Record Funding and Contract Scale
On July 1, AI inference chip company Etched exited stealth mode, revealing cumulative funding exceeding $800 million and customer contracts totaling more than $1 billion. The first rack systems are scheduled for delivery this summer. This funding scale places Etched among the highest-funded AI inference startups, reflecting strong market demand for specialized inference hardware.
Technological Milestones: A0 Tapeout, LVI and CSM
Etched has completed the tapeout of its first inference chip, A0, and built initial rack systems. Early customer testing shows that its inference system achieves state-of-the-art (SOTA) throughput, latency, and energy efficiency. The company also unveiled two key technologies: Low-Voltage Inference (LVI) for high-throughput scenarios, and Cluster-Scale Memory (CSM) for low-latency applications. These are designed to enhance inference performance for large-scale Mixture-of-Experts (MoE) models, long-context tasks, and AI agent workloads.
Team Strength and Production Capacity
Etched employs over 400 engineers, with core team members coming from NVIDIA, Google TPU, Broadcom, SK Hynix, and TSMC. To accelerate mass production, the company has built a 2MW datacenter and established a factory in Taiwan (China). Etched stated that more performance benchmarks and product roadmap updates will be released in the coming months.
Etched's progress underscores a broader industry shift: dedicated AI inference chips are moving from conceptual designs to large-scale commercial deployment, poised to play a critical role in next-generation AI computing architectures.