Ayar Labs, which develops optical interconnect chips for large-scale AI workloads, has secured $155 million in financing, including from competing processor companies Nvidia, Intel and AMD. Founded in 2017, the Silicon Valley-based company is pursuing a different processing path — combining photonic elements with electronic circuits on each chip for what it says provides faster, more efficient processing for artificial intelligence and high-performance computing. “This brings the company’s total funding to $370 million and raises the company’s valuation to above $1 billion,” Ayar notes, adding that the new funding allows the company to scale its optical I/O tech.

This latest round was led by Advent Global Opportunities and Light Street Capital, but it was other participants — notably “leading GPU providers AMD and Nvidia” along with “semiconductor foundries GlobalFoundries, Intel Foundry, and TSMC” — whose support “underscores the potential of our optical I/O technology to redefine the future of AI infrastructure,” Ayar Labs CEO and co-founder Mark Wade said in an announcement.

Ayar has developed what it says is the industry’s first standards-based, in-package optical I/O solution that is “commercial-ready and optimized for AI training and inference,” teeing it up to replace short-distance electrical bridges with optical interconnects that can communicate faster, reducing costs and energy consumption while increasing productivity, SiliconANGLE explains.

It will also fix “one of the main bottlenecks in AI — namely, the movement of data among the powerful accelerator chips that are used to process AI training and inference.”

“The artificial intelligence boom has been a power-hungry and expensive venture to fund,” writes Bloomberg, adding that AI and HPC functions make huge transmission demands on servers that are enabled mainly by Nvidia GPUs.

“The AI workload is really breaking the back of the existing hardware infrastructure, especially in interconnects. We’ve come up with a way to replace those,” Wade tells Bloomberg. (The Next Platform deep dives the AI data processing limitations of traditional copper.)

Ayar Labs’ two main products are the TeraPHY optical chiplet and a multi-wavelength light source called SuperNova. Additionally, UCIe 2.0, released in August, is “critical for the interconnect chiplet” and sets the stage for next-gen AI architectures, Ayar details in a blog post.

“The value of our optical solutions plays out when you’re looking at massive amounts of bandwidth escaping the package and running workloads that need hundreds of [system on a chip] SOC packages working together,” Wade told HPC Wire in an interview.

“As the industry heads toward putting one trillion transistors on a chip by 2030, breakthroughs in transistor and interconnect scaling are critical” Intel Foundry wrote in a news release from this month’s IEEE conference.