Etched: The AI Chip Company Now Valued at $10.3 Billion

From a Mattress-Free Floor to a Ten-Billion-Dollar Valuation

When three students left Harvard in 2022 to found an AI chip company, nobody took them entirely seriously. They didn't know how to raise funding, how to hire engineers, or even where they would live in Silicon Valley. Today, Etched announces the close of a Series C funding round worth $300 million, lifting its valuation to $10.3 billion — double what it was just seven months ago.

An Investor Lineup That Tells the Success Story

Sequoia Capital led the round, joined by notable names including Andreessen Horowitz, Korean memory giant SK Hynix, and trading firm Jane Street. Alongside institutional investors, high-profile individuals such as Peter Thiel, Andrej Karpathy, and Geoffrey Hinton also put money into the company — all of whom tested the hardware hands-on before making their investment decisions, according to co-founder and COO Robert Wachen.

Notably, this valuation makes Etched the holder of the highest valuation in the history of Sequoia-led Series C rounds — a remarkable figure in the fast-moving AI chip market.

What Exactly Does Etched Build?

The company manufactures integrated computing systems — not individual chips — optimized for the inference phase: the computational process that occurs after a user sends a request to a model. This phase breaks down into two components:

• Prefill phase: Analyzing context and understanding the request — a computationally intensive operation. Etched developed a dedicated chip for this stage that runs at very low voltage, reducing heat generation and allowing a greater number of transistors to be packed in.
• Decode phase: Generating the text response for the user, which requires massive memory capacity. Here, the company developed a group-level shared memory technology that links multiple chips in a single low-latency network.

The stated result: higher speed and lower operating costs compared to conventional solutions.

Correcting a Common Misconception

Wachen addresses a misconception that has followed the company since its founding — the idea that its chips only support specific language models. In reality, the systems work with a wide range of models, including:

• Mixture of Experts models such as DeepSeek and Qwen, which distribute tasks across specialized sub-models.
• Models not based on the Transformer architecture, such as Mamba, which is built on a State-Space Model.

In the same vein, the idea of embedding AI model components directly into the chip is no longer an exotic experiment; Google is reportedly pursuing a similar approach with its experimental Frozen v2 chip designed for the Gemini model.

From a Garage to a Ten-Megawatt Data Center

The road was far from smooth. In the early days, the founders ran chip design servers out of an early employee's home garage, with his wife manually rebooting them whenever they needed a restart. Wachen slept on the bare floor of a friend's unfurnished home, using a towel in place of a blanket.

Today, the company employs 400 people, and its San Jose headquarters houses a 2-megawatt data center. It recently opened a new 80,000-square-foot facility with 10 megawatts of capacity in nearby Milpitas. The company has also previously announced the successful manufacture of its first chips through partner TSMC and has secured $1 billion in orders.

A Lesson in Perseverance

Wachen closed with words that encapsulate the company's journey: "When you truly believe something is possible and keep working on it long enough, you can make it happen." Today he owns a mattress and a pillow — several pillows, in fact — but more importantly, he owns a company whose chips are now running in production with some of the world's leading AI firms.

✦ بقلم فريق دروب أيديا