By Jane Lanhee Lee
OAKLAND, California (Reuters) – Lightmatter, a Boston-based startup using light for AI computing, said on Wednesday it raised $154 million and tripled its valuation as customers in need of faster, more energy efficient computing for AI work are buying its systems.
“We use light to link computer chips together and we also use light to do calculations for deep learning,” Lightmatter co-founder and CEO Nick Harris said. “The reason that we’re getting these customers and data center scale deployments with our interconnect is that the generative AI boom is driving high-end chips like crazy.”
Light has been used for decades in the telecommunications industry to move data across the world with fiber optic cables. Now, large tech companies are looking to use light-based technology in their data centers to reduce power costs. The challenge has been to shrink the size of the devices used to create or control light.
As chips become more advanced, they consume more power. “It can be up to half of the entire operating cost of a data center just getting the power to the chips,” Harris said, adding that light transmits information far more energy efficiently than electrical signals traveling over wires.
Harris said its chip called Passage that connects other chips by translating electrons to photons and back will be mass produced and used in data centers next year. A large part of the funding will help roll out this product.
Its more futuristic product Envise is a system that uses light, instead of electrons, to do computing. Harris said Envise will be able to run large AI models but will not train them. Customers will test the system this year, he said.
GV (Google Ventures), the venture capital investment arm of Alphabet Inc, and Fidelity Management & Research Company were among the investors taking part in this latest round, the company said.
(Reporting by Jane Lanhee Lee; Editing by Lisa Shumaker)
Disclaimer: This report is auto generated from the Reuters news service. ThePrint holds no responsibilty for its content.