Etched has emerged from stealth with a working inference chip and over $1 billion in signed customer contracts. Since its seed round less than three years ago, the startup hit first-pass (A0) silicon success using TSMC’s N4P process. Now it’s testing a rack-scale system designed for both prefill and decode workloads, running models like DeepSeek, Qwen, Mamba and Llama at scale. The company raised $800 million across multiple undisclosed rounds, including a $500 million tranche in December that valued it at $5 billion post-money. Backers include VentureTech Alliance, Peter Thiel, Jane Street, Hudson River Trading, Jump Trading, Two Sigma, Stripes, Ribbit, Radical Ventures, Primary VC, Positive Sum and AI figures such as Andrej Karpathy, Geoffrey Hinton and Fei-Fei Li. That group signals a major foundry tie-up with one of the world’s top semiconductor manufacturers. Etched’s design teams co-developed hardware and software to drive inference throughput, cut latency and boost energy efficiency. The goal is to support models of any size without bottlenecks, making AI inference faster and cheaper. CEO Gavin Uberti says they saw early on that existing infrastructure couldn’t meet future demand for AI at scale. To meet that demand, Etched is ramping production in Taiwan and has built a data center, test house and NPI prototyping lab at its San Jose HQ. The company plans to push toward gigawatt-scale manufacturing by 2027, aiming to fill contracts and get its rack-scale systems into customer deployments soon.