Capital

A Quant Fund Bet $700M After Testing the Hardware

Jane Street tested Etched's AI chip in production, then led its $700M round at a $21B valuation. Here's the real signal for AI infrastructure buyers.

CapitalBy Servola Tech Desk2026-08-183 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

A Quant Fund Bet $700M After Testing the Hardware

Key takeaways

Jane Street ran Etched's Sohu AI chip in a real production workload before leading the startup's $700 million round, a stronger signal than any benchmark slide.
Etched's valuation doubled from $10.3 billion to $21 billion in one month, the fastest jump in the company's funding history.
Sohu is an ASIC built to run only transformer-architecture models, not a general-purpose GPU, so it could beat Nvidia on cost per token but also risks obsolescence if the dominant AI architecture changes.
Etched claims one Sohu chip can replace multiple Nvidia H100 GPUs and that an eight-chip server hits over 500,000 tokens per second on Llama-3 70B, but no independent lab has verified these numbers on physical hardware.
The signal worth tracking is not Etched's own benchmarks, it's whether more named, technically rigorous customers follow Jane Street's pattern of deploying first and endorsing second.

The Round: 700 Million Dollars in a Month of Doubling

On 18 August 2026, Etched announced a $700 million funding round led by quantitative trading firm Jane Street, valuing the AI chip startup at $21 billion (roughly EUR 19.3 billion). Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone also joined the round.

That is just one month after a $300 million Series C in July 2026 valued Etched at $10.3 billion, meaning the valuation doubled in four weeks. TechCrunch's own headline put it plainly: Etched's valuation doubles to $21B in a month, an unusually fast pace for a round of this size.

Why Jane Street's Signature Is Different From a Term Sheet

Jane Street has a reputation as one of the most rigorous, technically demanding trading firms in the world, and it does not invest on faith. Around July 2026, Etched shipped its first production rack of Sohu hardware to Jane Street, and Jane Street put it to work on real, trading-adjacent workloads, not a lab benchmark.

Only after seeing the hardware perform in live production did Jane Street lead the new round itself. That makes this Etched's first confirmed real deployment with a named enterprise customer, something different from investor due diligence or a demo run by the vendor.

What Sohu Actually Is, and What Etched Claims It Does

Sohu is an ASIC, a chip built for a single purpose, made exclusively to run transformer-architecture models, the architecture behind nearly all of today's large language models. Instead of programmable GPU cores like Nvidia's, it uses fixed-function circuits and is manufactured on TSMC's N4P process.

Etched itself claims a Sohu server can replace up to 160 Nvidia H100 GPUs, that each chip is roughly 20 times faster, and that an eight-chip server delivers more than 500,000 tokens per second on Llama-3 70B, but no independent benchmarking body has yet measured Sohu hardware under production conditions, so these remain the company's own figures.

The Tradeoff Owners Should Actually Weigh

A chip that only runs one model architecture is a real lock-in risk: if transformer architecture is ever superseded, the hardware advantage disappears with it, while a general-purpose GPU can adapt to whatever comes next. This is the same tradeoff behind the EU's push for its own Chips Act, since European AI data centers still depend heavily on imported Nvidia GPUs.

The signal worth watching over the next round or two is not Etched's own numbers, it's whether more named, technically credible customers repeat Jane Street's pattern of running the hardware in production first and endorsing it second.

Frequently asked questions

What exactly did Etched announce on 18 August 2026?

A $700 million funding round led by Jane Street, valuing the AI chip startup at $21 billion, roughly double the $10.3 billion valuation it set just one month earlier in a $300 million Series C.

Why does Jane Street's involvement matter more than a typical venture round?

Jane Street did not just write a check. It received and ran Etched's first production rack of Sohu hardware on real trading-adjacent workloads around July 2026, then led the new round itself after seeing it work, making this Etched's first confirmed real customer deployment rather than only investor due diligence.

Should a business trust Etched's performance claims?

Not without independent verification. Etched claims its Sohu chip can replace multiple Nvidia H100 GPUs and hit over 500,000 tokens per second on Llama-3 70B, but these are the company's own figures; no outside benchmarking body has yet measured Sohu hardware under production conditions.

Etched's own numbers are not proof, they are a pitch, and owners evaluating AI infrastructure should treat them as such. What changed on 18 August 2026 is that a firm with no reason to be generous put real hardware into a real workload before it put money behind the company, and that sequence, hardware first, capital second, is the only kind of validation worth updating a buying decision on.

EtchedJane StreetAI ChipsInference HardwareNvidia

The Market Doubted Nvidia's $500 Billion Deal

Nvidia announced MOUs with six Wall Street firms to mobilize over $500 billion for AI compute financing. Its stock fell 3 percent the same day. Here is why that gap matters.

4 min read

Capital

Nvidia's $3 Billion Bet Is on Grid Power, Not Chips

Nvidia is reportedly putting up to $3 billion into Lancium, the Blackstone-backed power developer behind Stargate's first site, betting on grid-connection rights over silicon.

3 min read

Decisions

The Third Pivot Was the Warning Magic Leap Missed

Magic Leap filed a WARN notice for 193 layoffs effective 1 October 2026, its fourth strategic pivot after $4 billion raised. The lesson is the pattern, not the pivot.