Technology

AI Chip Startup Etched Raises $300M at $10.3B Valuation

Martin HollowayPublished 2w ago4 min readBased on 5 sources
Reading level
AI Chip Startup Etched Raises $300M at $10.3B Valuation

AI chip startup Etched has closed a $300 million Series C round at a $10.3 billion valuation, with Sequoia leading the investment. Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital also participated in the round, which Etched says marks the highest valuation ever for a Sequoia-led Series C (TechCrunch).

The $10.3 billion figure effectively doubles the company's December 2025 valuation of $5 billion, achieved when Etched raised a $500 million round. That earlier round followed a $120 million raise in June 2024 and a $5.4 million seed round in March 2023 (Reuters, TechCrunch). In roughly seven months, the company's paper value has doubled.

Etched was founded in 2022 by three Harvard dropouts: CEO Gavin Uberti, COO Robert Wachen, and CTO Chris Zhu. At least one co-founder is a Harvard Thiel Fellow, and at least one co-founded Prod, described as having a $100B+ cohort valuation. Another co-founder venture includes Mentor Labs, which was acquired by Crimson Education (etched.com). Individual backers include Peter Thiel, Andrej Karpathy, Dylan Field, and Amjad Masad (TechCrunch).

A venture capital firm linked to TSMC has also invested in Etched, a detail first reported in late June (Bloomberg).

On the technical side, Etched has designed a custom chip for the "prefill" stage of AI inference, the initial phase where a model processes the input prompt before generating output. This chip runs at lower voltage than other AI accelerators, an approach the company calls low-voltage inference. Etched has also built a memory and interconnect technology called cluster scale memory, aimed at the "decode" phase, where the model generates tokens one at a time. By optimizing each phase separately rather than running both on a general-purpose GPU, the company aims to improve throughput and power efficiency for transformer-based workloads.

The company says it has successfully manufactured its own chips and that first full systems were being tested by clients as of June 2026. Etched had booked $1 billion worth of orders as of that month (TechCrunch). The TechCrunch report is based on a direct interview with co-founder and COO Robert Wachen, not a company blog post or press release.

The broader context here is an AI accelerator market where Nvidia's dominance remains largely unchallenged by credible alternatives. Etched is one of several startups betting that inference-specific silicon, purpose-built for transformer architectures rather than offering the flexibility of a general-purpose GPU, can deliver enough cost-per-token advantage to win deployments. SK Hynix's participation in the Series C is notable: the Korean memory giant brings expertise in high-bandwidth memory (HBM), a technology directly relevant to any inference architecture that pushes memory bandwidth and interconnect performance.

The TSMC-linked VC investment similarly signals foundry alignment, which matters for a company that has moved from design to manufactured silicon in roughly three years. Whether Etched's $1 billion order book converts to deployed revenue at scale is the question that will determine whether the $10.3 billion valuation holds. For now, the company has moved from design files to manufactured chips to client testing in a timeframe that is fast even by the compressed standards of the current AI hardware cycle.

The participation of Jane Street alongside strategic investors like SK Hynix is worth noting: quantitative trading firms have been among the most aggressive consumers of custom silicon for low-latency inference, and Jane Street's interest in Etched may reflect that demand profile as much as conventional venture returns.

In my view, the valuation trajectory from $5.4 million in seed funding to a $10.3 billion valuation in roughly three years places Etched in rare company, but the hard part of any chip startup is not raising capital or taping out silicon. It is shipping production volumes, hitting yield targets, and sustaining performance advantages against an incumbent that reinvests billions annually into its own roadmap. The next twelve months, with client testing already underway and orders on the books, will be the period that matters.