OFC 2026 drew nearly 18,000 attendees from 91 countries to the Los Angeles Convention Center, marking one of the largest and most commercially significant gatherings in the event’s history. The exhibition floor sold out with 706 exhibitors across 210,346 net square feet, reflecting strong global demand for optical networking technologies as AI infrastructure scales. Across five days of announcements, technical sessions, and live demonstrations, the event showcased an industry moving decisively from innovation to deployment.

The central theme was unmistakable: AI infrastructure is now setting the agenda for optical networking. Momentum around 224G lane speeds, early 448G development, and 1.6T system commercialization highlighted the rapid acceleration in performance requirements. At the same time, new architectural approaches—including co-packaged optics, optical I/O, optical circuit switching, and alternative light-source technologies—point to a fundamental redesign of how networks are built inside AI clusters. The industry focus has shifted from individual components to system-level performance, interoperability, and energy efficiency at scale.

A strong forward signal emerged during the week: OFC 2027 exhibit space is already 73% sold, underscoring continued investment and urgency across the optical ecosystem as vendors and hyperscalers align around the next generation of AI infrastructure.

New MSAs Signal Industry Alignment Around AI Interconnects

SDM4 MCF (Multicore Fiber) MSA

The SDM4 MSA introduces a four-core multicore fiber architecture designed to multiply fiber capacity without increasing physical footprint, addressing one of the most pressing constraints in AI data center design. By enabling parallel transmission paths within a single fiber strand, the approach significantly improves bandwidth density while reducing cabling complexity and bulk. This is particularly relevant as hyperscale operators confront exponential growth in east-west traffic across GPU clusters. Beyond immediate benefits, SDM4 represents a longer-term shift toward spatial division multiplexing as a foundational scaling mechanism for optical networks.

Open CPX (Co-Packaged Optics) MSA

The Open CPX initiative focuses on standardizing co-packaged optics architectures, where optical engines are integrated directly alongside switch ASICs to reduce power consumption and latency. This approach eliminates long electrical traces, which are increasingly inefficient at higher speeds such as 224G and beyond. The MSA aims to define interoperable interfaces and system architectures across vendors, helping accelerate adoption in hyperscale environments. Open CPX reflects a broader industry consensus that traditional pluggable optics alone will not meet future AI scaling requirements.

XPO (External Laser Optics) MSA

The XPO MSA promotes architectures that separate laser sources from switching silicon, enabling improved thermal management and operational flexibility. By moving lasers out of high-heat environments, XPO designs enhance reliability while also simplifying maintenance and replacement. This modular approach also allows operators to optimize laser performance independently of switching hardware. As co-packaged optics evolves, XPO represents a pragmatic compromise that balances integration benefits with real-world operational requirements.

OCI (Optical Compute Interconnect) MSA

The OCI MSA defines optical interconnect standards for tightly coupled compute environments, particularly within AI scale-up fabrics connecting GPUs and accelerators. Unlike traditional data center networking, these environments demand ultra-low latency and extremely high bandwidth between compute nodes. OCI aims to standardize how optical links are deployed within these domains, enabling interoperability and faster ecosystem development. This initiative signals a major shift toward optical fabrics becoming native to compute architectures rather than just connecting racks.

Low-Power Copper for 800G MSA

This initiative highlights continued innovation in electrical interconnects, focusing on ultra-low-power copper solutions for short-reach 800G links within racks. While optics dominates longer distances, optimized copper remains critical for cost and efficiency at very short ranges. The MSA aims to standardize performance and power targets, ensuring compatibility across vendors. This reflects a pragmatic approach to AI infrastructure design, where multiple interconnect technologies coexist depending on distance and application.

Optica Executive Forum Keynotes (Full Coverage)

Microsoft: AI Infrastructure and Network Architecture Evolution

Microsoft outlined the rapid evolution of AI infrastructure and emphasized that network design is increasingly the limiting factor in system performance. As AI workloads become more distributed and data-intensive, the balance between compute, memory, and interconnect becomes critical. The company highlighted the need for scalable, high-bandwidth networking architectures that can support dynamic workloads across large data center environments. Optical technologies were positioned as essential for enabling this scale while maintaining efficiency and reliability.

