Etched books $1 billion in contract orders for its AI inference chip at a $5 billion valuation after TSMC tape-out

AI chip startup Etched reported on June 30, 2026, that it has booked $1 billion in contract orders for its purpose-built transformer inference chip, following a successful manufacturing run at TSMC. The company, valued at $5 billion after closing a $500 million round in December 2025, is now testing complete inference systems with customers.

What's new

Etched's June 30 progress report reveals three milestones:

• TSMC tape-out successful: TSMC manufactured Etched's custom inference chip earlier in 2026, removing the key technical risk for a startup betting on custom silicon.
• $1 billion in contract orders: Enterprise customers have pre-committed $1B in orders for Etched's product — full inference cluster systems built around the custom chip, including custom racks and integrated software.
• Frontier inference cluster testing: Etched is currently testing these complete systems with a set of early customers, moving from tape-out to customer evaluation.

The company has raised $800 million in total funding, with a $500M round closed in December 2025 at the $5B valuation.

Context

Etched builds application-specific integrated circuits (ASICs) for transformer inference — silicon purpose-built for running large language models rather than adapted from general-purpose GPU designs. The core claim is that transformer-only ASICs can deliver substantially higher inference throughput per dollar than NVIDIA GPUs for production LLM workloads, because they omit the general-purpose compute overhead a GPU carries.

Etched is not alone in the dedicated inference ASIC space. Cerebras (which recently completed a $5.5B IPO) and Groq also compete in this category. But $1B in pre-committed orders represents a substantial commercial signal for a chip startup that, until recently, had not publicly confirmed manufacturing success.

Why it matters

The tape-out validation and $1B in contract orders are the two most important de-risking milestones for a semiconductor startup: manufacturing viability and commercial demand. Both arriving in the same progress report suggests Etched has crossed from "promising startup" to "real alternative" for AI inference procurement.

For enterprises building or expanding inference infrastructure, Etched's system — chips, custom racks, and software sold as an integrated product — represents a different procurement model than buying NVIDIA GPUs and assembling clusters independently. Whether it delivers on its cost-per-token claims at scale remains to be proven in customer deployments, but the $1B order volume suggests a significant number of buyers are willing to find out.

As inference costs remain one of the primary constraints on AI deployment economics, purpose-built silicon that reduces those costs is a direct input to how broadly organizations can deploy AI-powered products.