Etched Raises $500M at $5B for AI Inference Chips

Etched Takes on Nvidia in AI Inference

AI chip startup Etched has raised $500 million at a $5 billion valuation to build purpose-built inference hardware for transformer-based AI models. The company aims to challenge Nvidia's dominance by designing chips specifically optimized for AI inference workloads rather than the general-purpose GPU architecture that currently dominates the market.

Purpose-Built Inference Hardware

While Nvidia's GPUs excel at AI training, the inference market (running trained models in production) has different requirements: lower latency, higher throughput, and greater energy efficiency per computation. Etched's Application-Specific Integrated Circuits (ASICs) are designed from the ground up for transformer inference, potentially delivering 10-100x better performance-per-watt than general-purpose GPUs for this specific workload. As inference becomes the dominant cost in AI deployment, this advantage could capture significant market share.

The AI Semiconductor Arms Race

The $5B valuation reflects the enormous market opportunity in AI-specific semiconductors. Global spending on AI chips is projected to exceed $100 billion annually, with inference expected to account for the majority of that spending as more AI models are deployed into production. Etched's focused approach to inference optimization positions the company to capture a meaningful share of this market alongside, or at the expense of, Nvidia's GPU monopoly.