Intel and Ayar Labs Propel 224Gbps+ Scale-Up with Co-Packaged Optics, Extending NVIDIA NVLink

September 20, 2026

Silicon Co-Design USA

Overview

Co-Packaged Optics (CPO) technology is enabling 224Gbps+ scale-up interconnects, with Intel pursuing a truly integrated approach via its Optical Compute Interconnect (OCI) that heterogeneously integrates InP lasers on a Si wafer. Ayar Labs positions itself with an open, protocol-agnostic optical I/O chiplet platform, utilizing TeraPHY chiplets and SuperNova multi-wavelength light sources to extend GPU-to-GPU fabrics such as NVIDIA NVLink. Key CPO switch solutions from NVIDIA, Marvell/Celestial, and Broadcom aim to overcome the bandwidth, latency, and power efficiency limitations of traditional copper interconnects.

In Depth

Key Findings

The landscape of high-performance computing is being reshaped by Co-Packaged Optics (CPO), enabling data transfer speeds exceeding 224Gbps for scale-up architectures. Intel is pioneering a truly integrated CPO solution with its Optical Compute Interconnect (OCI), featuring heterogeneously integrated Indium Phosphide (InP) lasers directly on silicon wafers. Concurrently, Ayar Labs is advancing an open, protocol-agnostic optical I/O chiplet platform, designed to extend GPU-to-GPU fabrics like NVIDIA NVLink through its TeraPHY chiplets and SuperNova multi-wavelength light sources. These advancements collectively address the critical bandwidth, latency, and power efficiency challenges inherent in conventional copper interconnects for AI infrastructure.

Technical / Clinical Details

CPO integrates optical transceivers directly into the same package as electrical switching chips, dramatically shortening electrical trace lengths and thereby reducing signal loss, latency, and power consumption. Intel’s Optical Compute Interconnect (OCI) represents a sophisticated approach, achieving deep integration by heterogeneously combining InP lasers, a III-V compound semiconductor, onto a silicon photonics wafer manufactured through standard CMOS processes. This enables chip-level optical interconnects, resolving many issues associated with traditional pluggable optical modules. Ayar Labs’ platform offers a different but complementary strategy: it leverages TeraPHY optical I/O chiplets and SuperNova multi-wavelength light sources to enhance data movement between computational elements like GPUs, CPUs, and memory. Being open and protocol-agnostic, their solution is designed to opticalize existing electrical fabrics, such as NVIDIA’s NVLink and PCIe, extending their reach and performance capabilities significantly. Prominent CPO switch solutions, including NVIDIA Quantum-X Photonics, Marvell/Celestial Photonic Fabric, and Broadcom TH5/TH6, are also playing crucial roles, each contributing unique innovations to bolster data center interconnect performance and form the backbone of next-generation AI infrastructure.

Background & Context

The escalating scale and complexity of AI models have led to an exponential increase in data transfer requirements within data centers. Traditional copper-based interconnects are encountering severe limitations at data rates exceeding 224Gbps, struggling with signal integrity, power consumption, electromagnetic interference, and physical manageability. These issues create bottlenecks in bandwidth, latency, and power efficiency, particularly for GPU-to-GPU connections where extremely high bandwidth, as demanded by technologies like NVLink, is critical. CPO has emerged as one of the most promising technologies to overcome these limitations by placing optical engines much closer to the compute chips, thereby mitigating electrical constraints. This allows for higher density and more power-efficient designs for AI supercomputers and hyperscale data centers, paving the way for further advancements in AI computational capabilities.

Strategic Significance & Outlook

The ongoing advancements in CPO technology by Intel, Ayar Labs, and other leading vendors are poised to profoundly impact the future of AI infrastructure. The capability for 224Gbps+ scale-up is indispensable for meeting the demands of upcoming AI workloads. The broader adoption of CPO will significantly influence the optical module market, semiconductor manufacturing, and advanced packaging technologies across the entire supply chain, fostering new innovations and business models. An open platform strategy is key to enhancing vendor interoperability and accelerating the growth of the CPO ecosystem. Ultimately, CPO is expected to revolutionize data center and AI compute architectures, leading to substantial improvements in power efficiency and unlocking virtually unlimited bandwidth, thereby strongly propelling the further evolution and societal implementation of AI globally.

Source: https://www.siliconcodesign.com/p/co-packaged-optics-for-224gbps-scale

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.