Table of Contents
Sequoia and Jane Street reportedly anchor two simultaneous Etched tranches at different prices
In sum – what we know:
- A dual-round structure – Etched is negotiating two tranches simultaneously, a $10 billion round led by Sequoia and a $20 billion round led by Jane Street.
- Unproven at scale – The valuation jump comes before Etched’s Sohu chip has been commercially validated or independently benchmarked.
- Inference-only bet – Sohu is a transformer-only chip built solely for LLM inference, trading architectural flexibility for specialized performance claims.
AI chip startup Etched is in talks for two new funding rounds that could push its valuation to roughly $20 billion — about four times the $5 billion mark it hit just months ago. Neither round has closed, and terms remain fluid, but the numbers alone make this one of the more striking examples yet of how fast capital is moving through the AI infrastructure market.
The San Jose-based company is chasing a very specific opportunity. Investors are racing to fund hardware that can run large language models more cheaply at enterprise scale. Etched hasn’t commercially validated its chip at scale yet, which makes the proposed valuation something of a case study in how AI capital cycles now work — vast sums chasing infrastructure plays well ahead of widespread deployment or third-party benchmarks.
Much of that appetite comes down to Nvidia. Its GPUs dominate both AI training and inference globally, and investors are clearly willing to pay a premium for any specialized hardware alternative with a credible shot at chipping away at that dominance. Etched is pitching itself as exactly that.
Dual-round financing structure and key investors
The structure of the deal is unusual in itself. Etched is reportedly negotiating two funding tranches simultaneously — one at a $10 billion valuation and another at approximately $20 billion. The $10 billion round is said to be led by Sequoia Capital, bringing in traditional Silicon Valley venture money. The $20 billion round is reportedly spearheaded by Jane Street, the quantitative trading firm and an existing Etched backer, making that tranche more of an insider-led or strategic round at a much higher price.
There’s a logic to the sequencing. Anchoring a lower price with insiders lets Etched secure capital quickly and signal momentum, while it concurrently markets a massive markup to broader institutional and sovereign-fund investors. It’s an aggressive approach, and it mirrors a wider pattern in the current AI funding cycle — startups selling shares at one valuation, then raising new capital at a much higher one in rapid succession, effectively marking up their own value in real time.
Etched isn’t alone here, either. Cerebras Systems has similarly aggregated massive capital reserves through overlapping rounds in its bid to challenge the incumbents, and the dual-round playbook is starting to look like standard practice among the top tier of AI chip startups. Whether it’s sustainable practice is a different question.
The Sohu chip technology
Etched’s flagship product is “Sohu,” an inference chip built specifically around the repeating computational patterns of modern large language models. The architecture is what the company calls transformer-only — it abandons general-purpose GPU flexibility entirely and instead hardwires the operations that dominate LLM workloads, like attention mechanisms and matrix multiplications, directly into the silicon.
That’s a double-edged sword. Specialization is where the pitch gets compelling. Etched claims superior performance-per-watt and a substantially lower total cost of ownership for cloud providers running LLMs at deployable scale — claims that remain the company’s own, given the lack of independent benchmarks so far. But hardwiring your chip to one model architecture also means betting that transformers stay dominant. If AI architectures shift meaningfully, a transformer-only design has nowhere to pivot.
The other notable choice is what Sohu doesn’t do. The chip targets inference exclusively (not training), with the goal of maximizing token throughput. Strategically, that keeps Etched out of the heavily contested training market where Nvidia and AMD are entrenched, and lets it undercut the generalized GPU premiums those companies command on the inference side. It’s a narrower fight, but arguably a winnable one.
Who is Etched?
Etched was founded around 2022 and operated largely in stealth until 2024. The company closed a seed round at a roughly $34 million valuation in 2023. By December 2025, it had raised a $500 million round led by growth equity firm Stripes, with participation from Peter Thiel, at a post-money valuation of about $5 billion. Total disclosed funding stood at approximately $800 million as of June 2026 — a fairly modest figure against the $10 billion and $20 billion price tags now under negotiation.
The company says it has booked roughly $1 billion in customer contracts, targeting hyperscale cloud providers and dedicated AI platform companies. That’s a meaningful number, though it’s worth noting these are booked contracts rather than recognized revenue, and independent verification is limited. Commitments of this kind typically hinge on the hardware actually delivering.
Macro conditions are doing some of the work here too. Stabilizing interest rates, strong earnings from public AI leaders, and sovereign wealth funds hunting for aggressive AI infrastructure exposure have all helped push private valuations upward. Like Cerebras Systems before it, Etched is leveraging booked sales and projected compute shortages to justify a multibillion-dollar valuation before its chips are widely deployed. If Sohu delivers on the performance and cost claims, the math may well work out. If it doesn’t, a company that quadrupled its valuation in seven months will have set expectations that are very hard to meet.
