Etched is an American semiconductor start-up that is designing custom ASICs for AI workloads; its first generation product is Sohu, the first chip purpose built for transformer AI architecture, the kinds of large language models that power not only ChatGPT, but Anthropic PBC’s Claude and Google LLC’s Gemini, as well as image generators such as DALL-E and Stable Diffusion.[2]
Company type | Private |
---|---|
Industry | Semiconductors |
Founded | 2022 |
Founders |
|
Headquarters | , |
Key people |
|
Products | Sohu (transformer ASIC) |
Number of employees | 35 (2024)[note 1] |
Website | www |
Founded in 2022 by three Harvard University drop-outs, Gavin Uberti, Chris Zhu and Robert Wachen[3], the company raised US$120 million in a Series A round in June 2024 to fabricate its first chips at TSMC.[1] The round was co-led by venture-capital firms Primary Venture Partners and Patrick O’Shaughnessy’s Positive Sum Ventures. Other investors include ex-GitHub CEO Thomas Dohmke, Peter Thiel, and Replit CEO Amjad Masad [4] [2].
History
editUberti began experimenting with compiler optimizations for AI workloads in 2022 and recruited classmates Zhu and Wachen to pursue a chip start-up.[5] The trio incorporated the company in Silicon Valley and became one of the first teams to collectively receive the Thiel Fellowship.[6]
In March 2023 the start-up closed a US$5.4 million seed round that valued the company at US$34 million.[1]
Technology
editSohu is fabricated on TSMC's 4-nanometre process node and hard-wires the matrix multiplication patterns specific to transformer inference.[5] Because the chip omits hardware needed for other neural-network types, Etched has claimed in their launch in June 2024 that an eight-chip Sohu server can generate more than 500,000 tokens per second on Llama-70B, versus roughly 23,000 tokens per second for an eight-GPU Nvidia H100 system.[7] Uberti has asserted that “one Sohu server replaces 160 H100 GPUs,” positioning the device as a lower-cost, lower-power competitor to Nvidia's Blackwell architecture.[8]
Funding
editDate | Round | Amount (US$) | Lead investors |
---|---|---|---|
March 2023 | Seed | 5.4 million | Various angels |
June 2024 | Series A | 120 million | Primary Venture Partners, Positive Sum Ventures |
Other participants in the Series A included Cruise Automation cofounder and CEO Kyle Vogt. [9]
Reception
editTrade press and analysts describe Etched as part of a “second wave” of AI accelerator companies that aim to undercut Nvidia's general-purpose GPUs.[8] CNBC characterised the firm's focus on transformers as “the biggest bet in AI,” noting that its survival depends on the continued dominance of that architecture.[10]
See also
editNotes
editReferences
edit- ^ a b c Cherney, Max A. (June 25, 2024). "AI startup Etched raises $120 million to develop specialized chip". Reuters. Retrieved August 8, 2025.
- ^ a b "Transformer model chipmaker Etched.ai raises $120M to challenge Nvidia's market dominance". SiliconANGLE. 2024-06-25. Retrieved 2025-08-28.
- ^ "Harvard Dropouts With Plan to Disrupt Nvidia Nab Peter Thiel Funding". Bloomberg.com. Archived from the original on 2024-03-22. Retrieved 2025-08-28.
- ^ "AI Chip Startup Etched Raises $120 Million to Expand Production". Bloomberg.com. Archived from the original on 2024-06-25. Retrieved 2025-08-28.
- ^ a b Wiggers, Kyle (June 25, 2024). "Etched is building an AI chip that only runs one type of model". TechCrunch. Retrieved August 8, 2025.
- ^ "Harvard Dropouts With Plan to Disrupt Nvidia Nab Peter Thiel Funding". Bloomberg.com. Archived from the original on 2024-03-22. Retrieved 2025-08-28.
- ^ Trueman, Charlotte (June 26, 2024). "AI chip startup Etched.ai raises $120 million to take on Nvidia". DataCenterDynamics. Retrieved August 8, 2025.
- ^ a b Wheatley, Mike (June 25, 2024). "Transformer-model chipmaker Etched.ai raises $120M to challenge Nvidia's market dominance". SiliconANGLE. Retrieved August 8, 2025.
- ^ Leswing, Kif (2024-06-25). "Harvard dropouts raise $120 million to take on Nvidia's AI chips". CNBC. Retrieved 2025-08-28.
- ^ "Etched raises $120 million to build chip to take on Nvidia in AI". CNBC. June 25, 2024. Retrieved August 8, 2025.