One Customer Ran The Chip, And The Company's Value Doubled In A Month

On August 18, 2026, AI chip startup Etched announced it had raised another $700 million at a $21 billion valuation, led by the quantitative trading firm Jane Street after the fund tested and bought the startup's hardware.

The step-up is extreme. Etched was valued at $5 billion in December, raised a $300 million Series C at $10.3 billion in July, and doubled to $21 billion roughly a month later, adding nearly $11 billion in a single month.

What changed was not the technology but the evidence. Jane Street led the round and also became the company's first paying customer, deploying an Etched cluster in its own data center and running live workloads through it. As Jane Street put it, it tested the chip and was pleased with the early results.

Etched says it has more than $1 billion in customer contracts spanning public and private AI companies and cloud providers, and its systems are already running large models in production.

Take a look at this…

It's smaller than a fingertip. It's made of glass. And it's about to reshape AI from the ground up.

Jensen Huang, Nvidia's CEO, says this device is shattering the limitations of AI and without it, AI can't scale.

Google Ventures says it's the future of AI compute.

And Sequoia Capital – the firm that backed Anthropic and OpenAI – calls it a "holy grail".

Yet most Americans have never heard of it…

Wall Street insider Jason Bodner – the same man who called Nvidia at $4.50 – says this critical "light-speed" device could be bigger for AI than GPUs… and it's about to launch a whole new wave of AI winners. And to prove it, he's giving away his #1 stock involved with it – for free.

Click here for all the details.

Etched was founded in 2022 by Gavin Uberti, Chris Zhu, and Robert Wachen, who left Harvard to start the company. Its product, Sohu, is an ASIC built on TSMC's 4-nanometer process and designed for one job: running transformer models, the architecture behind today's major AI systems.

The engineering argument is about waste. A GPU runs inference through the CUDA software layer, which schedules work across thousands of general-purpose cores, and on transformer workloads that overhead can leave much of a GPU's theoretical capacity idle. Sohu removes the layer by hardwiring the computation into silicon.

The claimed payoff is large. Etched says Sohu can generate 500,000 tokens per second on Llama 70B, which it describes as 20 times the throughput of an eight-GPU H100 system.

The bet follows the money. Training happens periodically, but inference happens every time someone queries an assistant or runs an agent, so at scale the per-request cost becomes the dominant infrastructure expense. Shaving that cost is worth enormous sums.

Proof of deployment is what repriced the company. Design claims and signed orders were already there; a paying customer running production workloads converted a promise into evidence, and investors paid roughly double for the difference.

For Nvidia, the threat is narrow but real. Etched cannot replace GPUs across the board, but it is targeting the single largest and fastest-growing slice of AI compute spending.

Sohu is hardwired for transformers, which means it has no fallback if the field moves on. A GPU can run whatever comes next; this chip cannot. That is the entire risk of the design, and it is unhedged.

Customer proof is also thin. Jane Street is the one publicly confirmed deployment, and it is simultaneously the lead investor, which makes it a motivated reference rather than an independent one. Nvidia's newer chips keep narrowing the cost gap, other specialized inference startups are chasing the same bottleneck, and a $21 billion price assumes years of adoption that has barely begun.

The round matters because it shows AI hardware competition shifting from raw capability toward the economics of serving models at scale. The real impact depends on whether customers beyond a single investor deploy Sohu in volume, and whether the transformer stays dominant long enough for a chip built only for it to pay off.

Disclosure: This content is for educational and informational purposes only and does not constitute investment advice or recommendations. You should always conduct your own research or consult a qualified financial advisor before making investment decisions.