #ECOC26 #ECOCExhibition

Marvell delivered a dynamic presence at AI Infra Summit 2026, showcasing its comprehensive portfolio of AI data center connectivity and memory solutions across four speaking sessions and a full booth of live demonstrations. The Marvell Teralynx T100, the industry's first 102 Tbps Ethernet switch purpose-built for AI, was a centerpiece of the booth, making its first live public appearance since its announcement at Computex in June. Alongside it, demonstrations spanned Photonic Fabric technology for multi-rack optical memory sharing, the RELIANT interconnect telemetry platform, PCIe-based scale-up switching, CXL memory expansion and near-memory acceleration, and high-speed storage solutions. Among four Marvell presentations, Executive Vice President and General Manager of the Data Center Networking Business Group Dave Lazovsky delivered a keynote on AI inference at scale, underscoring the message that ran throughout the show: connectivity now sits at the center of AI infrastructure performance, shaping bandwidth, latency, power consumption, and the economics of deploying AI at scale. #AIInfraSummit

Marvell is demonstrating its co-packaged optics platform at @ECOC_Exhibition, including a 102.4 Tbps capable CPO implementation built on class-leading silicon photonics. The physics driving CPO adoption are straightforward. As per-lane speeds in scale-up fabrics increase, the distance over which signals can travel over passive copper falls. AI pods spanning multiple racks require longer interconnect distances than copper can support. CPO brings optical elements directly into the package alongside the IC, reducing distance, improving performance, and lowering power consumption. The demonstration at booth 2002 showcases 200G per lane with aggregate bandwidth in multiples of 6.4 Tbps, alongside detachable ELSFPs for rapid laser replacement and simplified manufacturability. #ECOC26 #ECOCExhibition

Memory capacity has become one of the defining constraints on AI infrastructure performance. Marvell is demonstrating a solution at @ECOC_Exhibition that approaches the problem from a different direction, and new category. The Photonic Fabric™ technology demonstration at Booth 2002 shows optical interconnects which can create shared memory pools across multiple racks, enabling compute and memory to scale independently. #ECOC26 #ECOCExhibition

When memory leaves the server and rides a switched fabric, is it still memory, or has it become a network? That question drove MemNetAI, the first workshop on Memory-Semantic Networking for AI-Scale Systems, launched by Marvell with researchers from IIT Hyderabad and IIIT Delhi at ACM SIGCOMM 2026 in Denver. The workshop brought together speakers from Cornell and industry experts to address a convergence that AI infrastructure is forcing: keeping accelerators fed with data is as much a networking problem as one of memory. Talks covered CXL memory expansion, GPU memory tracing, photonic scale-up interconnects, and memory disaggregation, with a closing panel debating whether the boundary between memory and network still means anything at the bandwidth and latency levels AI systems demand. Director of Custom Cloud Solutions Abed Mohammad Kamaluddin recaps the day's key themes and why they matter for the future of AI infrastructure: mrvl.co/3TK5ns0

Stop by Booth 2002 at @ECOC_Exhibition to witness one of the core building blocks of next-generation AI networking. Marvell's multi-protocol 224G LR SerDes IP block built on 2nm silicon platform, with high-speed data transmission over a 2.5-meter copper connection, enables 200G-per-lane connectivity, a critical step towards 1.6Tbps Ethernet, which helps AI and cloud infrastructure to move significantly more data with greater efficiency, paving the way for next-gen networking and the data center architectures that come with it. #ECOC26 #ECOCExhibition

At @ECOC_Exhibition, Marvell is showcasing an Ara T™ 1.6 T TRO PAM4 DSP optical connectivity solution, which is for short-reach AI and cloud optical links, built for the next generation of AI and hyperscale networks. The Marvell Ara TRO demonstration highlights the bandwidth, efficiency, and scalability required as the industry transitions to 200G-per-lane infrastructure. While a traditional PAM4 optical module typically has a DSP on the transmit (TX) side, and a DSP on the receive (RX) side, in a TRO architecture the DSP is retained only on the transmit path, and the receive path removes the receive DSP and uses a simpler TIA/front-end. Stop by Booth 2002 to see it in person. Learn more: mrvl.co/4iIJjIy

Network architecture has become one of the defining constraints on AI factory performance, scalability, power efficiency, and cost. Building infrastructure that supports hundreds of thousands of accelerators requires balancing scale-up, scale-out, and scale-across strategies while navigating a rapidly evolving technology landscape spanning CPO, LPO, silicon photonics, and next-generation Ethernet. At @ECOC_Exhibition, Marvell Vice President and General Manager of the Data Center Switch Business Rishi Chugh will join an Endeavor panel alongside leaders from Nokia and Lumentum to discuss how tomorrow's AI fabrics are being designed today, which technologies are ready for deployment, where the biggest bottlenecks remain, and what architectural decisions organizations should be making now. Learn more: mrvl.co/4rldibR

