en.Wedoany.com Reported - Etched recently publicly demonstrated a production-ready AI inference chip and disclosed that it has signed over $1 billion in customer contracts. The company raised a total of $800 million across multiple previously undisclosed funding rounds, with the latest round completed in December 2025, securing $500 million at a $5 billion post-money valuation. Investors include VentureTech Alliance, Peter Thiel, Jane Street, Hudson River Trading, Jump Trading, Two Sigma, Stripes, Ribbit Capital, Radical Ventures, Primary VC, Positive Sum, and several prominent AI researchers and entrepreneurs. Etched stated that its chip achieved first silicon success (A0) on TSMC's N4P process and plans to begin shipping its first rack-scale inference systems this summer.
The company, founded less than three years ago, is focused on building vertically integrated AI inference infrastructure, encompassing custom chips, racks, networking, cooling, software, and manufacturing. Its systems are already running production-grade AI models, including DeepSeek, Qwen, Mamba, and Llama, and support architectures ranging from dense to large-scale Mixture of Experts (MoE) systems with arbitrarily large parameter counts. To secure manufacturing capacity, Etched has opened a factory in Taiwan and is building a 2-megawatt data center, testing chambers, and a New Product Introduction (NPI) prototype lab at its San Jose headquarters, with the goal of achieving gigawatt-scale deployments starting in 2027.
Etched also disclosed two differentiated architectural technologies. Low Voltage Inference (LVI) technology reduces the voltage of the chip's compute array to less than half that of traditional AI accelerators, which the company claims enables sustained utilization exceeding 80% of peak FLOPs in trillion-parameter sparse MoE inference without thermal throttling. Cluster-Scale Memory (CSM) architecture combines HBM with shared low-latency memory connected via proprietary interconnects, aiming to reduce inference latency while maintaining high throughput. According to Etched, early customer testing has demonstrated leading throughput, latency, and energy efficiency for inference workloads, with more performance data expected to be released later this summer.
Key development milestones include: raising a total of $800 million across multiple funding rounds, with the latest round in December 2025 securing $500 million at a $5 billion valuation; signing over $1 billion in customer contracts; achieving first silicon success on TSMC's N4P process; planning to ship the first rack-scale inference systems in summer 2026; systems now running models such as DeepSeek, Qwen, Mamba, and Llama; completing a 2-megawatt data center, testing facilities, and NPI prototype lab in San Jose, with engineering and manufacturing facilities in Taiwan; assembling a team of over 400 engineers from NVIDIA, Google TPU, Broadcom, SK hynix, TSMC, and quantitative trading firms; and targeting gigawatt-scale AI inference infrastructure by 2027.
Gavin Uberti, co-founder and CEO of Etched, stated: "We recognized early on that frontier AI would become one of the most economically significant technologies ever, but the infrastructure needed to serve these models sustainably and economically simply did not exist."
