OFC 2026: Optical Scale-up consortium sets path for an open AI infrastructure specification
As hyperscalers continue to scale capacity to meet AI and cloud service demands, the need for greater capacity within data centers has never been more pressing.
One answer to guide this process came about this week with the launch of the Optical Compute Interconnect (OCI) Multi-Source Agreement (MSA) Consortium.
Led by Founding Members AMD, Broadcom, Meta, Microsoft, NVIDIA, and OpenAI, this industry consortium marks a pivotal shift toward a hyperscaler-driven open ecosystem to enable the development of a multi-vendor supply chain for optical-scale interconnects.
According to the founders, the goal is to enable tighter integration of optics with compute and networking silicon.
Further, OCI can enable gains in bandwidth density and system scalability while meeting the aggressive power targets of legacy copper-based connectivity.
Founding member Broadcom, which debuted its 3nm 400G/lane optical PAM-4 DSP, the Taurus™ BCM83640, said that it is an effort to drive 400G interoperability.
Khushrow Machhi, senior director, product marketing at Broadcom, said its effort will complement the 400G work being done by the IEEE and OIF. “We’re active in IEEE and OIF, which are in the early phases of defining the 400 gig per lane standards,” he said in a Broadband Pulse podcast interview. “So, in the meantime, Broadcom, along with NVIDIA, Cisco, MACOM, and Semtech, are founding members of a 400-gig optical MSA that’s intended to create an interoperable ecosystem around the 400-gig per lane optics.”
He added that “the eventual goal is to move some of this early work into IEEE and OIF so that it becomes standardized there also.”
Plug and play focus
By establishing an interoperable optical interface protocol, the founders said the OCI MSA enables a "plug-and-play" ecosystem.
As an open, interoperable specification, the OCI MSA aims to enable hyperscalers to disaggregate top-tier processor unit (XPU) engines and top-tier scale-up switches onto a common optical physical layer (PHY), ensuring that compute meets state-of-the-art optics.
By establishing a standardized roadmap, participants in the MSA can help reduce integration risk, shorten development cycles, and provide the full AI rack supply chain with a clear, de-risked path for multi-generational, multi-vendor optical interconnect deployments.
Copper to optical scale-up
While traditional copper has been sufficient for data movement across short distances within data centers, its limits will drive a migration to all-fiber solutions.
Traditional copper-based connectivity is reaching physical limits, which are impacting AI cluster-scale-up domain architectures. OCI will enable migration from copper-based to optical-based scale-up architectures, alleviating copper interconnect bottlenecks.
The OCI specification, available at www.oci-msa.org, combines non-return-to-zero (NRZ) modulation and WDM optical technology, shifting the connectivity paradigm from a module-centric to a silicon-centric model.
By enabling tighter integration of optics with compute and networking silicon, OCI aims to offer meaningful gains in bandwidth density and system scalability while meeting the aggressive power targets of legacy copper-based connectivity.
Offering a unified roadmap for the full AI rack supply chain, the MSA enables multi-vendor optical PHY and interconnect deployments across multiple hardware generations.
Standardization and scale are key focuses. The MSA highlights the OCI GEN1 4-wavelength x 50Gbps NRZ (200Gbps/direction) and the OCI GEN2 400Gbps/direction bidirectional (BiDi) technologies, achieving up to 800 Gbps per fiber.
Looking beyond 800G, the MSA creates a roadmap to scale wavelength counts and data rates to reach 3.2 Tbps per fiber and higher, and to enable scale-up domains with both higher GPU counts and higher bandwidth per GPU. Finally, the MSA will have support for pluggable, on-board, and co-packaged optics (CPO).
Meta, which has set an aggressive pace to expand its data center footprint, with 26 of its data centers under construction or operational in the US, sees the potential of the MSA to address the power and cost constraints impacting AI cluster design as real and imminent.
“We encourage adoption of this OCI protocol to decouple the need for larger scale-up domains from the limitations of electrical backplanes in high-performance AI clusters,” said Dan Rabinovitsj, Vice President of Hardware Systems at Meta.
For related articles, visit the Broadband Topic Center.
For more information on high-speed transmission systems and suppliers, visit the Lightwave Buyer’s Guide.
To stay abreast of fiber network deployments, subscribe to Lightwave’s Service Providers and Datacom/Data Center newsletters.
About the Author
Sean Buckley
Sean is responsible for establishing and executing the editorial strategy of Lightwave across its website, email newsletters, events, and other information products.