Optical interconnect startup Ayar Labs is the latest name to join NVLink Fusion, Nvidia’s scale-up centric backplane.

NVLink Fusion allows partners to build semi-custom AI infrastructure linking non-Nvidia accelerators with Nvidia hardware via chiplet-based interconnect (UCIe) providing the bridge. Ayar Labs joining the cohort will allow its co-packaged optics (CPO) products to pair up with Nvidia’s own tech and bring it to rack-scale infrastructure.

“AI infrastructure is being co-designed from the ground up, and customers need more options to scale performance efficiently as bandwidth continues to rise,” said Mark Wade, CEO at Ayar Labs. “By joining the NVLink Fusion ecosystem, we’re introducing co-packaged optics as a foundational building block for customers deploying heterogeneous compute in Nvidia AI factories.”

NVLink Fusion was unveiled last May, already counting partners, including Fujitsu, Synopsys, and Qualcomm. Prior to Ayar Labs, arguably the biggest get was Arm, with the processor pioneers able to make its Neoverse designs interoperable with Nvidia hardware.

Despite its startup status, the addition of Ayar is another coup in itself, given co-packaged optics are set to become a huge part of Nvidia’s AI infrastructure stack. The chip giant is betting on photonics to slash power consumption while boosting the bandwidth needed to power ever-increasing AI workloads.

Ayar claimed to have unveiled the world’s first UCIe optical interconnect chiplet – with ASIC company Alchip working to develop scalable AI infrastructure based on the technology.

As part of the NVLink Fusion ecosystem, Ayar’s CPO technology looks to provide high-bandwidth, low-latency, power-efficient optical connectivity for hyperscalers and cluster builders to create optically-connected AI infrastructure.

“NVLink Fusion, combined with Ayar Labs’ CPO technology, gives customers more options to build heterogeneous AI factories,” said Ashish Karandikar, VP of engineering at Nvidia. “By expanding the NVLink Fusion ecosystem with Ayar Labs’ optical connectivity, innovators can scale bandwidth and bring heterogeneous AI infrastructure to market faster.”

The NVLink Fusion closely follows another Ayar milestone, which saw it raise $500 million back in March. Alchip participated in the Series E funding round, along with AMD and fellow NVLink Fusion player MediaTek, with the cash going towards scaling high-volume production of its CPO offering.

Nvidia also took part, with its backing central to its growing CPO bet, having also invested in Lumentum, Corning, and Coherent to strengthen its supply chain amid existing component issues.

Another of those firms – Marvell – was anointed “the next trillion-dollar company” by Nvidia's Jensen Huang during a keynote at Computex this week.

Also this week, we got our first look at one of its upcoming CPO switches. Lambda, the Microsoft-backed neocloud, was given early access to a Quantum-X Q3450-LD switch, which, due to its optical engines being placed next to its ASICs, eliminates the need for traditional pluggable transceivers.

CPO versions of Quantum-X and Spectrum-X are expected to launch later this year, with Huang stating that the hardware was in “full production” around the time of GTC earlier this year.