Etched

Etched

Etched is the AI chip company building Sohu — a transformer-specific application-specific integrated circuit (ASIC) designed to run transformer inference faster and cheaper than general-purpose GPUs. Etched's bet: now that transformers have stabilized as the dominant AI architecture, fixed-function silicon will outperform NVIDIA's flexible H100/B200 for transformer workloads. Etched raised approximately $120M Series A led by Primary Venture Partners.

Production credibility: $120M Series A led by Primary Venture Partners with Two Sigma Ventures, Skybox Data Centers, and others participating; founded 2022 by Gavin Uberti and Chris Zhu; Cupertino, California HQ. Sohu chip targets transformer-specific inference workloads at substantially lower cost per token than equivalent NVIDIA inference deployments. Backed by Peter Thiel, Stanley Druckenmiller, and other prominent individual investors.

Key Features

• Sohu — transformer-specific ASIC chip optimized for inference (not training)
• Founded 2022 by Gavin Uberti and Chris Zhu in Cupertino, California
• $120M Series A led by Primary Venture Partners
• Individual investors include Peter Thiel and Stanley Druckenmiller
• Targets transformer inference at lower cost per token than NVIDIA H100/B200
• Architecture bet: transformer dominance is durable enough to justify fixed-function silicon
• Compete with NVIDIA inference for hyperscale customers and AI-native inference clouds

Ideal Use Case

AI inference providers and hyperscalers serving transformer-heavy production workloads at scale — where the lower cost-per-token of dedicated silicon outweighs the flexibility loss vs NVIDIA's general-purpose GPUs. Particularly fits inference-only deployments where training happens elsewhere.

How Etched differentiates

NVIDIA's H100 and B200 are general-purpose AI accelerators — they run transformers, diffusion models, traditional ML, and arbitrary CUDA workloads. Etched's Sohu is the opposite — fixed silicon optimized for one architecture (transformers). The trade-off is flexibility vs efficiency: if transformers stay dominant for 5+ years, Sohu's cost-per-token advantage compounds; if a new architecture displaces transformers, Sohu is obsolete. The bet is the dominance question. Cerebras and Groq pursue different silicon strategies (wafer-scale and LPU respectively) but Etched is the purest transformer-only bet in the market.

FAQ

Q: What is Etched? A: Etched is an AI chip company building Sohu — a transformer-specific ASIC designed to run transformer inference faster and cheaper than general-purpose NVIDIA GPUs.

Q: Who founded Etched? A: Gavin Uberti and Chris Zhu co-founded Etched in 2022. The company is headquartered in Cupertino, California.

Q: How much has Etched raised? A: $120M Series A led by Primary Venture Partners with Two Sigma, Skybox Data Centers, Peter Thiel, and Stanley Druckenmiller participating.

Q: Etched vs NVIDIA H100 vs Groq vs Cerebras? A: NVIDIA H100 is general-purpose AI silicon. Groq builds LPUs optimized for inference latency. Cerebras builds wafer-scale chips. Etched's Sohu is the most architecture-specific bet — transformer-only ASIC, lowest cost-per-token if transformers stay dominant.

Q: Is Sohu shipping? A: Sohu has been announced and is in customer engagement; production volume timing depends on customer ramp.

tl;dr

Etched builds Sohu — a transformer-specific ASIC chip. $120M Series A (Primary Venture Partners, Peter Thiel, Stanley Druckenmiller). Bet that transformer dominance is durable enough to justify fixed-function silicon. Most architecture-specific AI chip bet in the market vs general-purpose NVIDIA / Groq LPU / Cerebras wafer-scale.

Related

Looking for more options? Browse the AI Infrastructure directory or read our best AI infrastructure tools listicle. Etched is also tracked on Crunchbase.