Marvell (Radha Nagarajan): Optical Interconnect Roadmap for AI

Marvell’s Radha Nagarajan presented a detailed roadmap for next-generation optical interconnects, focusing on the transition to higher lane speeds and tighter integration. The discussion highlighted challenges related to packaging, signal integrity, and power consumption as systems scale toward 1.6T and beyond. Nagarajan emphasized the importance of co-design across silicon, photonics, and system architecture to achieve optimal performance. The roadmap reflects the complexity of scaling optical technologies in parallel with AI compute demands.

Marvell (Dave L.): Optical-First AI Infrastructure Vision

Marvell described how AI infrastructure bottlenecks have shifted from compute to data movement, with GPU utilization often limited by memory and interconnect performance. The company argued that optical fabrics are essential for unlocking higher efficiency and enabling next-generation AI systems. By adopting an optical-first approach, operators can reduce latency, improve bandwidth, and lower power consumption. This vision underscores the growing importance of photonics in system-level design.

Broadcom: Optical Scale-Up Networks

Broadcom presented its vision for optical scale-up architectures designed for tightly coupled GPU clusters. These systems require extremely high bandwidth and low latency, which optical interconnects are uniquely positioned to provide. The company emphasized the importance of ecosystem collaboration and emerging MSAs to ensure interoperability. The presentation highlighted how optical fabrics are becoming central to AI infrastructure design.

OpenAI: Infrastructure Requirements for Frontier Models

OpenAI discussed the infrastructure demands of training and deploying large-scale AI models, emphasizing exponential growth in data movement. As models become more complex, the need for high-bandwidth, low-latency interconnects becomes critical. The organization highlighted the role of optical networking in enabling scalable and efficient AI systems. These requirements are driving innovation across the optical ecosystem.

Meta: Scaling AI Data Center Networks

Meta outlined its approach to scaling AI infrastructure across global data centers, focusing on efficiency, reliability, and performance. The company emphasized disaggregated architectures and high-performance interconnects as key design principles. Optical technologies were identified as essential for supporting the growing demands of AI workloads. Meta’s strategy reflects broader industry trends toward scalable, modular infrastructure.

Read more: https://convergedigest.com/optica-executive-forum-meta-ai-networking/

Arista Networks: High-Performance AI Networking Architectures

Arista discussed the evolution of Ethernet-based networking to support AI workloads, emphasizing scalability and low latency. The company highlighted the importance of integrating optical technologies into switching platforms. Operational simplicity and reliability were also key themes, reflecting the needs of hyperscale environments. Arista’s approach underscores the convergence of Ethernet and optical innovation.

Major News from the OFC 2026 Show Floor

Broadcom expands AI networking stack with scale-up fabrics, 400G/lane optics, and next-gen switch silicon

Broadcom continued to build out its end-to-end AI infrastructure portfolio with a focus on tightly integrated switching, optical interconnect, and scale-up fabrics. The company has been advancing its roadmap toward 400G-per-lane signaling and next-generation switch architectures, including platforms aligned with emerging optical scale-up requirements for GPU clusters. Recent activity also highlights Broadcom’s role in driving ecosystem alignment through initiatives such as OCI and co-packaged optics efforts, where interoperability and system-level performance are critical. By combining switching silicon, optical DSPs, and connectivity technologies, Broadcom is positioning itself as a central supplier for hyperscale AI networks, where bandwidth density, latency, and power efficiency must all scale simultaneously.

