Etched Valued at $10.3 Billion: Its Chips Speed Up AI Models Without GPUs

Etched, an AI chip startup founded by three Harvard dropouts, reached a $10.3 billion valuation on July 23, 2026, after a funding round from major investors. The company built its own chips and memory modules that, by its own claim, accelerate inference on any AI model without using GPUs.

What exactly did Etched build

Etched did not just develop an accelerator. It built a combination of specialized chips and memory components for the inference stage, the point at which a trained neural network produces answers for the user. According to TechCrunch, the company's key promise is that its solution speeds up inference on any AI model while doing without graphics processors.

Key facts about the round and the product:

• Company valuation: $10.3 billion
• Founders: three Harvard dropouts
• Date of the news: July 23, 2026
• Product: custom chips plus memory modules for accelerating inference
• Main selling point: works without GPUs, with any AI model

"Accelerating inference on any AI model, without GPUs" is how

Etched itself states its main promise, as reported by TechCrunch.

Why this is a challenge for Nvidia

Etched is hitting the most sensitive spot in the market: the AI industry's dependence on Nvidia GPUs. Today the vast majority of the compute for training and running neural networks goes through GPUs, and Nvidia has built a market capitalization in the trillions of dollars on that demand. A startup that promises to accelerate inference without any GPUs is offering an alternative architecture for the part of the workload that costs the most in production.

The $10.3 billion valuation shows that major investors are willing to put money behind the idea. TechCrunch's headline says outright that Etched is "defying skeptics." In other words, the project had a reputation for being dubious, and the valuation it reached partly dispels those doubts.

Who is behind the company

Etched was founded by three people who left Harvard to start a company, a path familiar in the American tech sector that echoes the stories of Microsoft and Facebook. The bet on a young team with no degree but with working hardware paid off: a round from name-brand investors moved the company into the multibillion-dollar ranks.

It is important to understand the difference between training and inference. Training a model is a one-time, expensive operation, whereas inference is repeated for every user request, billions of times. Savings at this stage translate directly into money for anyone deploying AI products at scale, from chatbots to search assistants.

What this means

The emergence of a multibillion-dollar player betting on inference without GPUs is a signal that Nvidia's monopoly on AI hardware is no longer unconditional. If Etched's promises are confirmed in production, companies deploying AI services will get a cheaper alternative for the most high-volume part of their compute.

Frequently asked questions

What is Etched's valuation?

According to TechCrunch, as of July 23, 2026, Etched is valued at $10.3 billion after a funding round from major investors.

Do Etched's chips need GPUs?

No. By the company's own claim, its chips and memory modules accelerate inference on any AI model without using graphics processors.

*Meta is designated an extremist organization and banned in Russia.