The All-or-Nothing Wedge
Etched’s bet is an application-specific integrated circuit (ASIC) designed exclusively for transformer inference. By stripping away the general-purpose silicon needed to run everything from video games to scientific simulations, the company claims it can achieve radical efficiency. A server with eight Sohu chips can generate over 500,000 tokens per second on a Llama 70B model [Etched X, Jun 2024]. The company says that performance replaces a cluster of 160 Nvidia H100 GPUs [Etched X, Jun 2024].
Funding and Leadership
Etched’s co-founders, Gavin Uberti, Chris Zhu, and Robert Wachen, are all Thiel Fellows [Rambus, 2024]. Uberti previously worked on compiler technology at OctoML. The technical team includes chip veterans like Chief Architect Saptadeep Pal [LinkedIn, 2026]. After a $23 million seed in 2023, Etched raised a $120 million Series A in mid-2024 led by Primary Venture Partners and Positive Sum Ventures [TechCrunch, Jun 2024]. A reported $500 million round at a $5 billion valuation was led by Stripes [Data Center Dynamics, 2025].
Evidence of Demand
Etched has stated that unnamed customers have reserved "tens of millions of dollars" worth of its hardware [TechCrunch, Jun 2024]. Recent hires like Chase Holmes suggest a push to land enterprise deals [Primary VC Job Board, 2026].
Risks
- Architectural Shift. If a significantly more efficient AI model architecture emerges that isn't a transformer, Etched’s chip becomes a very expensive paperweight.
- Execution. Designing and fabricating a cutting-edge chip is a multibillion-dollar game of perfect. Etched is working with TSMC to produce Sohu [AIBase], but any yield issues or delays could be fatal.
- The Ecosystem Moat. Nvidia’s dominance is supported by the CUDA ecosystem. Etched must convince customers to port their models to a new, unproven stack.
The Next Twelve Months
For Etched, 2025 is the year of proof. The reported $500 million war chest must translate into volume production, public customer announcements, and real-world benchmarks. The key milestone to watch is the first named hyperscaler or major AI lab going into production with Sohu servers.