Etched AI chip $10.3B US valuation 2026: Etched, a US AI chip startup founded by three Harvard dropouts, raises $300M Series C at $10.3B valuation led by Sequoia for AI inference chips, with $1B in booked orders from US customers.<\/strong> The company has successfully manufactured its chips and is testing first systems with US clients, marking a significant milestone in the race to build specialized AI hardware for transformer models. Etched’s valuation trajectory\u2014from $2B in 2025 to $10.3B in 2026\u2014reflects growing investor confidence in dedicated AI inference chips for the US market.<\/p>\n Etched announced a $300 million Series C on July 23, 2026, at $10.3 billion valuation. Sequoia led with Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital participating. This is the highest valuation ever for a Sequoia-led Series C, surpassing Stripe’s $9.2B Series I in 2022.<\/p>\n The rapid valuation increase reflects Etched’s progress in manufacturing and commercialization. The company raised $80 million Series B in February 2025 at $2 billion valuation, then $220 million Series B+ in December 2025 at $5 billion. The $300 million Series C brings total funding to $600 million. This development is significant for Etched AI chip $10.3B US valuation 2026.<\/p>\n CEO Gavin Uberti attributed the oversubscribed round to strong demand from US data center operators and AI companies seeking alternatives to Nvidia’s expensive GPU clusters. Etched’s chips offer 3-5x better performance per dollar for transformer inference workloads, according to third-party benchmarks.<\/p>\n Etched builds custom AI inference chips designed specifically for transformer models\u2014the architecture powering GPT-4, Claude, and other large language models. Unlike GPUs that handle many types of workloads, Etched’s chips optimize exclusively for transformer computations, offering US companies lower cost per inference.<\/p>\n The company’s first chip, Sohu, is manufactured on TSMC’s 5nm process and contains 140 billion transistors. Sohu delivers 500,000 tokens per second for Llama 70B inference, compared to 50,000 tokens per second on Nvidia H100 GPUs. This 10x performance improvement at one-third the cost makes Etched’s chips attractive for US AI companies processing millions of daily requests.<\/p>\n Etched’s architecture uses a systolic array design optimized for matrix multiplications common in transformer models. The chips include on-chip SRAM for low-latency access and support for FP8, INT8, and INT4 precision formats used in modern AI inference.<\/p>\n In June 2026, Etched successfully manufactured its chips with first systems being tested by US clients. The company has $1 billion in booked orders from US data center operators, AI companies, and cloud providers. Manufacturing is handled by TSMC in Taiwan, with packaging and testing in the US.<\/p>\n The manufacturing milestone is significant because many AI chip startups have failed to move from design to production. Etched’s ability to manufacture at scale validates its technical approach and positions the company to compete with established players like Nvidia and AMD in the US market.<\/p>\n First production systems are being deployed at three US data centers for testing, with general availability planned for Q4 2026. Etched has signed supply agreements with TSMC for 2027 production capacity, ensuring it can meet growing US demand for AI inference chips.<\/p>\n Etched’s valuation has grown 5x in 18 months, reflecting the AI chip market’s explosive growth and strong investor confidence in the company’s technology.<\/p>\nTable of Contents<\/h2>\n
\n
Funding Details<\/h2>\n
What Etched Does in US<\/h2>\n
Manufacturing Milestone<\/h2>\n
US Valuation Trajectory<\/h2>\n