AI boom triples valuation of US startup Lightmatter, which uses light for computing

OAKLAND, Calif., May 31 (Reuters) – Boston-based startup Lightmatter, which uses light for AI computing, said Wednesday that customers who need faster, more energy-efficient computing for AI work It announced that it had raised $154 million and tripled its valuation because of the purchase. that system.

“We’re using light to link computer chips, and we’re using light for deep learning calculations,” said Nick Harris, co-founder and CEO of Lightmatter. rice field. “The reason we are seeing these customer and datacenter-scale deployments in our interconnect is because the generative AI boom is driving high-end his chips at breakneck speed.”

For decades, the telecommunications industry has used light to move data around the world using fiber optic cables. Major technology companies are now looking to use optical-based technology in their data centers to reduce power costs. The challenge has been to shrink the size of the devices used to generate or control light.

As chips get more powerful, they consume more power. “Just powering a chip can be up to half the cost of running an entire data center,” Harris said, noting that light can transmit information much more energy efficiently than electrical signals traveling on wires. added.

Harris said a chip called Passage, which connects other chips by converting electrons to photons and vice versa, will be in mass production next year and used in data centers. Most of the funds will go towards the deployment of this product.

A more futuristic product, Envise, is a system that uses light instead of electrons for computing. Harris said Envise can run AI models at scale, but won’t train them. He said customers are planning to test the system this year.

GV (Google Ventures), the venture capital investment arm of Alphabet (GOOGL.O), and Fidelity Management & Research Company were among the investors in the latest round, the company said.

Reported by Jane Lanhee Lee.Editing: Lisa Shoemaker

Our standards: Thomson Reuters Trust Principles.