Inference chip startup Etched had launched from stealth with $800 million in funding.
The company also announced it had developed a working chip and already signed more than $1 billion in customer contracts.
The funding was raised over multiple rounds, the company said, the latest being a $500 million round that provided Etched with a $5 billion post-money valuation in December. Investors in the startup include VentureTech Alliance, Jane Street, Hudson River Trading, Stripes, Radical Ventures, Primary VC, alongside Peter Thiel. Andrej Karpathy and Geoffrey Hinton.
Etched’s unnamed chip was produced on TSMC’s N4P process technology, and the company is now validating its rack-scale product with customers. The chip uses an architecture that creates a shared low-latency memory pool across the entire scale-up domain, the company said, adding that its proprietary ultra-low-latency, high-bandwidth interconnect enables “dramatically faster” memory access across chips.
Headquartered in San Jose, the company said that in addition to developing the chip, it has also stood up a factory in Taiwan and has built a 2MW data center, test house, and NPI prototyping lab at its offices in California. No further details about the facilities have been provided.
“Our approach from the beginning has been to build for gigawatt-scale,” said Rob Wachen, co-founder of Etched. “Production is the product. We are living through one of the largest infrastructure buildouts in history, and the companies that matter will be the ones that can translate technology into systems that can be manufactured, deployed, and operated at massive scale. We’re as focused on operations as we are on frontier research.”
Gavin Uberti, co-founder and CEO of Etched, added: “We recognized early on that frontier AI would become one of the most economically significant technologies ever created, but that the infrastructure needed to serve those models in a sustainable and economically viable way simply did not exist. As AI rapidly embeds in every industry and application, the need for accelerated inference infrastructure has never been greater.”