After obtaining $58 million in backing from many major tech companies, the startup for optical networking chips, nEye Systems Inc., hopes to become the hub for artificial intelligence data centers.

Leading today’s Series B round was CapitalG, a growth-stage fund supported by Alphabet Inc., the parent company of Google LLC. Microsoft Corp.’s M12, as well as Micron Ventures, Nvidia Corp., and Socratic Partners, participated. After a prior, secret initial fundraising round, it gives the company a total of over $72 million raised.

The business, situated in Emeryville, California, is creating a networking device that uses optical technology instead of electrical signals to transfer data between AI chips. Because this technique can enable quicker chip-to-chip connections at a significantly lower cost than current network interconnects, companies like Google and Nvidia are very interested in it.

The majority of contemporary data centers still use outdated electrical switch-based interconnects, which have high energy consumption and severe bandwidth restrictions. Because they can’t communicate with one another fast enough, this creates a bottleneck for AI workloads that depend on massive clusters of interconnected graphics processing units. Because of the significant power use, it also means higher prices.

Since its direct optical connections allow data transfers with nearly infinite bandwidth, the company’s wafer-scale optical circuit switch is said to be a more effective and economical option. It claims that compared to current data center interconnects, its chips are roughly 100 times smaller, 1,000 times more energy efficient, 10,000 times faster, and 10 times less expensive.

There are more startups constructing optical interconnects. Other well-funded businesses like Lightmatter Inc., Celestial AI Inc., Ayar Labs Inc., Xscape Photonics, and Lightium AG are developing comparable technology. Another term for the same idea is silicon photonics, and traditional chip titans like Intel Corp. and IBM Corp. are also working on it.

Nevertheless, nEye seems to have a secret weapon in the form of an intelligent optical circuit switch that can instantly modify the network of chip-to-chip connections in data centers. By choosing the best connection between several chips depending on the program being used, this technique can improve data center performance.

Alphabet has already conducted internal experiments with the concept. Google said to have constructed an AI supercomputer a number of years ago that was far more potent than Nvidia’s most cutting-edge products at the time. Google has never commercialized this technology, despite using it internally and promoting it as an energy saver today at its Cloud Next conference in Las Vegas.

Developing the idea to produce its own intelligent optical circuit switches is precisely what nEye intends to accomplish. It offers significant gains in data center efficiency for workloads including AI and other computations.

Ming Wu, a professor at the University of California Berkeley and one of the co-founders of Eye, told the report, “Google is a pioneer, it led the way.” “Instead of developing this technology themselves, other AI companies and hyperscaler AI data center operators will be looking to acquire it.”

According to the business, it has already constructed chip prototypes and hopes to be able to provide customers with manufacturing chip samples the following year. It hasn’t stated when it will be able to ship them in greater quantities, though.

Eye says that its chips can be equally useful for conventional data center workloads, which also have significant energy costs, even though its primary market is the AI sector.

According to James Luo, general partner at CapitalG, Eye is tackling important bottlenecks in both artificial intelligence and conventional high-performance computing. He emphasized that “the beauty of it is that it’s applicable to both models.”