Broadcom doubles down on Tomahawk 6 and 400G/lane optics to challenge NVIDIA AI fabric dominance

Broadcom continues to extend its leadership in data center switching with its Tomahawk 6 platform, designed to support next-generation AI workloads with higher bandwidth density and improved energy efficiency. The company is aligning its switching roadmap with 400G-per-lane optical technologies, enabling 1.6T and future 3.2T interconnects that are critical for scaling AI clusters. This approach positions Broadcom as a key alternative to vertically integrated AI networking stacks, particularly in environments where open Ethernet-based fabrics are preferred. By combining high-performance switching silicon with advanced optical connectivity and ecosystem-driven standards, Broadcom is strengthening its position against competing AI fabric solutions and reinforcing its role in hyperscale deployments.

Lightmatter demos 1.6 Tbps per fiber optical interconnect

Lightmatter demonstrated a DWDM-based optical interconnect delivering 1.6 Tbps per fiber using multiple wavelengths and 112G PAM4 signaling. This approach significantly reduces fiber count while increasing bandwidth density, addressing key challenges in scaling AI clusters. By leveraging wavelength multiplexing, the architecture enables more efficient use of physical infrastructure. The demonstration highlights a broader industry shift toward optical fabrics optimized for high-performance computing environments.

Lightmatter unveils 6.4 Tbps BiDi optical engine

The company introduced a bidirectional optical engine designed to maximize bandwidth per fiber through simultaneous transmit and receive operation. This innovation improves port density and reduces overall infrastructure requirements. It also enhances efficiency in dense AI deployments where space and power are constrained. The development reflects ongoing efforts to optimize optical interconnect performance at scale.

Salience Labs launches 32-port all-optical switch

Salience Labs introduced an all-optical switch designed to support dynamic AI workloads with lower latency and higher throughput. By eliminating electronic switching components, the platform improves performance and scalability. The roadmap toward higher port counts suggests growing adoption of optical switching architectures. This represents a significant shift in how data center networks may be designed in the future.

iPronics scales optical switching manufacturing

iPronics expanded its manufacturing capabilities for silicon photonics-based optical switches through a partnership with Fabrinet. This move enables the transition from prototype systems to volume production. The expansion supports growing demand for optical switching in AI infrastructure. It marks an important step toward commercialization.

HyperLight introduces 400G-per-lane TFLN PICs

HyperLight unveiled photonic integrated circuits based on thin-film lithium niobate capable of 400G per lane operation. These devices offer high bandwidth with low power consumption, making them suitable for AI workloads. The technology represents an alternative to traditional silicon photonics. It highlights ongoing diversification in optical materials and approaches.

Tower Semiconductor and Coherent demonstrate 400G/lane silicon modulator

The companies demonstrated a silicon photonics modulator achieving 400G per lane in a production-ready process. This validates the scalability of CMOS-compatible photonics for next-generation optical systems. The result supports the transition toward higher-capacity modules. It also reinforces silicon photonics as a key technology for AI infrastructure.

Coherent advances 224G ecosystem and InP platform

Coherent highlighted its roadmap toward 224G signaling and 1.6T optical modules, supported by indium phosphide technology. The company emphasized vertical integration across components. This approach enables improved performance and reliability. It positions InP as a critical platform for high-speed optical systems.

Lumentum showcases 1060nm VCSEL optical interconnect

Lumentum demonstrated a VCSEL-based optical interconnect optimized for short-reach AI networking. The design reduces power consumption and improves efficiency. It also enables higher density connectivity within racks. VCSEL technology continues to evolve for next-generation applications.

Avicena unveils microLED optical I/O platform

Avicena introduced a microLED-based optical interconnect that eliminates traditional laser sources. This approach reduces complexity and improves efficiency. It represents a novel direction for short-reach connectivity. The platform could enable new system architectures.

AOI highlights high-power external laser for CPO

Applied Optoelectronics showcased an external laser source designed for co-packaged optics systems. The architecture improves thermal management and serviceability. It supports scalable optical deployments. External laser designs are gaining traction. The company also indicated it has secured a major hyperscaler design win, signaling strong commercial traction for its external laser strategy as AI infrastructure deployments accelerate.

Corning introduces PRIZM TMT fiber platform

Corning addressed fiber density challenges with a new platform for AI data centers. The system enables higher fiber counts within limited space. This supports scaling of GPU clusters. It reflects growing infrastructure demands.

