Стартап Etched оценили в $10,3 млрд: его чипы ускоряют ИИ-модели без GPU

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

What exactly did Etched build

Etched developed not just an accelerator, but a combination of specialized chips and memory components for the inference stage — the moment when a trained neural network generates responses 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 students who dropped out of Harvard
• News date — July 23, 2026
• Product — proprietary chips plus memory modules for accelerating inference
• Main feature — works without a GPU, with any AI model

"Accelerating inference on any AI model — without a GPU," is how

Etched itself formulates its main promise, according to TechCrunch.

Why this is a challenge for Nvidia

Etched strikes at the market's most sensitive spot — the AI industry's dependence on Nvidia's graphics processors. Today, the overwhelming majority of computations for training and running neural networks run through GPUs, and it is precisely on this demand that Nvidia built a market capitalization worth trillions of dollars. A startup promising to accelerate inference entirely without a GPU offers an alternative architecture for the part of the workload that is the most expensive in production.

The $10.3 billion valuation shows that major investors are willing to bet money on this idea. TechCrunch's headline directly notes that Etched "defies skeptics" — meaning the project had a reputation for being dubious, and the valuation it achieved partially removes those doubts.

Who is behind the company

Behind Etched are three founders who left Harvard for the startup — a classic path for the American tech sector, echoing the stories of Microsoft and Facebook. The bet on a young team without a degree, but with ready-made hardware, paid off: a funding round from name-brand investors pushed the company into multi-billion-dollar territory.

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

What this means

The emergence of a multi-billion-dollar player betting on GPU-free inference 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 gain a cheaper alternative for the most massive part of the computation.

Frequently asked questions

What is Etched's valuation?

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

Do Etched's chips need GPUs?

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

*Meta is recognized as an extremist organization and is banned in Russia.