Etched Lands $700M to Scale AI Inference Chips

The big picture: Etched, a US-based startup specializing in processors built to run trained AI models, has raised a $700 million funding round led by Jane Street, doubling its valuation to $21 billion within a month. Jane Street not only led the funding but also became Etched’s first production customer by operating its own rack within the company. Etched has received orders totaling over $1 billion.

Why it matters:

• Rapid Valuation Growth: Etched’s valuation doubled to $21 billion in less than a month, underscoring intense investor confidence in specialized AI hardware solutions.
• Strategic Customer Validation: Lead investor Jane Street’s role as Etched’s first production customer provides significant validation for its inference technology in demanding, latency-sensitive workloads.
• Market Consolidation: The funding strengthens Etched’s position as one of the few remaining independent companies in the AI chip market, following the acquisition or IPO of key competitors.

How it works:

• Inference Specialization: Etched designs processors specifically for running already-trained AI models, based on the premise that inference will eventually require more computing power than training.
• Two-Stage Processing: The company’s approach involves a compute-intensive prefill phase for prompt understanding and a memory-intensive decode phase for generating responses.
• Low-Voltage Design: Etched’s prefill chip uses less than half the voltage of ordinary AI accelerators, allowing for more transistors without overheating and improving responses per rack for latency-sensitive applications.

The catch: Etched’s $21 billion valuation is largely based on over $1 billion in orders and market potential rather than established revenue. The company must now convert these contracts into operational customer clusters beyond its investor base to justify its rapid growth and high market capitalization in a competitive, consolidating AI chip landscape still dominated by Nvidia.