Farrsal
Financial news and analysis

Global insights for better decisions.

Economic, market and geopolitical coverage with clear context.

FarrsalNews for a Smarter Tomorrow
Financial news and analysis•5 min readPage 1 of 4

AI Chip Startup Etched Eyes $20 Billion Valuation in Latest Funding Round

A Bold Ask in Silicon Valley Etched, a startup developing specialized processors for artificial intelligence, is reportedly targeting a $20 billion valuation in…

Farrsal Editorial
Farrsal · News for a Smarter Tomorrow
f𝕏WAin
AI Chip Startup Etched Eyes $20 Billion Valuation in Latest Funding Round
Article image: AI Chip Startup Etched Eyes $20 Billion Valuation in Latest Funding Round

Article Highlights

Excerpts from the article

  • Etched, a startup developing specialized processors for artificial intelligence, is reportedly targeting a $20 billion valuation in an upcoming…
  • The Cupertino- and San Jose-based firm is betting that the future of AI inference belongs not to general-purpose graphics processors, but to silicon…
  • Etched was founded in 2022 by Harvard dropouts Gavin Uberti and Chris Zhu. The pair set out to solve what they saw as a fundamental inefficiency in…

A Bold Ask in Silicon Valley

Etched, a startup developing specialized processors for artificial intelligence, is reportedly targeting a $20 billion valuation in an upcoming funding round, according to a Wall Street Journal report cited by Investing.com. The fresh capital push would mark a dramatic leap for a company that was valued at just $5 billion barely a year ago.

The Cupertino- and San Jose-based firm is betting that the future of AI inference belongs not to general-purpose graphics processors, but to silicon designed exclusively for transformer models. That thesis is now about to face its biggest test yet.

From a Dorm Room to a $5 Billion Valuation

Etched was founded in 2022 by Harvard dropouts Gavin Uberti and Chris Zhu. The pair set out to solve what they saw as a fundamental inefficiency in the AI chip market — running transformer-based workloads on ever-larger GPUs that were never purpose-built for that task.

Their answer is Sohu, a chip manufactured by TSMC on a cutting-edge 4-nanometer process. Unlike Nvidia's H100 and B200 accelerators, which are designed to handle a wide variety of parallel computations, Sohu is optimized for the matrix multiplication and attention mechanisms central to transformer models. The result, the company argues, is dramatically better performance per watt and per dollar for the kind of inference that powers models like GPT-4, Claude, and Gemini.