Strategic Alliance to Bring Co-Packaged Optics to Scale
Global Unichip Corp. (GUC), a leading Advanced ASIC company, and Ayar Labs, a pioneer in co-packaged optics (CPO) for large-scale AI workloads, have announced a strategic partnership aimed at integrating CPO technology into hyperscale data centers. This collaboration targets the rapid deployment of high-bandwidth, low-latency interconnects that are essential for training and inference in modern AI systems. By combining GUC’s custom silicon capabilities with Ayar Labs’ photonics-based data paths, the duo seeks to reduce power consumption, cabling complexity, and latency across massive AI clusters.
What Co-Packaged Optics Brings to Hyperscalers
Co-packaged optics place optical communications components in close proximity to the silicon die, creating shorter signal paths and higher bandwidth per watt. For hyperscalers deploying AI models with trillions of operations, CPO can dramatically improve data throughput while lowering the energy footprint. The partnership envisions leveraging Ayar Labs’ integrated photonic modules alongside GUC’s ASIC prowess to create scalable, production-ready solutions for data centers that demand consistent performance and operate within tight power envelopes.
Key Benefits for AI Workloads
- Higher bandwidth per rack with reduced electrical interconnects
- Lower latency between accelerators, memory, and storage
- Improved energy efficiency, aligning with hyperscale sustainability goals
- Greater scalability to support evolving AI model sizes and training regimes
Collaborative Roadmap and Target Markets
The partners plan to co-develop reference designs and production-ready reference platforms that enable hyperscalers to evaluate CPO in real-world workloads. The initial focus will be on large-scale AI training clusters and inference fabrics used in data centers supporting enterprises, cloud providers, and AI research facilities. By aligning Ayar Labs’ photonics integration with GUC’s ASIC design services, the alliance aims to shorten time-to-market for CPO-enabled systems and provide a path for customers to adopt high-bandwidth interconnects without overhauling existing server architectures.
Deployment Scenarios and Technical Considerations
In practical deployments, CPO can consolidate optical links directly with compute tiles, reducing the density of wiring and the associated thermal and mechanical challenges. The collaboration will address system-level considerations such as thermal management, signal integrity across optical links, and firmware/software integration to support existing AI frameworks. With data center operators under pressure to improve performance-per-watt, the GUC-Ayar Labs alliance positions itself to offer a compelling option for next-generation AI accelerators, memory hierarchies, and accelerator interconnects.
Industry Context and Outlook
Co-packaged optics is gaining momentum as hyperscalers seek to sustain exponential AI throughput while managing total cost of ownership. The joint effort between GUC and Ayar Labs aligns with broader industry trends toward photonics-enabled scalability in data centers. As AI models grow more complex, the demand for high-bandwidth, energy-efficient interconnects will only increase, potentially reshaping how hyperscalers architect their AI compute fabrics.
About the Partners
Global Unichip Corp. (GUC) provides advanced ASIC development and manufacturing services, helping customers translate complex silicon ideas into production-grade chips. Ayar Labs specializes in co-packaged optics that integrate photonics with compute, aiming to deliver scalable optical interconnects tailored for modern AI workloads. Together, they seek to accelerate the adoption of CPO in hyperscale environments and enable higher performance AI systems.
