Etched company profile: valuation, funding & revenue

This Etched company profile covers the company's valuation, revenue, funding rounds, investors and history. Etched works in semiconductors, with a focus on AI inference chips (transformer ASICs). It was founded in 2022 and is headquartered in San Jose, California, United States. Its latest valuation on record is US$10.3 billion (July 2026). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.

Revenue

Funding history

Etched timeline

Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.

• 2022 — Etched is founded in San Jose by Gavin Uberti (CEO), Chris Zhu and Robert Wachen.
• Dec 2025 — Equity round: US$500M, valuing the company at US$5B. Investors: Stripes (lead).
• Jul 2026 — Series C: US$300M, valuing the company at US$10.3B. Investors: Sequoia Capital (lead); Andreessen Horowitz, Jane Street, Diffusion.
• 2026 — Booked orders: More than US$1 billion in booked customer orders (company-stated).
• August 2026 — Later report: A single report described a further round at about US$21 billion in August 2026; not independently confirmed.

Investors

Overview

Etched designs AI chips that do one thing: run transformer models, the neural network architecture behind ChatGPT, Claude, Gemini, Llama and most modern AI. Its chip, called Sohu, is an application-specific integrated circuit (ASIC) that hard-wires the transformer architecture into silicon. By giving up the flexibility to run other types of AI models, Etched argues it can serve transformer models far faster and more cheaply than general-purpose GPUs such as Nvidia's. The company has become one of the most talked-about challengers in AI hardware, with a valuation that rose from about US$5 billion to more than US$10 billion within months.

Gavin Uberti, the CEO, Chris Zhu and Robert Wachen founded Etched in 2022 after leaving Harvard University. Their bet, made before ChatGPT's launch, was that transformers would become the dominant AI architecture for years, making a specialised chip worthwhile. Etched is headquartered in San Jose, California, in Silicon Valley's semiconductor heartland.

The Sohu bet

GPUs are programmable processors that can run almost any parallel computation, from graphics to scientific simulations to many kinds of neural networks. That flexibility costs space and energy on the chip. Etched's Sohu removes everything not needed for transformer inference, devoting more of the silicon to the specific matrix calculations transformers require. Etched has claimed that a server of Sohu chips can generate far more tokens per second than comparable GPU servers. These claims haven't yet been independently verified at scale, and the company has said it expected to ship its first racks in 2026. If transformers are replaced by a different architecture, a transformer-only chip would lose its advantage; that's the central risk of the strategy.

Funding

Etched raised a US$120 million Series A in 2024, with backing from Primary Venture Partners, Positive Sum and investors including Peter Thiel. In December 2025 it raised about US$500 million at a US$5 billion valuation, led by Stripes, and said it had more than US$1 billion in booked customer orders. In July 2026 it raised a US$300 million Series C at a US$10.3 billion valuation, led by Sequoia Capital, with Andreessen Horowitz, Jane Street and Diffusion, according to Reuters as cited by Startup Fortune. One later report described a further round at about US$21 billion in August 2026; because we found only a single source, we list it separately and use US$10.3 billion as the valuation.

Business model

Etched sells servers and racks built around Sohu chips to AI labs, cloud providers and large companies running inference at scale. Revenue depends on delivering chips on schedule and in volume, which requires partnerships with chip foundries such as TSMC, memory suppliers and server manufacturers. The company hasn't disclosed revenue, so we show it as not disclosed; booked orders indicate demand but aren't the same as delivered revenue.

Competition

Etched competes primarily with Nvidia, whose GPUs and CUDA software dominate AI computing, and with AMD. Other AI chip companies include Groq, which licensed its technology to Nvidia in a large deal, Cerebras, which went public in 2026, SambaNova, Tenstorrent and photonic interconnect specialist Lightmatter. Large technology companies also design their own chips: Google's TPUs, Amazon's Trainium and Inferentia, Microsoft's Maia and Meta's MTIA. Our guide to AI chip startups compares these approaches.

Etched at a glance

Etched is a San Jose AI chip company founded in 2022 by Gavin Uberti, Chris Zhu and Robert Wachen. Its Sohu chip is built only to run transformer models. It raised about US$500 million at US$5 billion in December 2025 and US$300 million at US$10.3 billion in July 2026, led by Sequoia, and says it has more than US$1 billion in booked orders.

Timeline

• 2022: founded by three Harvard students who left university.
• June 2024: announces the Sohu chip and a US$120 million Series A.
• December 2025: raises about US$500 million at US$5 billion, citing more than US$1 billion in booked orders.
• 2026: expects to ship its first Sohu racks.
• July 2026: raises US$300 million at US$10.3 billion, led by Sequoia.

Why inference chips are attracting so much money

As AI products scale to hundreds of millions of users, the cost of inference, running models to answer requests, has become the largest part of AI computing spending. Reasoning models and agents, which generate many tokens per task, multiply that demand. Companies are eager for alternatives to Nvidia that could lower the cost per token, reduce power use and ease supply constraints. Investors see a chance that specialised inference chips could capture part of a market worth hundreds of billions of dollars a year. That's why companies like Etched can raise large rounds before shipping products in volume.

The software challenge

Nvidia's strength lies not only in its hardware but in CUDA, the software platform that developers have used for nearly two decades, along with libraries optimised for every major AI framework. Any new chip must make it easy to run popular models without months of engineering work. Etched's focus on transformers simplifies this: because it supports only one architecture, it can optimise its software stack tightly for the most popular open and closed models. Whether customers find it easy to deploy will be an important test.

Risks

• Architecture risk: a shift away from transformers would undermine the chip's value.
• Execution: first-generation chips often face delays, yield problems or performance below expectations.
• Nvidia's pace: Nvidia releases new generations rapidly and can cut prices.
• Supply chain: access to advanced manufacturing and high-bandwidth memory is constrained.

Questions for buyers

• What independent benchmarks show performance on your models?
• How easily can your models be deployed on the chip?
• What delivery schedules and support can the vendor commit to?
• What happens if your workloads move beyond transformers?

Latest developments (October 2026)

Etched's valuation rose from about US$5 billion in early 2026 to US$10.3 billion after a July 2026 round, as described above. The company says its Sohu chip, built only for transformer AI models, is manufactured on TSMC's 4-nanometre-class process, that first systems were scheduled to ship in summer 2026, and that it has more than US$1 billion of pre-orders. Independent benchmarks were not yet available in the sources we reviewed. See our AI chip startups guide.

What to watch

Watch independent performance results, volume shipments, named customers, confirmation of further funding and Nvidia's response with its own inference-focused products.

Takeaways

AI infrastructure buyers may gain a cheaper inference option if Sohu delivers. Founders can see the payoff and risk of an early, focused technical bet. Investors should weigh booked orders against unproven volume production.

Sources

Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.

1. Startup Fortune (citing Reuters)
2. Inbenta
3. Kami Labs