← Back to feed

Etched has closed a $300 million Series C at a $10.3 billion valuation. The round was led by Sequoia and includes participation from a16z, Jane Street, Diffusion, and SK Hynix. Etched describes it as the highest valuation ever achieved in a Sequoia-led Series C.

The raise comes less than one month after Etched emerged from stealth at an $800 million valuation with $800M in initial funding and over $1 billion in signed contracts. The implied valuation jump — from $800M to $10.3B in under 30 days — is roughly 13x.

What the Money Buys

Etched is using the Series C to scale production and customer deployments. On the hardware side, two facilities are now active:

• A Taiwan factory established at stealth exit
• A new 80,000-square-foot facility in Milpitas, California, 15 minutes from headquarters — a 10 MW site housing an NPI (new product introduction) lab and an in-house SMT line

The Milpitas site is designed to compress the cycle between silicon design, prototyping, and customer deployment.

What Etched Builds

Etched’s Sohu chip is built specifically for transformer inference — not training, not general GPU workloads. The company’s founding claim is that transformer inference is the defining compute workload of the AI era, and that a chip designed around it outperforms H100s substantially per dollar and per watt. The company claims Sohu can replace 160 H100s for inference tasks.

That is a large claim that has not yet been independently validated at scale. Etched’s $1B+ in contracts suggests some customers are betting on it.

Investor Signal Reading

Sequoia leading at a record Series C valuation is the headline. It signals conviction that AI inference hardware has a standalone market — not just a market that exists until hyperscalers build custom silicon.

Jane Street backing is consistent with their earlier CoreWeave bet. The top quant shop is systematically acquiring positions across the AI compute stack — inference clusters, hardware, and now inference chips.

SK Hynix participation is strategically notable. HBM now represents 63% of AI chip component costs, and SK Hynix supplies it. An equity stake in a chip designer gives the memory maker visibility into future HBM demand curves and a preferred customer relationship to negotiate.

The Market Context

The AI inference chip market is fragmenting. NVIDIA dominates training and has a strong inference position, but the economics of inference-only workloads — high throughput, repetitive attention patterns, cost-per-token constraints — create room for specialized silicon. CoreWeave has demonstrated that inference neoclouds can scale to $99.4B backlogs. Etched is betting that the chip layer can capture similar margin.

Whether a purpose-built transformer chip can build a sustainable business before NVIDIA’s Rubin architecture closes the efficiency gap is the central question. The $10.3B valuation says investors think Etched has a window.