AI chip startup Etched doubles to $10.3B valuation in under a month, armed with $1B in contracts

Etched doubles to $10.3B as $1B in orders tests whether Nvidia's inference monopoly holds

Etched, the AI chip startup, has closed a $300 million Series C at a $10.3 billion valuation 1. Sequoia led the round, with SK Hynix, Jane Street, and Diffusion Capital among the participants. Previous backer Andreessen Horowitz also joined 1. Etched says this marks the highest valuation ever for a Sequoia-led Series C 1.

The figure more than doubles the $5 billion valuation from Etched's prior round 1. Etched is separately in talks for a round that would value it at $20 billion 2.

But the valuation is not the story. The story is $1 billion in booked customer orders for Etched's custom inference processors 1. Companies that run AI models at scale are committing real money to silicon that does not come from Nvidia.

That is a market-structure shift. Each dollar of those orders represents a customer calculating that the cost of depending entirely on Nvidia for inference compute exceeds the risk of backing a supplier whose chips have barely left the lab. The investors pricing Etched at $10.3 billion, and reportedly eyeing $20 billion next, are running the same math from the other side: the inference monopoly is breakable, or at least someone will pay $20 billion to find out.

The cap table adds operational weight to the financial bet. SK Hynix is an investor 1. Bloomberg reported in late June that Jane Street and a venture fund linked to TSMC had also backed the company 2. TSMC manufactured Etched's first batch of silicon 1. Having your foundry and your memory supplier on the cap table is how a company without shipping product earns credibility that financial backing alone cannot buy.

Individual backers include Peter Thiel, Andrej Karpathy, Dylan Field, and Amjad Masad 1. COO Robert Wachen told TechCrunch that these investors, alongside figures from OpenAI and Anthropic, tested the hardware through private demonstrations before committing 1.

Etched's chips are designed to accelerate inference, the computational work that happens after a prompt is submitted and the model generates its reply 1. The company built two components from scratch for the two stages of inference. The first processes prompts at lower voltage than any competing AI chip, according to Wachen, who says the approach generates less heat and packs in more transistors. The second is a new memory architecture that connects many chips to a shared pool for generating output tokens 1. The systems can run any AI model, including Mixture of Experts designs like DeepSeek and Qwen and non-transformer architectures such as Mamba 1.

The skepticism has not disappeared. Etched originally pitched its chips as purpose-built for transformer models, the architecture behind ChatGPT and Claude, and critics questioned whether specialized silicon could keep pace as model designs evolve 1. Access remains limited to investors and early customers, which has kept doubts alive 1. The company says it is processing tokens in its lab with leading AI companies, but mass production and delivery are still ahead 1. Wachen told TechCrunch the founders remain humbled by what it will take to reach scale 1.

Three years in, Etched has grown to 400 employees and operates a 2-megawatt data center 1. The founders started the company from a garage after leaving Harvard, rebooting their design servers by hand 1.

A chip company with limited production, $1 billion in booked orders, and a valuation arc from $5 billion toward a potential $20 billion inside a year is telling you where inference compute is heading. Capital and customers are arriving at the same conclusion from opposite ends: remaining the last company paying Nvidia's margins indefinitely costs more than funding the alternative.