Most hardware startups spend years trying to convince investors their chip can do more things. Etched has spent three years trying to convince investors its chip should do exactly one thing — and that bet just got a lot more expensive to ignore.
The Numbers
TechCrunch reported August 18 that Etched’s valuation has doubled to $21 billion in a single month, an increase of nearly $11 billion. That follows an already rapid climb: the company exited stealth on June 30 at a reported $5 billion valuation with $800 million raised and $1 billion in signed customer contracts, then closed a $300 million Series C led by Sequoia on July 23 at $10.3 billion — the highest valuation Sequoia had ever backed at that stage. Less than a month later, investors have doubled the number again.
Backers across Etched’s funding history include Sequoia, Andreessen Horowitz, Kleiner Perkins, Jane Street, Peter Thiel, Tiger Global, Bain Capital Ventures, and Blackstone, alongside individual investors like Nobel laureate Geoffrey Hinton, Stanford AI researcher Fei-Fei Li, and former Tesla Autopilot lead Andrej Karpathy.
The Bet: One Chip, One Job
Etched’s core product is Sohu, an application-specific integrated circuit — an ASIC, as opposed to a general-purpose GPU — designed to do exactly one thing: run inference for transformer models, the architecture underlying nearly every major AI system today, including ChatGPT, Claude, and Gemini. Unlike Nvidia’s GPUs, built to flexibly handle a wide range of workloads, Sohu can’t train models and can’t run any architecture other than transformers. CEO Gavin Uberti has framed that narrowness as the point: “The infrastructure required to serve frontier AI sustainably and economically was never going to come from incremental improvements to existing hardware.” Etched claims Sohu delivers 10 to 100 times better performance-per-watt than Nvidia’s H100 for inference.
What Changed This Round
Co-founder and COO Robert Wachen told TechCrunch that this latest valuation jump is specifically tied to two new components Etched designed from scratch to speed up the two distinct stages of AI inference: the compute-intensive “prefill” phase, where a system processes and understands an incoming prompt, and the memory-intensive “decode” phase, where it actually generates the output tokens a user sees. Etched built a dedicated prefill chip that runs at low voltage, which the company says lets it pack in significantly more transistors without the heat problems that typically come with high-end AI chip designs.
Trading firm Jane Street, an investor and early customer, offered a concrete endorsement in Etched’s funding announcement: “We tested the chip and are pleased with the early results… We’re excited to now have our own rack running in our datacenter” — a real deployment, not just a funding commitment.
The Real Risk Investors Are Pricing
Etched’s entire value proposition rests on one architectural bet holding: that transformers remain the dominant way frontier AI models get built. If the field shifts to a fundamentally different architecture — something researchers periodically explore — a chip that can only run transformers loses its reason to exist almost overnight, unlike a flexible GPU that could run whatever comes next. That’s a meaningfully different risk profile than most hardware bets, and it’s why Etched’s rapid valuation climb has drawn as much scrutiny as excitement.
Why This Fits the Broader Moment
Etched’s rise is part of a broader surge of investor interest in inference-specific hardware as AI companies shift spending from training new models to serving existing ones at massive scale — the same dynamic driving TSMC’s blowout quarter, covered in our earlier reporting on chip demand. As inference costs become the dominant line item in AI companies’ compute budgets, specialized silicon that can meaningfully undercut general-purpose GPUs on cost-per-token is attracting capital at a pace few sectors are matching right now.
What to Watch Next
The open questions the reporting hasn’t fully resolved are whether this round represents primary or secondary capital, how the $1 billion-plus in customer contracts breaks down by named customer, and how quickly Sohu chips move from evaluation racks like Jane Street’s into large-scale production deployment. Those answers will determine whether Etched’s valuation reflects real, durable demand or an unusually fast markup on the same underlying business.
Sources: TechCrunch, AI Weekly, Tech Startups
Disclaimer: This content is meant to inform and should not be considered financial advice. The views expressed in this article may include the author’s personal opinions and do not represent Times Tabloid’s opinion. Readers are advised to conduct thorough research before making any investment decisions. Any action taken by the reader is strictly at their own risk. Times Tabloid is not responsible for any financial losses.
