Etched Exits Stealth With Working AI Inference Chip

Etched exited stealth on June 30, 2026, revealing a working AI inference chip built on TSMC's N4P process that already runs models like DeepSeek and Llama. The San Jose startup has raised $800 million from investors including Peter Thiel and Geoffrey Hinton, secured over $1 billion in customer contracts, and plans to ship first racks this summer, challenging Nvidia's dominance in AI inference.

For AI practitioners watching inference costs eat into unit economics, a credible new silicon challenger to Nvidia matters more than another frontier model. Etched, the San Jose startup behind the long-rumored transformer-focused chip effort, came out of stealth on June 30, 2026 with a working part rather than a roadmap: first-pass A0 silicon success on TSMC's N4P process, rack-scale systems already running DeepSeek, Qwen, Mamba, and Llama, and, per the company, over $1 billion in signed customer contracts. Etched said it has raised $800 million across multiple previously undisclosed rounds, the latest a $500 million round at a $5 billion post-money valuation in December, from investors including Peter Thiel, Jane Street, Two Sigma, Ribbit Capital, and researchers Andrej Karpathy, Geoffrey Hinton, and Fei-Fei Li. First racks are slated to ship this summer, with the company claiming state-of-the-art throughput, latency, and power efficiency on inference workloads.