Gavin Uberti was 21 when an interviewer introduced him as the Harvard dropout who meant to take on Nvidia. It is a strange sentence to hang on a person, and it sums up the problem of telling his story: the public record gives us the bet far more clearly than the boy.
What the record does say is short. He studied mathematics and computer science at Harvard and left without finishing. He was a 2024 Thiel Fellow, a programme that pays young people to skip college. His company's site lists him as a Math 55 alumnus, the notoriously hard Harvard course, and as an expert in AI compilers. Two others, Chris Zhu and Robert Wachen, are named alongside him as founders of Etched, and all three are described as Harvard dropouts. Where they grew up, what their parents said, what money they had: none of that has been published in anything we could read. We will not guess.
He studied mathematics and computer science at Harvard and left without finishing.
So start with the choice. Nvidia's GPUs can run almost anything. Etched decided to make a chip that runs one thing, the transformer models behind ChatGPT, Claude and Gemini. The chip is called Sohu, an ASIC made by TSMC on a 4 nm process. The company says one Sohu server can stand in for 160 Nvidia H100s when running trained models. Nobody outside the company had tested that claim as of August 2026.
What almost nobody outside the industry saw is how long it all sat in silence. Etched stayed in stealth for years. A $120 million Series A is reported, then a valuation near $5 billion in December 2025 and a $500 million round in January 2026 that the company only confirmed months later. On 30 June 2026 it came out with a working chip, about $800 million raised and more than $1 billion in signed customer contracts.
Then the money got louder. A $300 million Series C led by Sequoia in July put the price at $10.3 billion. In August, $700 million led by Jane Street lifted it to $21 billion.
That deal carries its own awkwardness. Jane Street is also Etched's first paying customer, so the loudest vote of confidence comes from a buyer. By then only a single rack had been delivered. No revenue has been published.
The cost of the bet is built into it. Sohu works only for transformers. If the field moves to something else, a chip that took years to make loses most of its use. A few founders who left college early have put their whole company behind one prediction about the future.
The lesson for a young founder is not about being brave. It is about being specific. Etched did not promise to be a bit better at everything. It chose the single assumption it could not hedge, and then waited years, quietly, to find out whether it was right.