Startups & Funding

Etched hits $10.3B valuation with $300M Series C led by Sequoia

Harvard dropouts build custom AI chips that beat Nvidia on inference speed and cost.

Deep Dive

Etched, the AI chip startup founded by Harvard dropouts Gavin Uberti, Robert Wachen, and Chris Zhu, has closed a $300 million Series C funding round at a $10.3 billion valuation, nearly doubling its $5 billion valuation from December. The round was led by Sequoia, with participation from Andreessen Horowitz, SK Hynix, Jane Street, Diffusion Capital, and prominent angel investors like Peter Thiel, Andrej Karpathy, and Dylan Field. The company claims this is the highest valuation ever for a Sequoia-led Series C. Etched's chips, manufactured by TSMC, are designed to accelerate inference—the compute process after a user submits a prompt—by splitting it into two stages: prefill (understanding the prompt) and decode (generating output). The company created a low-voltage prefill chip that runs dramatically faster and a new cluster-scale memory interconnect for fast, low-latency decoding across many chips. Etched's systems can run any AI model, including Mixture of Experts (e.g., DeepSeek, Qwen) and non-transformer designs like Mamba. The startup has already booked $1 billion in orders and is testing its first full systems with clients. Despite skepticism from those who doubted a transformer-specific chip approach, Etched's investors and early customers—including AI luminaries like Geoffrey Hinton and Noam Brown—have been impressed by private demos. The founders acknowledge the road to mass production remains challenging but are confident in their technology's ability to lower costs and speed up AI inference.

Key Points
  • Etched raised $300M at a $10.3B valuation, doubling in seven months, led by Sequoia with participation from Andreessen Horowitz, SK Hynix, and others.
  • The startup designed custom chips for AI inference with low-voltage prefill and cluster-scale memory, claiming faster speeds and lower costs than Nvidia.
  • Etched has already secured $1B in orders, and its silicon is manufactured by TSMC; systems can run any model including Mixture of Experts and Mamba.

Why It Matters

Custom AI inference chips could slash costs and energy use, challenging Nvidia's dominance in the data center.

📬 Get the top 10 AI stories daily