Several weeks after Ayar Labs announced it was collaborating with Taiwanese ASIC company AIchip, the two have released further details about the terms of their partnership.

In early September, the partners said they were working together to develop scalable AI infrastructure based on co-packaged optics (CPO) technology, with the solution set to offer an ultra-high bandwidth, low latency, and energy-efficient solution for AI clusters.

The partners will also work with TSMC on the project, using the chipmaker’s advanced packaging and process technology, including COUPE, TSMC-SoIC, and advanced process nodes.

According to a presentation delivered by the two companies at TSMC’s North America OIP Ecosystem Forum last week, the new solution will feature Ayar Labs’ TeraPHY optical engines co-packaged with Alchip’s advanced solutions on a common substrate, bringing optical I/O directly to the AI accelerator interface.

This integration allows for more than 100Tbps of scale-up bandwidth per accelerator, the partners said, adding that it also supports more than 256 optical scale-up ports per device, in addition to enabling multi-rack network scale-up without the power and latency penalties of pluggable optics – achieved by minimizing electrical trace lengths and placing optical connections close to the compute core.

Ayar Labs’ TeraPHY is protocol agnostic and can be integrated with customer-designed chiplets and fabrics. The solution also supports the UCIe (Universal Chiplet Interconnect Express) standard for die-to-die interconnect, and has flexible protocol endpoints at the package boundary, meaning design teams can slot the joint solution alongside existing compute tiles, memory stacks, and accelerators.

The two companies are currently engaged with select customers to integrate CPO technology into next-generation AI accelerators and scale-up switches, and said they will share additional collateral, reference architectures, and build options with qualified design teams.

“AI cluster scale-up networks are bounded by copper connectivity distances. At the same time, network power efficiency is limited by power density and cooling,” explained Erez Shaizaf, Alchip’s CTO. “This means that we want to optimize space efficiency, but the challenge is that copper doesn’t have the I/O connectivity to put everything together. CPO will break these boundaries.”

Vladimir Stojanovic, CTO and co-founder of Ayar Labs, added: “AI has reached an inflection point where traditional interconnects are limiting performance, power, and scalability. Our collaboration with Alchip enables a CPO solution for AI scale-up, pioneering new, energy-efficient system architectures that will power innovations for hyperscalers and enterprise AI customers, and accelerate AI proliferation globally.”

Founded in 2015, California-based Ayar Labs develops optical interconnect solutions based on open standards, optimized for AI training and inference workloads.

In April 2025, the company unveiled the world’s first UCIe optical interconnect chiplet. Designed to support AI workloads, Ayar Labs said the offering will eliminate bottlenecks, maximizing artificial intelligence (AI) infrastructure performance and efficiency whilst reducing latency and power consumption.