Marvell is showcasing three industry-first 2nm optical demonstrations at ECOC 2026, advancing the next generation of AI data center connectivity. These include the first 2nm 400G/lane optical PAM4 demo; the first live demonstration of 2nm 800G ZR/ZR+ with integrated MACsec encryption, powered by the Marvell Libra coherent DSP; and the first 2nm 1.6T ZR and coherent-lite O-band demonstrations, paving the way for 3.2T connectivity within and between data center facilities. Overall, Marvell and its ecosystem partners will showcase 38 advanced technology demonstrations, including the Marvell 102.4T co-packaged optics platform. Read the full press release: mrvl.co/4jg6bPD

Photonics has moved from a supporting role in data center connectivity to a central one, and the ecosystem forming around it is still taking shape. At ECOC, Marvell Senior Vice President and General Manager of the Connectivity Business Unit Xi Wang will join a fireside chat with other industry leaders to discuss the growing role of photonics in next-generation AI and machine learning interconnects, the challenges and opportunities that come with it, and the emerging ecosystem being built around these technologies. Learn more: mrvl.co/4rldibR

The AI infrastructure opportunity doesn’t stop at the XPU. @MarvellTech’s SVP Will Chu says “the networking, memory, storage, and security around a chip means 3 - 4 more “XPU attach” design wins per custom XPU. Dave Lazovsky (fmr. CEO of Celestial.AI) and now SVP, says reasoning models are why: KV cache is up 10x in 9 months, pushing scale-up past a single rack. Cumulative data center capex forecasts through 2030 climbed from $10.7 trillion to $12 trillion within a single quarter. Chu and Lazovsky’s combined perspectives make up the system Marvell is counting on to capture that spend. $MRVL

The shift from chassis-based coherent transport to compact pluggable optics is reshaping how operators scale metro, regional, long-haul, and data center interconnect networks. Lower cost per bit, improved power efficiency, and network disaggregation are accelerating the transition, and the technology roadmap continues to advance rapidly. At @ECOC_Exhibition, Marvell Senior Vice President and General Manager of Data Center Interconnect Russ Esmacher will present on advances in coherent DSPs, photonic integration, and pluggable form factors supporting 400G, 800G, and higher-capacity services, with a look ahead at 800G and 1.6T ZR/ZR+ solutions and tighter optical network integration. Learn more: mrvl.co/4rldibR

The choice of optical interconnect architecture has significant implications for bandwidth, power, cost, and deployment flexibility as GPU clusters grow in size and complexity. Pluggable optics offer ecosystem maturity and operational simplicity. Co-packaged optics offer bandwidth density and power efficiency. Near-packaged optics sits between the two. At @ECOC_Exhibition, Marvell Vice President of Connectivity Lian Qin will present a comparative analysis of all three approaches, examining performance, power consumption, serviceability, ecosystem readiness, and total cost of ownership, and exploring how NPO addresses the evolving requirements of AI scale-up infrastructure. Learn more: mrvl.co/4rldibR

Payment security has resisted the move to the cloud longer than any other cryptographic workload. The reason is not the hardware. It is decades of infrastructure built around proprietary payment software that had no cloud equivalent, leaving banks and payment processors colocating physical HSM appliances inside hyperscaler data centers as a workaround. That gap is closing. Azure Cloud HSM v2, built jointly by Marvell, @Microsoft, and @Utimaco, brings the Atalla payment software, the same platform that has underpinned PIN processing and payment cryptography since the 1970s, onto Marvell LiquidSecurity 2 HSM hardware already deployed across hyperscaler infrastructure. Existing applications require no code changes. Customers gain elastic capacity instead of provisioning for peak. Director of Security Sales Eyal Worthalter details the history behind payment HSMs, why payments lagged behind the rest of the cryptography world, and what this joint offering represents for banks, payment processors, and financial technology companies. Learn more: mrvl.co/4yNQ9Bn

At Booth 704 at AI Infra Summit 2026, Marvell is demonstrating the Structera CXL portfolio across three memory use cases. Structera X expands DDR5 memory capacity up to 8 TB per controller, enabling infrastructure to scale memory independently from compute. Structera A near-memory accelerators bring processing directly to where data resides, reducing data movement across the fabric by 60 times and improving XPU efficiency for KV cache, vector embeddings, and graph data. A third demonstration shows Structera A powering a 10 billion-scale vector database in a Kubernetes cluster, purpose-built for retrieval-intensive AI workloads. #AIInfraSummit

We are expanding our multi-year agreement with @MarvellTech to add SiGe capacity in Burlington, Vermont for next-generation optical connectivity in AI data centers. Read more: bit.ly/4yJhOmN

Scaling AI infrastructure requires different connectivity technologies at every distance, from within the rack to across the data center. Marvell is demonstrating that full range at AI Infra Summit Booth 704. The demonstration includes Marvell-powered active electrical cables, active optical cables, and PCIe retimer cables, each optimized for different combinations of bandwidth, power, latency, cost, and reach. Alongside those, Marvell is showcasing the Structera S 60260, the industry's highest radix PCIe 6 switch with 256 lanes of data traffic, bringing an open, standards-based option to scale-up fabrics that today rely on either PCIe or NVLink. Learn how to Expand memory resources through Marvell PCIe/CXL end-to-end platform, with @intel Xeon® 6 processors with P-Cores.