Nokia explores PON for data center operations

Nokia proposed using passive optical networking for data center management. The approach reduces complexity and power consumption. It applies telecom principles to new environments. This reflects broader convergence trends.

VIAVI introduces 224G validation platform

VIAVI launched a platform for testing next-generation optical systems. It addresses signal integrity challenges at higher speeds. Validation is critical for deployment. The tool supports ecosystem readiness.

Keysight launches 224G test solutions

Keysight introduced tools for validating high-speed interfaces. These solutions support both R&D and manufacturing. They ensure interoperability. Testing remains a key enabler.

PhotonIC launches ROCS qualification platform

PhotonIC introduced a system for qualifying optoelectronic components. It improves supply chain reliability. The platform addresses growing complexity. It supports scalable manufacturing.

NewPhotonics heaterless micro-ring modulator

NewPhotonics introduced a design that eliminates thermal tuning. This reduces power consumption. It improves efficiency. The approach advances silicon photonics.

NewPhotonics 3.2T DR8 transmitter-on-chip PIC

The company unveiled a highly integrated photonic chip targeting 3.2T systems. It combines lasers and modulators. This improves density and efficiency. It supports hyperscale deployments.

Adtran LiteWave800 low-power 800G LPO module

Adtran introduced a low-power 800G optics module for AI data centers. It reduces energy consumption. It improves thermal performance. LPO continues to evolve.

Credo targets AI scale-out with high-efficiency optical DSP and SerDes solutions

Credo highlighted its portfolio of low-power, high-performance connectivity silicon designed for AI scale-out networks, including optical DSPs and SerDes technologies optimized for 800G and 1.6T interconnects. The company emphasized energy efficiency and signal integrity as critical differentiators, particularly as hyperscalers push toward higher port densities and tighter power envelopes. Credo’s approach centers on minimizing power per bit while maintaining robust performance across longer reach links, positioning its solutions for both pluggable optics and emerging optical architectures. The announcement underscores the increasing importance of connectivity silicon vendors in enabling scalable AI infrastructure.

Ayar Labs advances optical I/O for chip-to-chip AI interconnects

Ayar Labs showcased its optical I/O technology designed to replace traditional electrical interconnects with high-bandwidth, low-latency optical links directly at the chip level. The company’s solution integrates optical engines with silicon to enable efficient data movement between processors, addressing one of the key bottlenecks in AI system design. By reducing power consumption and increasing bandwidth density, Ayar Labs’ approach supports next-generation scale-up architectures where chip-to-chip communication becomes critical. The technology reflects a broader industry shift toward optical interconnects not just between systems, but within compute packages themselves.

Looking Ahead

OFC 2026 underscored the rapid alignment of the optical networking industry around AI infrastructure requirements, with innovation spanning architectures, components, and system integration.

OFC 2027 will take place March 7–11, 2027, at the Los Angeles Convention Center.

Did I miss something?

Reach out to [email protected] to let me know. Up next: 30 short videos filmed at OFC.

Tags: OFC26

🎡 OFC 2026 News Rodeo: AI Infrastructure Drives Record Week of Optical Innovation

Optica Executive Forum: Microsoft’s Yawei Yin Explores “Scale-Across”

Digital Realty Expands into Milan with 84 MW Data Center Campus

Jim Carroll

Related Posts

TeraSignal Introduces Intelligent Redriver for 200G+ Copper Interconnects

NLM Photonics Advances Silicon-Organic Hybrid PICs

Cisco’s Bill Gartner on Optics Qualification, Testing, and Supply Chain Strategy

1Finity Ultra Optical System OLS Built for the AI Era

Optica Executive Forum: Microsoft’s Yawei Yin Explores “Scale-Across”

Optica Executive Forum: Marvell’s Radha Nagarajan on Optical Interconnects for AI

Digital Realty Expands into Milan with 84 MW Data Center Campus

Categories

Archives