[
The aspiring competitor to Nvidia plans to start shipping chips to some customers this summer.
It was earlier reported that Etched raised a $500 million round valuing the company at $5 billion. Led by investment firm Stripes with participation from billionaire Peter Thiel, Positive Sum and Ribbit Capital, that round closed in December and also included an investment from Jane Street, Hudson River Trading, Two Sigma and VentureTech Alliance, which has a strategic partnership with TSMC, Etched said.
Jane Street led a previously unannounced funding round, according to Etched co-founder and president Robert Wachen, and has invested additional funds since then. The trading firm has invested a total of more than $100 million in Etched, according to people familiar with the matter.
Founded in 2022, Etched designs chips to run AI models. The company is currently testing its products and has signed sales contracts worth $1 billion, said Wachen, who declined to name any customers. This is the first time Etched has publicly discussed its funding and chip plans in about two years. “We’ve been very quiet until we had stuff to show off,” said Gavin Uberti, co-founder and CEO.
Other Etched investors include Geoffrey Hinton, who won a Nobel Prize for developing some of the key concepts in modern AI, computer vision pioneer Fei-Fei Li and hedge fund manager Stanley Druckenmiller.
Etched is among a group of startups vying to compete with Nvidia as the industry shifts from training AI models to running them, a process known as inference. Groq, another inference chip company, in December agreed to license its technology to Nvidia for a reported $20 billion. Under the deal, Nvidia will hire most of Groq’s team. In April, Alphabet Inc.’s Google announced that a version of its AI chips will focus on inference. Working with TSMC, Etched has developed a technology it calls low-voltage inference, essentially running the chips at lower voltage to prevent overheating. That lets Etched squeeze more performance out of its silicon. The company also developed a memory system that combines two different kinds — the High-Bandwidth Memory that’s critical to Nvidia’s chips and the Static Random-Access Memory that’s increasingly being used in chips meant for fast responses to AI queries.
Besides creating custom chips, Etched designs its entire server rack — the circuit boards, plates for cooling the chips, networking connections and more. Etched is the only chip startup to have done this, Wachen said.
“If you’re gonna do a hard thing, do the whole hard thing,” Uberti said. “You have to go build a lot of products if you want to go compete at scale.”
It hasn’t been easy. After a supplier in Bangalore began having issues that would have caused a one- to two-year delay, Etched sent about a dozen of its best engineers to India for six months and fixed the issue.
More than half of the company’s 400 employees live close to the firm’s San Jose, California, headquarters — making it possible for them to work quickly and solve knotty challenges, Wachen said.
Having products available soon and focusing on the whole rack is a big advantage in gaining customers, said Patrick O’Shaughnessy, CEO of Positive Sum, a venture capital firm that led an earlier investment round in Etched.
“Timing matters,” he said. “If you have compute now, people will buy it.”
https://img.etimg.com/thumb/msid-132103921,width-1200,height-630,imgsize-18803,overlay-etmarkets/articleshow.jpg
https://economictimes.indiatimes.com/markets/us-stocks/news/jane-street-tsmc-linked-ai-startup-etched-raises-800-m/articleshow/132103929.cms




