Ayar Labs, a chip startup backed by NVIDIA and AMD, has raised $500 million in a Series E funding round, valuing the company at $3.8 billion.
The round includes investors such as Neuberger Berman, MediaTek, and the Qatar Investment Authority, as well as Alchip Technologies, ARK Invest, Insight Partners, Sequoia Capital, and 1789 Capital.
The company operates in a specialised area known as co-packaged optics, part of the broader field of silicon photonics. Its goal is to replace copper interconnects inside servers with optical, fibre-based links.
Instead of moving data as electrical signals through copper, Ayar’s chips send data as light. This approach is designed to reduce energy use and increase data transfer speed between processors in AI systems.
As reported by The Wall Street Journal, Ayar Labs is among a small group of companies working to change how semiconductors are connected inside server racks. The publication noted that the company has developed “some of the most advanced applications” of co-packaged optics, according to its investors and executives.
The shift comes as the artificial intelligence industry moves from training large language models to running them in real-world applications such as chatbots and AI agents. This stage, known as inference, requires chips to respond quickly and efficiently to user queries. As AI systems scale, moving data between chips consumes significant power, creating bottlenecks.
“We’re solving one of the biggest hardware issues that’s causing bottlenecks in AI,” said Mark Wade, Ayar Labs’ chief executive. He said the company has spent 15 years developing its core technologies and anticipated that copper connectivity would eventually limit computing performance in the 2020s or 2030s.
Specialised in co-packaged optics
Ayar Labs says its optical interconnect chips can deliver between four and twenty times more computing throughput per watt compared to traditional copper-based connections.
Copper links, while widely used, face physical limitations. As more electrical current flows through copper at higher speeds, signal quality degrades, and energy losses increase.
Gabe Cahill, managing director at Neuberger Berman, said, ” Companies tend to stay with copper ‘for as long as they can,’ but pushing more electrons through copper reduces signal fidelity. He described using light for data transmission as a significant step forward. Cahill also said the firm sees a strong opportunity for Ayar Labs to pursue an initial public offering in the near future.
NVIDIA has long relied on high-speed copper connections, such as NVLink, to connect chips within data centres. Over time, it has improved these systems to support more powerful processors. However, as AI clusters grow larger and inference demand increases, the limits of copper become more apparent.
Pat Gelsinger, former CEO of Intel and now a board member at Ayar Labs, said he began researching silicon photonics more than two decades ago. He acknowledged that early predictions about the rapid end of copper were premature. However, he now believes the scale of modern AI infrastructure has changed the equation.
As companies build larger AI clusters, the physical limits of copper wiring are constraining how large and efficient those systems can become. Developing co-packaged optics has been slow due to technical challenges. Integrating optical components alongside silicon chips requires precise manufacturing and careful heat management.
Packaging optical links without affecting performance or reliability has been a major hurdle for the industry. Gelsinger said Ayar Labs has addressed these issues and is preparing for high-volume production.
