Nvidia Is Losing Top Talent to $21B AI Chip Startup: Why Investors Are Betting on the 'Kids in Chips'

Etched, a San Jose-based AI chip startup founded by three 24-year-old Harvard dropouts, has raised its valuation to $21 billion after a $700 million funding round led by Jane Street, and has attracted nearly 400 former Nvidia employees, including a dozen engineers recruited by veteran systems engineer Brian Loiler. The startup's chips, designed for fast AI inference, complete certain communication operations in 700 nanoseconds versus 4,000 nanoseconds on Nvidia's Blackwell architecture, and it has secured Jane Street as its first client with orders worth $1 billion. Investor Michael Burry highlighted the competition in a Tuesday X post, saying 'This is serious competition for Nvidia.'

Nvidia Is Losing Top Talent to $21B AI Chip Startup: Why Investors Are Betting on the 'Kids in Chips' Nearly 400 Etched employees previously worked at Nvidia, with the company recruiting a dozen Nvidia engineers to improve its server design Nvidia's AI chip dominance could come under pressure from a new semiconductor startup founded recently by three 24-year-old Harvard dropouts. The Big Short 's Michael Burry highlighted in a Tuesday X post that the startup had claimed that it took 44 days after getting its test chips back from Taiwan Semiconductor Manufacturing to set them up and running inference workloads, which usually takes at least six months. 'This is serious competition for Nvidia,' he wrote. Etched, a San Jose-based AI chip company, has seen its valuation surge to $21 billion after raising $700 million in a funding round led by Jane Street. The latest deal more than doubled its valuation within a month, according to Reuters. The startup built a 2-megawatt in-office data centre and has some serious control over its supply chains. Generally, chip startups outsource testing of their products, but Etched has a team of 20 people in Taiwan, where it owns a factory designed to build server components. Sequoia Capital also simultaneously led Etched's $300 million series C round. Overall, the company has raised close to $2 billion on a newly conceptualised fast inference computing product, and has already secured Jane Street as its first client with orders worth $1 billion. The startup appears to be targeting the quantitative trading industry, and in early 2025, it hired former renowned quant trader Tim Perevozchikov. Quant trading firms generally seek cutting-edge processors to quickly run their investing models for trading edge, Perevozchikov had said. Among tech investors, the prevailing wisdom was 'don't back the kids in chips,' said Sequoia Capital's general partner Sonya Huang. 'There was a sense that these kids think they can do anything, and maybe on the internet, they can, but that in chips, startups are guilty until proven innocent.' An Army of Nvidia Veterans Join Etched A striking pattern at the company is that nearly 400 employees had previously worked at Nvidia, according to The Wall Street Journal . Elsewhere, Burry mentioned in his X post that Etched hired Nvidia's veteran systems engineer Brian Loiler in 2024 to advance server design. 'Loiler has since recruited to Etched about a dozen engineers from Nvidia who, in some cases, turned down attractive counteroffers,' Burry had stated. Burry's comments coincide with his ongoing warnings https://www.ibtimes.co.uk/michael-burry-bet-against-oracle-nebius-1813014 about how AI profits do not add up, and tech giants are artificially inflating earnings to make balance sheets appear better than they actually are. Furthermore, systems engineer Anthony Gamerman joined Etched from Nvidia in April, as employees at the multi-trillion-dollar company continued to leave rapidly. 'One leaves, and then another one, you hear a couple weeks later, and you go, "Whoa, I can't believe so-and-so is heading over to Etched,"' he had reportedly said. Back in 2023, the startup had also onboarded Saptadeep Pal as its chief silicon architect. The exec researched multi-GPU parallelization at Nvidia, and later co-founded bitcoin miner Auradine, while claiming to have ended up designing the lowest-voltage chip ever shipped. Etched's Cluster-Scale Memory Takes on Nvidia's Blackwell Etched's chips are designed around a 'cluster-scale memory' architecture that enables processors within a system to communicate more efficiently. While certain communication operations take 4,000 nanoseconds on Nvidia's Blackwell architecture, they can be completed in 700 nanoseconds on Etched's systems, according to the company. In case AI inference becomes a major bottleneck, customised chips could gain share in a market currently dominated by Nvidia GPUs. Nvidia itself has likely acknowledged the importance of specialised inference technology. In December 2025, it entered a non-exclusive licensing agreement with Groq https://www.ibtimes.co.uk/nvidia-ai-chip-demand-2027-new-platforms-1786114 and hired founder Jonathan Ross, president Sunny Madra, and other employees. Nvidia's annual report says it paid $13 billion at closing, with another $4 billion due within a year, for the Groq technology and workforce-related deal. Groq remained an independent company. This development makes Etched's valuation even more revealing as investors appear to be increasingly assigning value to companies developing alternatives to Nvidia's dominant architecture. 'We may be the only AI chip that was built by a startup that was successful on the first try,' Etched president and co-founder Robert Wachen had told a media outlet. Two other founders, Gavin Uberti and Chris Zhu, met at an undergraduate course in theoretical mathematics. Meanwhile, Wachen, who reportedly lived on the same floor as Uberti, was running an edutech firm. Currently, the company has solid backing from popular names, including Blackstone, Andreessen Horowitz, Kleiner Perkins, and Tiger Global. © Copyright IBTimes 2025. All rights reserved.